]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/mmap.c | |
3 | * | |
4 | * Written by obz. | |
5 | * | |
046c6884 | 6 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
7 | */ |
8 | ||
b1de0d13 MH |
9 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
10 | ||
e8420a8e | 11 | #include <linux/kernel.h> |
1da177e4 | 12 | #include <linux/slab.h> |
4af3c9cc | 13 | #include <linux/backing-dev.h> |
1da177e4 | 14 | #include <linux/mm.h> |
615d6e87 | 15 | #include <linux/vmacache.h> |
1da177e4 LT |
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> | |
c59ede7b | 21 | #include <linux/capability.h> |
1da177e4 LT |
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> | |
c01d5b30 | 28 | #include <linux/shmem_fs.h> |
1da177e4 | 29 | #include <linux/profile.h> |
b95f1b31 | 30 | #include <linux/export.h> |
1da177e4 LT |
31 | #include <linux/mount.h> |
32 | #include <linux/mempolicy.h> | |
33 | #include <linux/rmap.h> | |
cddb8a5c | 34 | #include <linux/mmu_notifier.h> |
82f71ae4 | 35 | #include <linux/mmdebug.h> |
cdd6c482 | 36 | #include <linux/perf_event.h> |
120a795d | 37 | #include <linux/audit.h> |
b15d00b6 | 38 | #include <linux/khugepaged.h> |
2b144498 | 39 | #include <linux/uprobes.h> |
d3737187 | 40 | #include <linux/rbtree_augmented.h> |
1640879a AS |
41 | #include <linux/notifier.h> |
42 | #include <linux/memory.h> | |
b1de0d13 | 43 | #include <linux/printk.h> |
19a809af | 44 | #include <linux/userfaultfd_k.h> |
d977d56c | 45 | #include <linux/moduleparam.h> |
62b5f7d0 | 46 | #include <linux/pkeys.h> |
1da177e4 LT |
47 | |
48 | #include <asm/uaccess.h> | |
49 | #include <asm/cacheflush.h> | |
50 | #include <asm/tlb.h> | |
d6dd61c8 | 51 | #include <asm/mmu_context.h> |
1da177e4 | 52 | |
42b77728 JB |
53 | #include "internal.h" |
54 | ||
3a459756 KK |
55 | #ifndef arch_mmap_check |
56 | #define arch_mmap_check(addr, len, flags) (0) | |
57 | #endif | |
58 | ||
d07e2259 DC |
59 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS |
60 | const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; | |
61 | const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; | |
62 | int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; | |
63 | #endif | |
64 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS | |
65 | const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; | |
66 | const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; | |
67 | int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; | |
68 | #endif | |
69 | ||
f4fcd558 | 70 | static bool ignore_rlimit_data; |
d977d56c | 71 | core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); |
d07e2259 | 72 | |
e0da382c HD |
73 | static void unmap_region(struct mm_struct *mm, |
74 | struct vm_area_struct *vma, struct vm_area_struct *prev, | |
75 | unsigned long start, unsigned long end); | |
76 | ||
1da177e4 LT |
77 | /* description of effects of mapping type and prot in current implementation. |
78 | * this is due to the limited x86 page protection hardware. The expected | |
79 | * behavior is in parens: | |
80 | * | |
81 | * map_type prot | |
82 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC | |
83 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
84 | * w: (no) no w: (no) no w: (yes) yes w: (no) no | |
85 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
cc71aba3 | 86 | * |
1da177e4 LT |
87 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes |
88 | * w: (no) no w: (no) no w: (copy) copy w: (no) no | |
89 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
90 | * | |
91 | */ | |
92 | pgprot_t protection_map[16] = { | |
93 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, | |
94 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
95 | }; | |
96 | ||
804af2cf HD |
97 | pgprot_t vm_get_page_prot(unsigned long vm_flags) |
98 | { | |
b845f313 DK |
99 | return __pgprot(pgprot_val(protection_map[vm_flags & |
100 | (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | | |
101 | pgprot_val(arch_vm_get_page_prot(vm_flags))); | |
804af2cf HD |
102 | } |
103 | EXPORT_SYMBOL(vm_get_page_prot); | |
104 | ||
64e45507 PF |
105 | static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) |
106 | { | |
107 | return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); | |
108 | } | |
109 | ||
110 | /* Update vma->vm_page_prot to reflect vma->vm_flags. */ | |
111 | void vma_set_page_prot(struct vm_area_struct *vma) | |
112 | { | |
113 | unsigned long vm_flags = vma->vm_flags; | |
114 | ||
115 | vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); | |
116 | if (vma_wants_writenotify(vma)) { | |
117 | vm_flags &= ~VM_SHARED; | |
118 | vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, | |
119 | vm_flags); | |
120 | } | |
121 | } | |
122 | ||
1da177e4 | 123 | /* |
c8c06efa | 124 | * Requires inode->i_mapping->i_mmap_rwsem |
1da177e4 LT |
125 | */ |
126 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
127 | struct file *file, struct address_space *mapping) | |
128 | { | |
129 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 130 | atomic_inc(&file_inode(file)->i_writecount); |
1da177e4 | 131 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 132 | mapping_unmap_writable(mapping); |
1da177e4 LT |
133 | |
134 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 135 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
136 | flush_dcache_mmap_unlock(mapping); |
137 | } | |
138 | ||
139 | /* | |
6b2dbba8 | 140 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 141 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 142 | */ |
a8fb5618 | 143 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
144 | { |
145 | struct file *file = vma->vm_file; | |
146 | ||
1da177e4 LT |
147 | if (file) { |
148 | struct address_space *mapping = file->f_mapping; | |
83cde9e8 | 149 | i_mmap_lock_write(mapping); |
1da177e4 | 150 | __remove_shared_vm_struct(vma, file, mapping); |
83cde9e8 | 151 | i_mmap_unlock_write(mapping); |
1da177e4 | 152 | } |
a8fb5618 HD |
153 | } |
154 | ||
155 | /* | |
156 | * Close a vm structure and free it, returning the next. | |
157 | */ | |
158 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
159 | { | |
160 | struct vm_area_struct *next = vma->vm_next; | |
161 | ||
a8fb5618 | 162 | might_sleep(); |
1da177e4 LT |
163 | if (vma->vm_ops && vma->vm_ops->close) |
164 | vma->vm_ops->close(vma); | |
e9714acf | 165 | if (vma->vm_file) |
a8fb5618 | 166 | fput(vma->vm_file); |
f0be3d32 | 167 | mpol_put(vma_policy(vma)); |
1da177e4 | 168 | kmem_cache_free(vm_area_cachep, vma); |
a8fb5618 | 169 | return next; |
1da177e4 LT |
170 | } |
171 | ||
5d22fc25 | 172 | static int do_brk(unsigned long addr, unsigned long len); |
e4eb1ff6 | 173 | |
6a6160a7 | 174 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 | 175 | { |
8764b338 | 176 | unsigned long retval; |
1da177e4 LT |
177 | unsigned long newbrk, oldbrk; |
178 | struct mm_struct *mm = current->mm; | |
a5b4592c | 179 | unsigned long min_brk; |
128557ff | 180 | bool populate; |
1da177e4 | 181 | |
dc0ef0df MH |
182 | if (down_write_killable(&mm->mmap_sem)) |
183 | return -EINTR; | |
1da177e4 | 184 | |
a5b4592c | 185 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
186 | /* |
187 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
188 | * randomize_va_space to 2, which will still cause mm->start_brk | |
189 | * to be arbitrarily shifted | |
190 | */ | |
4471a675 | 191 | if (current->brk_randomized) |
5520e894 JK |
192 | min_brk = mm->start_brk; |
193 | else | |
194 | min_brk = mm->end_data; | |
a5b4592c JK |
195 | #else |
196 | min_brk = mm->start_brk; | |
197 | #endif | |
198 | if (brk < min_brk) | |
1da177e4 | 199 | goto out; |
1e624196 RG |
200 | |
201 | /* | |
202 | * Check against rlimit here. If this check is done later after the test | |
203 | * of oldbrk with newbrk then it can escape the test and let the data | |
204 | * segment grow beyond its set limit the in case where the limit is | |
205 | * not page aligned -Ram Gupta | |
206 | */ | |
8764b338 CG |
207 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
208 | mm->end_data, mm->start_data)) | |
1e624196 RG |
209 | goto out; |
210 | ||
1da177e4 LT |
211 | newbrk = PAGE_ALIGN(brk); |
212 | oldbrk = PAGE_ALIGN(mm->brk); | |
213 | if (oldbrk == newbrk) | |
214 | goto set_brk; | |
215 | ||
216 | /* Always allow shrinking brk. */ | |
217 | if (brk <= mm->brk) { | |
218 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
219 | goto set_brk; | |
220 | goto out; | |
221 | } | |
222 | ||
1da177e4 LT |
223 | /* Check against existing mmap mappings. */ |
224 | if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) | |
225 | goto out; | |
226 | ||
227 | /* Ok, looks good - let it rip. */ | |
5d22fc25 | 228 | if (do_brk(oldbrk, newbrk-oldbrk) < 0) |
1da177e4 | 229 | goto out; |
128557ff | 230 | |
1da177e4 LT |
231 | set_brk: |
232 | mm->brk = brk; | |
128557ff ML |
233 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
234 | up_write(&mm->mmap_sem); | |
235 | if (populate) | |
236 | mm_populate(oldbrk, newbrk - oldbrk); | |
237 | return brk; | |
238 | ||
1da177e4 LT |
239 | out: |
240 | retval = mm->brk; | |
241 | up_write(&mm->mmap_sem); | |
242 | return retval; | |
243 | } | |
244 | ||
d3737187 ML |
245 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
246 | { | |
247 | unsigned long max, subtree_gap; | |
248 | max = vma->vm_start; | |
249 | if (vma->vm_prev) | |
250 | max -= vma->vm_prev->vm_end; | |
251 | if (vma->vm_rb.rb_left) { | |
252 | subtree_gap = rb_entry(vma->vm_rb.rb_left, | |
253 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
254 | if (subtree_gap > max) | |
255 | max = subtree_gap; | |
256 | } | |
257 | if (vma->vm_rb.rb_right) { | |
258 | subtree_gap = rb_entry(vma->vm_rb.rb_right, | |
259 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
260 | if (subtree_gap > max) | |
261 | max = subtree_gap; | |
262 | } | |
263 | return max; | |
264 | } | |
265 | ||
ed8ea815 | 266 | #ifdef CONFIG_DEBUG_VM_RB |
acf128d0 | 267 | static int browse_rb(struct mm_struct *mm) |
1da177e4 | 268 | { |
acf128d0 | 269 | struct rb_root *root = &mm->mm_rb; |
5a0768f6 | 270 | int i = 0, j, bug = 0; |
1da177e4 LT |
271 | struct rb_node *nd, *pn = NULL; |
272 | unsigned long prev = 0, pend = 0; | |
273 | ||
274 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
275 | struct vm_area_struct *vma; | |
276 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
5a0768f6 | 277 | if (vma->vm_start < prev) { |
ff26f70f AM |
278 | pr_emerg("vm_start %lx < prev %lx\n", |
279 | vma->vm_start, prev); | |
5a0768f6 ML |
280 | bug = 1; |
281 | } | |
282 | if (vma->vm_start < pend) { | |
ff26f70f AM |
283 | pr_emerg("vm_start %lx < pend %lx\n", |
284 | vma->vm_start, pend); | |
5a0768f6 ML |
285 | bug = 1; |
286 | } | |
287 | if (vma->vm_start > vma->vm_end) { | |
ff26f70f AM |
288 | pr_emerg("vm_start %lx > vm_end %lx\n", |
289 | vma->vm_start, vma->vm_end); | |
5a0768f6 ML |
290 | bug = 1; |
291 | } | |
acf128d0 | 292 | spin_lock(&mm->page_table_lock); |
5a0768f6 | 293 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { |
8542bdfc | 294 | pr_emerg("free gap %lx, correct %lx\n", |
5a0768f6 ML |
295 | vma->rb_subtree_gap, |
296 | vma_compute_subtree_gap(vma)); | |
297 | bug = 1; | |
298 | } | |
acf128d0 | 299 | spin_unlock(&mm->page_table_lock); |
1da177e4 LT |
300 | i++; |
301 | pn = nd; | |
d1af65d1 DM |
302 | prev = vma->vm_start; |
303 | pend = vma->vm_end; | |
1da177e4 LT |
304 | } |
305 | j = 0; | |
5a0768f6 | 306 | for (nd = pn; nd; nd = rb_prev(nd)) |
1da177e4 | 307 | j++; |
5a0768f6 | 308 | if (i != j) { |
8542bdfc | 309 | pr_emerg("backwards %d, forwards %d\n", j, i); |
5a0768f6 | 310 | bug = 1; |
1da177e4 | 311 | } |
5a0768f6 | 312 | return bug ? -1 : i; |
1da177e4 LT |
313 | } |
314 | ||
d3737187 ML |
315 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
316 | { | |
317 | struct rb_node *nd; | |
318 | ||
319 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
320 | struct vm_area_struct *vma; | |
321 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
96dad67f SL |
322 | VM_BUG_ON_VMA(vma != ignore && |
323 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma), | |
324 | vma); | |
1da177e4 | 325 | } |
1da177e4 LT |
326 | } |
327 | ||
eafd4dc4 | 328 | static void validate_mm(struct mm_struct *mm) |
1da177e4 LT |
329 | { |
330 | int bug = 0; | |
331 | int i = 0; | |
5a0768f6 | 332 | unsigned long highest_address = 0; |
ed8ea815 | 333 | struct vm_area_struct *vma = mm->mmap; |
ff26f70f | 334 | |
ed8ea815 | 335 | while (vma) { |
12352d3c | 336 | struct anon_vma *anon_vma = vma->anon_vma; |
ed8ea815 | 337 | struct anon_vma_chain *avc; |
ff26f70f | 338 | |
12352d3c KK |
339 | if (anon_vma) { |
340 | anon_vma_lock_read(anon_vma); | |
341 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
342 | anon_vma_interval_tree_verify(avc); | |
343 | anon_vma_unlock_read(anon_vma); | |
344 | } | |
345 | ||
5a0768f6 | 346 | highest_address = vma->vm_end; |
ed8ea815 | 347 | vma = vma->vm_next; |
1da177e4 LT |
348 | i++; |
349 | } | |
5a0768f6 | 350 | if (i != mm->map_count) { |
8542bdfc | 351 | pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); |
5a0768f6 ML |
352 | bug = 1; |
353 | } | |
354 | if (highest_address != mm->highest_vm_end) { | |
8542bdfc | 355 | pr_emerg("mm->highest_vm_end %lx, found %lx\n", |
ff26f70f | 356 | mm->highest_vm_end, highest_address); |
5a0768f6 ML |
357 | bug = 1; |
358 | } | |
acf128d0 | 359 | i = browse_rb(mm); |
5a0768f6 | 360 | if (i != mm->map_count) { |
ff26f70f AM |
361 | if (i != -1) |
362 | pr_emerg("map_count %d rb %d\n", mm->map_count, i); | |
5a0768f6 ML |
363 | bug = 1; |
364 | } | |
96dad67f | 365 | VM_BUG_ON_MM(bug, mm); |
1da177e4 LT |
366 | } |
367 | #else | |
d3737187 | 368 | #define validate_mm_rb(root, ignore) do { } while (0) |
1da177e4 LT |
369 | #define validate_mm(mm) do { } while (0) |
370 | #endif | |
371 | ||
d3737187 ML |
372 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
373 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) | |
374 | ||
375 | /* | |
376 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or | |
377 | * vma->vm_prev->vm_end values changed, without modifying the vma's position | |
378 | * in the rbtree. | |
379 | */ | |
380 | static void vma_gap_update(struct vm_area_struct *vma) | |
381 | { | |
382 | /* | |
383 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback | |
384 | * function that does exacltly what we want. | |
385 | */ | |
386 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); | |
387 | } | |
388 | ||
389 | static inline void vma_rb_insert(struct vm_area_struct *vma, | |
390 | struct rb_root *root) | |
391 | { | |
392 | /* All rb_subtree_gap values must be consistent prior to insertion */ | |
393 | validate_mm_rb(root, NULL); | |
394 | ||
395 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
396 | } | |
397 | ||
398 | static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) | |
399 | { | |
400 | /* | |
401 | * All rb_subtree_gap values must be consistent prior to erase, | |
402 | * with the possible exception of the vma being erased. | |
403 | */ | |
404 | validate_mm_rb(root, vma); | |
405 | ||
406 | /* | |
407 | * Note rb_erase_augmented is a fairly large inline function, | |
408 | * so make sure we instantiate it only once with our desired | |
409 | * augmented rbtree callbacks. | |
410 | */ | |
411 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
412 | } | |
413 | ||
bf181b9f ML |
414 | /* |
415 | * vma has some anon_vma assigned, and is already inserted on that | |
416 | * anon_vma's interval trees. | |
417 | * | |
418 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
419 | * vma must be removed from the anon_vma's interval trees using | |
420 | * anon_vma_interval_tree_pre_update_vma(). | |
421 | * | |
422 | * After the update, the vma will be reinserted using | |
423 | * anon_vma_interval_tree_post_update_vma(). | |
424 | * | |
425 | * The entire update must be protected by exclusive mmap_sem and by | |
426 | * the root anon_vma's mutex. | |
427 | */ | |
428 | static inline void | |
429 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
430 | { | |
431 | struct anon_vma_chain *avc; | |
432 | ||
433 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
434 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
435 | } | |
436 | ||
437 | static inline void | |
438 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
439 | { | |
440 | struct anon_vma_chain *avc; | |
441 | ||
442 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
443 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
444 | } | |
445 | ||
6597d783 HD |
446 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
447 | unsigned long end, struct vm_area_struct **pprev, | |
448 | struct rb_node ***rb_link, struct rb_node **rb_parent) | |
1da177e4 | 449 | { |
6597d783 | 450 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
1da177e4 LT |
451 | |
452 | __rb_link = &mm->mm_rb.rb_node; | |
453 | rb_prev = __rb_parent = NULL; | |
1da177e4 LT |
454 | |
455 | while (*__rb_link) { | |
456 | struct vm_area_struct *vma_tmp; | |
457 | ||
458 | __rb_parent = *__rb_link; | |
459 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
460 | ||
461 | if (vma_tmp->vm_end > addr) { | |
6597d783 HD |
462 | /* Fail if an existing vma overlaps the area */ |
463 | if (vma_tmp->vm_start < end) | |
464 | return -ENOMEM; | |
1da177e4 LT |
465 | __rb_link = &__rb_parent->rb_left; |
466 | } else { | |
467 | rb_prev = __rb_parent; | |
468 | __rb_link = &__rb_parent->rb_right; | |
469 | } | |
470 | } | |
471 | ||
472 | *pprev = NULL; | |
473 | if (rb_prev) | |
474 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
475 | *rb_link = __rb_link; | |
476 | *rb_parent = __rb_parent; | |
6597d783 | 477 | return 0; |
1da177e4 LT |
478 | } |
479 | ||
e8420a8e CH |
480 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
481 | unsigned long addr, unsigned long end) | |
482 | { | |
483 | unsigned long nr_pages = 0; | |
484 | struct vm_area_struct *vma; | |
485 | ||
486 | /* Find first overlaping mapping */ | |
487 | vma = find_vma_intersection(mm, addr, end); | |
488 | if (!vma) | |
489 | return 0; | |
490 | ||
491 | nr_pages = (min(end, vma->vm_end) - | |
492 | max(addr, vma->vm_start)) >> PAGE_SHIFT; | |
493 | ||
494 | /* Iterate over the rest of the overlaps */ | |
495 | for (vma = vma->vm_next; vma; vma = vma->vm_next) { | |
496 | unsigned long overlap_len; | |
497 | ||
498 | if (vma->vm_start > end) | |
499 | break; | |
500 | ||
501 | overlap_len = min(end, vma->vm_end) - vma->vm_start; | |
502 | nr_pages += overlap_len >> PAGE_SHIFT; | |
503 | } | |
504 | ||
505 | return nr_pages; | |
506 | } | |
507 | ||
1da177e4 LT |
508 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
509 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
510 | { | |
d3737187 ML |
511 | /* Update tracking information for the gap following the new vma. */ |
512 | if (vma->vm_next) | |
513 | vma_gap_update(vma->vm_next); | |
514 | else | |
515 | mm->highest_vm_end = vma->vm_end; | |
516 | ||
517 | /* | |
518 | * vma->vm_prev wasn't known when we followed the rbtree to find the | |
519 | * correct insertion point for that vma. As a result, we could not | |
520 | * update the vma vm_rb parents rb_subtree_gap values on the way down. | |
521 | * So, we first insert the vma with a zero rb_subtree_gap value | |
522 | * (to be consistent with what we did on the way down), and then | |
523 | * immediately update the gap to the correct value. Finally we | |
524 | * rebalance the rbtree after all augmented values have been set. | |
525 | */ | |
1da177e4 | 526 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
d3737187 ML |
527 | vma->rb_subtree_gap = 0; |
528 | vma_gap_update(vma); | |
529 | vma_rb_insert(vma, &mm->mm_rb); | |
1da177e4 LT |
530 | } |
531 | ||
cb8f488c | 532 | static void __vma_link_file(struct vm_area_struct *vma) |
1da177e4 | 533 | { |
48aae425 | 534 | struct file *file; |
1da177e4 LT |
535 | |
536 | file = vma->vm_file; | |
537 | if (file) { | |
538 | struct address_space *mapping = file->f_mapping; | |
539 | ||
540 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 541 | atomic_dec(&file_inode(file)->i_writecount); |
1da177e4 | 542 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 543 | atomic_inc(&mapping->i_mmap_writable); |
1da177e4 LT |
544 | |
545 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 546 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 LT |
547 | flush_dcache_mmap_unlock(mapping); |
548 | } | |
549 | } | |
550 | ||
551 | static void | |
552 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
553 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
554 | struct rb_node *rb_parent) | |
555 | { | |
556 | __vma_link_list(mm, vma, prev, rb_parent); | |
557 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
1da177e4 LT |
558 | } |
559 | ||
560 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
561 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
562 | struct rb_node *rb_parent) | |
563 | { | |
564 | struct address_space *mapping = NULL; | |
565 | ||
64ac4940 | 566 | if (vma->vm_file) { |
1da177e4 | 567 | mapping = vma->vm_file->f_mapping; |
83cde9e8 | 568 | i_mmap_lock_write(mapping); |
64ac4940 | 569 | } |
1da177e4 LT |
570 | |
571 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
572 | __vma_link_file(vma); | |
573 | ||
1da177e4 | 574 | if (mapping) |
83cde9e8 | 575 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
576 | |
577 | mm->map_count++; | |
578 | validate_mm(mm); | |
579 | } | |
580 | ||
581 | /* | |
88f6b4c3 | 582 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
6b2dbba8 | 583 | * mm's list and rbtree. It has already been inserted into the interval tree. |
1da177e4 | 584 | */ |
48aae425 | 585 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 586 | { |
6597d783 | 587 | struct vm_area_struct *prev; |
48aae425 | 588 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 589 | |
6597d783 HD |
590 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
591 | &prev, &rb_link, &rb_parent)) | |
592 | BUG(); | |
1da177e4 LT |
593 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
594 | mm->map_count++; | |
595 | } | |
596 | ||
597 | static inline void | |
598 | __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, | |
599 | struct vm_area_struct *prev) | |
600 | { | |
d3737187 | 601 | struct vm_area_struct *next; |
297c5eee | 602 | |
d3737187 ML |
603 | vma_rb_erase(vma, &mm->mm_rb); |
604 | prev->vm_next = next = vma->vm_next; | |
297c5eee LT |
605 | if (next) |
606 | next->vm_prev = prev; | |
615d6e87 DB |
607 | |
608 | /* Kill the cache */ | |
609 | vmacache_invalidate(mm); | |
1da177e4 LT |
610 | } |
611 | ||
612 | /* | |
613 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
614 | * is already present in an i_mmap tree without adjusting the tree. | |
615 | * The following helper function should be used when such adjustments | |
616 | * are necessary. The "insert" vma (if any) is to be inserted | |
617 | * before we drop the necessary locks. | |
618 | */ | |
5beb4930 | 619 | int vma_adjust(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
620 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) |
621 | { | |
622 | struct mm_struct *mm = vma->vm_mm; | |
623 | struct vm_area_struct *next = vma->vm_next; | |
624 | struct vm_area_struct *importer = NULL; | |
625 | struct address_space *mapping = NULL; | |
6b2dbba8 | 626 | struct rb_root *root = NULL; |
012f1800 | 627 | struct anon_vma *anon_vma = NULL; |
1da177e4 | 628 | struct file *file = vma->vm_file; |
d3737187 | 629 | bool start_changed = false, end_changed = false; |
1da177e4 LT |
630 | long adjust_next = 0; |
631 | int remove_next = 0; | |
632 | ||
633 | if (next && !insert) { | |
287d97ac LT |
634 | struct vm_area_struct *exporter = NULL; |
635 | ||
1da177e4 LT |
636 | if (end >= next->vm_end) { |
637 | /* | |
638 | * vma expands, overlapping all the next, and | |
639 | * perhaps the one after too (mprotect case 6). | |
640 | */ | |
641 | again: remove_next = 1 + (end > next->vm_end); | |
642 | end = next->vm_end; | |
287d97ac | 643 | exporter = next; |
1da177e4 LT |
644 | importer = vma; |
645 | } else if (end > next->vm_start) { | |
646 | /* | |
647 | * vma expands, overlapping part of the next: | |
648 | * mprotect case 5 shifting the boundary up. | |
649 | */ | |
650 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
287d97ac | 651 | exporter = next; |
1da177e4 LT |
652 | importer = vma; |
653 | } else if (end < vma->vm_end) { | |
654 | /* | |
655 | * vma shrinks, and !insert tells it's not | |
656 | * split_vma inserting another: so it must be | |
657 | * mprotect case 4 shifting the boundary down. | |
658 | */ | |
cc71aba3 | 659 | adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); |
287d97ac | 660 | exporter = vma; |
1da177e4 LT |
661 | importer = next; |
662 | } | |
1da177e4 | 663 | |
5beb4930 RR |
664 | /* |
665 | * Easily overlooked: when mprotect shifts the boundary, | |
666 | * make sure the expanding vma has anon_vma set if the | |
667 | * shrinking vma had, to cover any anon pages imported. | |
668 | */ | |
287d97ac | 669 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
c4ea95d7 DF |
670 | int error; |
671 | ||
b800c91a | 672 | importer->anon_vma = exporter->anon_vma; |
c4ea95d7 | 673 | error = anon_vma_clone(importer, exporter); |
3fe89b3e | 674 | if (error) |
c4ea95d7 | 675 | return error; |
5beb4930 RR |
676 | } |
677 | } | |
678 | ||
37f9f559 KS |
679 | vma_adjust_trans_huge(vma, start, end, adjust_next); |
680 | ||
1da177e4 LT |
681 | if (file) { |
682 | mapping = file->f_mapping; | |
27ba0644 KS |
683 | root = &mapping->i_mmap; |
684 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); | |
682968e0 | 685 | |
27ba0644 KS |
686 | if (adjust_next) |
687 | uprobe_munmap(next, next->vm_start, next->vm_end); | |
682968e0 | 688 | |
83cde9e8 | 689 | i_mmap_lock_write(mapping); |
1da177e4 | 690 | if (insert) { |
1da177e4 | 691 | /* |
6b2dbba8 | 692 | * Put into interval tree now, so instantiated pages |
1da177e4 LT |
693 | * are visible to arm/parisc __flush_dcache_page |
694 | * throughout; but we cannot insert into address | |
695 | * space until vma start or end is updated. | |
696 | */ | |
697 | __vma_link_file(insert); | |
698 | } | |
699 | } | |
700 | ||
bf181b9f ML |
701 | anon_vma = vma->anon_vma; |
702 | if (!anon_vma && adjust_next) | |
703 | anon_vma = next->anon_vma; | |
704 | if (anon_vma) { | |
81d1b09c SL |
705 | VM_BUG_ON_VMA(adjust_next && next->anon_vma && |
706 | anon_vma != next->anon_vma, next); | |
4fc3f1d6 | 707 | anon_vma_lock_write(anon_vma); |
bf181b9f ML |
708 | anon_vma_interval_tree_pre_update_vma(vma); |
709 | if (adjust_next) | |
710 | anon_vma_interval_tree_pre_update_vma(next); | |
711 | } | |
012f1800 | 712 | |
1da177e4 LT |
713 | if (root) { |
714 | flush_dcache_mmap_lock(mapping); | |
6b2dbba8 | 715 | vma_interval_tree_remove(vma, root); |
1da177e4 | 716 | if (adjust_next) |
6b2dbba8 | 717 | vma_interval_tree_remove(next, root); |
1da177e4 LT |
718 | } |
719 | ||
d3737187 ML |
720 | if (start != vma->vm_start) { |
721 | vma->vm_start = start; | |
722 | start_changed = true; | |
723 | } | |
724 | if (end != vma->vm_end) { | |
725 | vma->vm_end = end; | |
726 | end_changed = true; | |
727 | } | |
1da177e4 LT |
728 | vma->vm_pgoff = pgoff; |
729 | if (adjust_next) { | |
730 | next->vm_start += adjust_next << PAGE_SHIFT; | |
731 | next->vm_pgoff += adjust_next; | |
732 | } | |
733 | ||
734 | if (root) { | |
735 | if (adjust_next) | |
6b2dbba8 ML |
736 | vma_interval_tree_insert(next, root); |
737 | vma_interval_tree_insert(vma, root); | |
1da177e4 LT |
738 | flush_dcache_mmap_unlock(mapping); |
739 | } | |
740 | ||
741 | if (remove_next) { | |
742 | /* | |
743 | * vma_merge has merged next into vma, and needs | |
744 | * us to remove next before dropping the locks. | |
745 | */ | |
746 | __vma_unlink(mm, next, vma); | |
747 | if (file) | |
748 | __remove_shared_vm_struct(next, file, mapping); | |
1da177e4 LT |
749 | } else if (insert) { |
750 | /* | |
751 | * split_vma has split insert from vma, and needs | |
752 | * us to insert it before dropping the locks | |
753 | * (it may either follow vma or precede it). | |
754 | */ | |
755 | __insert_vm_struct(mm, insert); | |
d3737187 ML |
756 | } else { |
757 | if (start_changed) | |
758 | vma_gap_update(vma); | |
759 | if (end_changed) { | |
760 | if (!next) | |
761 | mm->highest_vm_end = end; | |
762 | else if (!adjust_next) | |
763 | vma_gap_update(next); | |
764 | } | |
1da177e4 LT |
765 | } |
766 | ||
bf181b9f ML |
767 | if (anon_vma) { |
768 | anon_vma_interval_tree_post_update_vma(vma); | |
769 | if (adjust_next) | |
770 | anon_vma_interval_tree_post_update_vma(next); | |
08b52706 | 771 | anon_vma_unlock_write(anon_vma); |
bf181b9f | 772 | } |
1da177e4 | 773 | if (mapping) |
83cde9e8 | 774 | i_mmap_unlock_write(mapping); |
1da177e4 | 775 | |
2b144498 | 776 | if (root) { |
7b2d81d4 | 777 | uprobe_mmap(vma); |
2b144498 SD |
778 | |
779 | if (adjust_next) | |
7b2d81d4 | 780 | uprobe_mmap(next); |
2b144498 SD |
781 | } |
782 | ||
1da177e4 | 783 | if (remove_next) { |
925d1c40 | 784 | if (file) { |
cbc91f71 | 785 | uprobe_munmap(next, next->vm_start, next->vm_end); |
1da177e4 | 786 | fput(file); |
925d1c40 | 787 | } |
5beb4930 RR |
788 | if (next->anon_vma) |
789 | anon_vma_merge(vma, next); | |
1da177e4 | 790 | mm->map_count--; |
3964acd0 | 791 | mpol_put(vma_policy(next)); |
1da177e4 LT |
792 | kmem_cache_free(vm_area_cachep, next); |
793 | /* | |
794 | * In mprotect's case 6 (see comments on vma_merge), | |
795 | * we must remove another next too. It would clutter | |
796 | * up the code too much to do both in one go. | |
797 | */ | |
d3737187 ML |
798 | next = vma->vm_next; |
799 | if (remove_next == 2) | |
1da177e4 | 800 | goto again; |
d3737187 ML |
801 | else if (next) |
802 | vma_gap_update(next); | |
803 | else | |
804 | mm->highest_vm_end = end; | |
1da177e4 | 805 | } |
2b144498 | 806 | if (insert && file) |
7b2d81d4 | 807 | uprobe_mmap(insert); |
1da177e4 LT |
808 | |
809 | validate_mm(mm); | |
5beb4930 RR |
810 | |
811 | return 0; | |
1da177e4 LT |
812 | } |
813 | ||
814 | /* | |
815 | * If the vma has a ->close operation then the driver probably needs to release | |
816 | * per-vma resources, so we don't attempt to merge those. | |
817 | */ | |
1da177e4 | 818 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
19a809af AA |
819 | struct file *file, unsigned long vm_flags, |
820 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 821 | { |
34228d47 CG |
822 | /* |
823 | * VM_SOFTDIRTY should not prevent from VMA merging, if we | |
824 | * match the flags but dirty bit -- the caller should mark | |
825 | * merged VMA as dirty. If dirty bit won't be excluded from | |
826 | * comparison, we increase pressue on the memory system forcing | |
827 | * the kernel to generate new VMAs when old one could be | |
828 | * extended instead. | |
829 | */ | |
830 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) | |
1da177e4 LT |
831 | return 0; |
832 | if (vma->vm_file != file) | |
833 | return 0; | |
834 | if (vma->vm_ops && vma->vm_ops->close) | |
835 | return 0; | |
19a809af AA |
836 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) |
837 | return 0; | |
1da177e4 LT |
838 | return 1; |
839 | } | |
840 | ||
841 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
965f55de SL |
842 | struct anon_vma *anon_vma2, |
843 | struct vm_area_struct *vma) | |
1da177e4 | 844 | { |
965f55de SL |
845 | /* |
846 | * The list_is_singular() test is to avoid merging VMA cloned from | |
847 | * parents. This can improve scalability caused by anon_vma lock. | |
848 | */ | |
849 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
850 | list_is_singular(&vma->anon_vma_chain))) | |
851 | return 1; | |
852 | return anon_vma1 == anon_vma2; | |
1da177e4 LT |
853 | } |
854 | ||
855 | /* | |
856 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
857 | * in front of (at a lower virtual address and file offset than) the vma. | |
858 | * | |
859 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
860 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
861 | * | |
862 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
863 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
864 | * wrap, nor mmaps which cover the final page at index -1UL. | |
865 | */ | |
866 | static int | |
867 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
868 | struct anon_vma *anon_vma, struct file *file, |
869 | pgoff_t vm_pgoff, | |
870 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 871 | { |
19a809af | 872 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 873 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
874 | if (vma->vm_pgoff == vm_pgoff) |
875 | return 1; | |
876 | } | |
877 | return 0; | |
878 | } | |
879 | ||
880 | /* | |
881 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
882 | * beyond (at a higher virtual address and file offset than) the vma. | |
883 | * | |
884 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
885 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
886 | */ | |
887 | static int | |
888 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
889 | struct anon_vma *anon_vma, struct file *file, |
890 | pgoff_t vm_pgoff, | |
891 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 892 | { |
19a809af | 893 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 894 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 895 | pgoff_t vm_pglen; |
d6e93217 | 896 | vm_pglen = vma_pages(vma); |
1da177e4 LT |
897 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
898 | return 1; | |
899 | } | |
900 | return 0; | |
901 | } | |
902 | ||
903 | /* | |
904 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
905 | * whether that can be merged with its predecessor or its successor. | |
906 | * Or both (it neatly fills a hole). | |
907 | * | |
908 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
909 | * certain not to be mapped by the time vma_merge is called; but when | |
910 | * called for mprotect, it is certain to be already mapped (either at | |
911 | * an offset within prev, or at the start of next), and the flags of | |
912 | * this area are about to be changed to vm_flags - and the no-change | |
913 | * case has already been eliminated. | |
914 | * | |
915 | * The following mprotect cases have to be considered, where AAAA is | |
916 | * the area passed down from mprotect_fixup, never extending beyond one | |
917 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
918 | * | |
919 | * AAAA AAAA AAAA AAAA | |
920 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
921 | * cannot merge might become might become might become | |
922 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
923 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
924 | * mremap move: PPPPNNNNNNNN 8 | |
925 | * AAAA | |
926 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
927 | * might become case 1 below case 2 below case 3 below | |
928 | * | |
929 | * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: | |
930 | * mprotect_fixup updates vm_flags & vm_page_prot on successful return. | |
931 | */ | |
932 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
933 | struct vm_area_struct *prev, unsigned long addr, | |
934 | unsigned long end, unsigned long vm_flags, | |
cc71aba3 | 935 | struct anon_vma *anon_vma, struct file *file, |
19a809af AA |
936 | pgoff_t pgoff, struct mempolicy *policy, |
937 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 LT |
938 | { |
939 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
940 | struct vm_area_struct *area, *next; | |
5beb4930 | 941 | int err; |
1da177e4 LT |
942 | |
943 | /* | |
944 | * We later require that vma->vm_flags == vm_flags, | |
945 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
946 | */ | |
947 | if (vm_flags & VM_SPECIAL) | |
948 | return NULL; | |
949 | ||
950 | if (prev) | |
951 | next = prev->vm_next; | |
952 | else | |
953 | next = mm->mmap; | |
954 | area = next; | |
955 | if (next && next->vm_end == end) /* cases 6, 7, 8 */ | |
956 | next = next->vm_next; | |
957 | ||
958 | /* | |
959 | * Can it merge with the predecessor? | |
960 | */ | |
961 | if (prev && prev->vm_end == addr && | |
cc71aba3 | 962 | mpol_equal(vma_policy(prev), policy) && |
1da177e4 | 963 | can_vma_merge_after(prev, vm_flags, |
19a809af AA |
964 | anon_vma, file, pgoff, |
965 | vm_userfaultfd_ctx)) { | |
1da177e4 LT |
966 | /* |
967 | * OK, it can. Can we now merge in the successor as well? | |
968 | */ | |
969 | if (next && end == next->vm_start && | |
970 | mpol_equal(policy, vma_policy(next)) && | |
971 | can_vma_merge_before(next, vm_flags, | |
19a809af AA |
972 | anon_vma, file, |
973 | pgoff+pglen, | |
974 | vm_userfaultfd_ctx) && | |
1da177e4 | 975 | is_mergeable_anon_vma(prev->anon_vma, |
965f55de | 976 | next->anon_vma, NULL)) { |
1da177e4 | 977 | /* cases 1, 6 */ |
5beb4930 | 978 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
979 | next->vm_end, prev->vm_pgoff, NULL); |
980 | } else /* cases 2, 5, 7 */ | |
5beb4930 | 981 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 | 982 | end, prev->vm_pgoff, NULL); |
5beb4930 RR |
983 | if (err) |
984 | return NULL; | |
6d50e60c | 985 | khugepaged_enter_vma_merge(prev, vm_flags); |
1da177e4 LT |
986 | return prev; |
987 | } | |
988 | ||
989 | /* | |
990 | * Can this new request be merged in front of next? | |
991 | */ | |
992 | if (next && end == next->vm_start && | |
cc71aba3 | 993 | mpol_equal(policy, vma_policy(next)) && |
1da177e4 | 994 | can_vma_merge_before(next, vm_flags, |
19a809af AA |
995 | anon_vma, file, pgoff+pglen, |
996 | vm_userfaultfd_ctx)) { | |
1da177e4 | 997 | if (prev && addr < prev->vm_end) /* case 4 */ |
5beb4930 | 998 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
999 | addr, prev->vm_pgoff, NULL); |
1000 | else /* cases 3, 8 */ | |
5beb4930 | 1001 | err = vma_adjust(area, addr, next->vm_end, |
1da177e4 | 1002 | next->vm_pgoff - pglen, NULL); |
5beb4930 RR |
1003 | if (err) |
1004 | return NULL; | |
6d50e60c | 1005 | khugepaged_enter_vma_merge(area, vm_flags); |
1da177e4 LT |
1006 | return area; |
1007 | } | |
1008 | ||
1009 | return NULL; | |
1010 | } | |
1011 | ||
d0e9fe17 LT |
1012 | /* |
1013 | * Rough compatbility check to quickly see if it's even worth looking | |
1014 | * at sharing an anon_vma. | |
1015 | * | |
1016 | * They need to have the same vm_file, and the flags can only differ | |
1017 | * in things that mprotect may change. | |
1018 | * | |
1019 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1020 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1021 | * there is a vm_ops->close() function, because that indicates that the | |
1022 | * driver is doing some kind of reference counting. But that doesn't | |
1023 | * really matter for the anon_vma sharing case. | |
1024 | */ | |
1025 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1026 | { | |
1027 | return a->vm_end == b->vm_start && | |
1028 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1029 | a->vm_file == b->vm_file && | |
34228d47 | 1030 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && |
d0e9fe17 LT |
1031 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1036 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1037 | * the same as 'old', the other will be the new one that is trying | |
1038 | * to share the anon_vma. | |
1039 | * | |
1040 | * NOTE! This runs with mm_sem held for reading, so it is possible that | |
1041 | * the anon_vma of 'old' is concurrently in the process of being set up | |
1042 | * by another page fault trying to merge _that_. But that's ok: if it | |
1043 | * is being set up, that automatically means that it will be a singleton | |
1044 | * acceptable for merging, so we can do all of this optimistically. But | |
4db0c3c2 | 1045 | * we do that READ_ONCE() to make sure that we never re-load the pointer. |
d0e9fe17 LT |
1046 | * |
1047 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1048 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1049 | * is to return an anon_vma that is "complex" due to having gone through | |
1050 | * a fork). | |
1051 | * | |
1052 | * We also make sure that the two vma's are compatible (adjacent, | |
1053 | * and with the same memory policies). That's all stable, even with just | |
1054 | * a read lock on the mm_sem. | |
1055 | */ | |
1056 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1057 | { | |
1058 | if (anon_vma_compatible(a, b)) { | |
4db0c3c2 | 1059 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); |
d0e9fe17 LT |
1060 | |
1061 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1062 | return anon_vma; | |
1063 | } | |
1064 | return NULL; | |
1065 | } | |
1066 | ||
1da177e4 LT |
1067 | /* |
1068 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1069 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1070 | * to allocate a new anon_vma. It checks because a repetitive | |
1071 | * sequence of mprotects and faults may otherwise lead to distinct | |
1072 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1073 | * mprotect. | |
1074 | */ | |
1075 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1076 | { | |
d0e9fe17 | 1077 | struct anon_vma *anon_vma; |
1da177e4 | 1078 | struct vm_area_struct *near; |
1da177e4 LT |
1079 | |
1080 | near = vma->vm_next; | |
1081 | if (!near) | |
1082 | goto try_prev; | |
1083 | ||
d0e9fe17 LT |
1084 | anon_vma = reusable_anon_vma(near, vma, near); |
1085 | if (anon_vma) | |
1086 | return anon_vma; | |
1da177e4 | 1087 | try_prev: |
9be34c9d | 1088 | near = vma->vm_prev; |
1da177e4 LT |
1089 | if (!near) |
1090 | goto none; | |
1091 | ||
d0e9fe17 LT |
1092 | anon_vma = reusable_anon_vma(near, near, vma); |
1093 | if (anon_vma) | |
1094 | return anon_vma; | |
1da177e4 LT |
1095 | none: |
1096 | /* | |
1097 | * There's no absolute need to look only at touching neighbours: | |
1098 | * we could search further afield for "compatible" anon_vmas. | |
1099 | * But it would probably just be a waste of time searching, | |
1100 | * or lead to too many vmas hanging off the same anon_vma. | |
1101 | * We're trying to allow mprotect remerging later on, | |
1102 | * not trying to minimize memory used for anon_vmas. | |
1103 | */ | |
1104 | return NULL; | |
1105 | } | |
1106 | ||
40401530 AV |
1107 | /* |
1108 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1109 | * low as possible but still greater than mmap_min_addr | |
1110 | */ | |
1111 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1112 | { | |
1113 | hint &= PAGE_MASK; | |
1114 | if (((void *)hint != NULL) && | |
1115 | (hint < mmap_min_addr)) | |
1116 | return PAGE_ALIGN(mmap_min_addr); | |
1117 | return hint; | |
1118 | } | |
1119 | ||
363ee17f DB |
1120 | static inline int mlock_future_check(struct mm_struct *mm, |
1121 | unsigned long flags, | |
1122 | unsigned long len) | |
1123 | { | |
1124 | unsigned long locked, lock_limit; | |
1125 | ||
1126 | /* mlock MCL_FUTURE? */ | |
1127 | if (flags & VM_LOCKED) { | |
1128 | locked = len >> PAGE_SHIFT; | |
1129 | locked += mm->locked_vm; | |
1130 | lock_limit = rlimit(RLIMIT_MEMLOCK); | |
1131 | lock_limit >>= PAGE_SHIFT; | |
1132 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
1133 | return -EAGAIN; | |
1134 | } | |
1135 | return 0; | |
1136 | } | |
1137 | ||
1da177e4 | 1138 | /* |
27f5de79 | 1139 | * The caller must hold down_write(¤t->mm->mmap_sem). |
1da177e4 | 1140 | */ |
1fcfd8db | 1141 | unsigned long do_mmap(struct file *file, unsigned long addr, |
1da177e4 | 1142 | unsigned long len, unsigned long prot, |
1fcfd8db ON |
1143 | unsigned long flags, vm_flags_t vm_flags, |
1144 | unsigned long pgoff, unsigned long *populate) | |
1da177e4 | 1145 | { |
cc71aba3 | 1146 | struct mm_struct *mm = current->mm; |
62b5f7d0 | 1147 | int pkey = 0; |
1da177e4 | 1148 | |
41badc15 | 1149 | *populate = 0; |
bebeb3d6 | 1150 | |
e37609bb PK |
1151 | if (!len) |
1152 | return -EINVAL; | |
1153 | ||
1da177e4 LT |
1154 | /* |
1155 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1156 | * | |
1157 | * (the exception is when the underlying filesystem is noexec | |
1158 | * mounted, in which case we dont add PROT_EXEC.) | |
1159 | */ | |
1160 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
90f8572b | 1161 | if (!(file && path_noexec(&file->f_path))) |
1da177e4 LT |
1162 | prot |= PROT_EXEC; |
1163 | ||
7cd94146 EP |
1164 | if (!(flags & MAP_FIXED)) |
1165 | addr = round_hint_to_min(addr); | |
1166 | ||
1da177e4 LT |
1167 | /* Careful about overflows.. */ |
1168 | len = PAGE_ALIGN(len); | |
9206de95 | 1169 | if (!len) |
1da177e4 LT |
1170 | return -ENOMEM; |
1171 | ||
1172 | /* offset overflow? */ | |
1173 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
cc71aba3 | 1174 | return -EOVERFLOW; |
1da177e4 LT |
1175 | |
1176 | /* Too many mappings? */ | |
1177 | if (mm->map_count > sysctl_max_map_count) | |
1178 | return -ENOMEM; | |
1179 | ||
1180 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1181 | * that it represents a valid section of the address space. | |
1182 | */ | |
1183 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
de1741a1 | 1184 | if (offset_in_page(addr)) |
1da177e4 LT |
1185 | return addr; |
1186 | ||
62b5f7d0 DH |
1187 | if (prot == PROT_EXEC) { |
1188 | pkey = execute_only_pkey(mm); | |
1189 | if (pkey < 0) | |
1190 | pkey = 0; | |
1191 | } | |
1192 | ||
1da177e4 LT |
1193 | /* Do simple checking here so the lower-level routines won't have |
1194 | * to. we assume access permissions have been handled by the open | |
1195 | * of the memory object, so we don't do any here. | |
1196 | */ | |
62b5f7d0 | 1197 | vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | |
1da177e4 LT |
1198 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1199 | ||
cdf7b341 | 1200 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1201 | if (!can_do_mlock()) |
1202 | return -EPERM; | |
ba470de4 | 1203 | |
363ee17f DB |
1204 | if (mlock_future_check(mm, vm_flags, len)) |
1205 | return -EAGAIN; | |
1da177e4 | 1206 | |
1da177e4 | 1207 | if (file) { |
077bf22b ON |
1208 | struct inode *inode = file_inode(file); |
1209 | ||
1da177e4 LT |
1210 | switch (flags & MAP_TYPE) { |
1211 | case MAP_SHARED: | |
1212 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
1213 | return -EACCES; | |
1214 | ||
1215 | /* | |
1216 | * Make sure we don't allow writing to an append-only | |
1217 | * file.. | |
1218 | */ | |
1219 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1220 | return -EACCES; | |
1221 | ||
1222 | /* | |
1223 | * Make sure there are no mandatory locks on the file. | |
1224 | */ | |
d7a06983 | 1225 | if (locks_verify_locked(file)) |
1da177e4 LT |
1226 | return -EAGAIN; |
1227 | ||
1228 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
1229 | if (!(file->f_mode & FMODE_WRITE)) | |
1230 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
1231 | ||
1232 | /* fall through */ | |
1233 | case MAP_PRIVATE: | |
1234 | if (!(file->f_mode & FMODE_READ)) | |
1235 | return -EACCES; | |
90f8572b | 1236 | if (path_noexec(&file->f_path)) { |
80c5606c LT |
1237 | if (vm_flags & VM_EXEC) |
1238 | return -EPERM; | |
1239 | vm_flags &= ~VM_MAYEXEC; | |
1240 | } | |
80c5606c | 1241 | |
72c2d531 | 1242 | if (!file->f_op->mmap) |
80c5606c | 1243 | return -ENODEV; |
b2c56e4f ON |
1244 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1245 | return -EINVAL; | |
1da177e4 LT |
1246 | break; |
1247 | ||
1248 | default: | |
1249 | return -EINVAL; | |
1250 | } | |
1251 | } else { | |
1252 | switch (flags & MAP_TYPE) { | |
1253 | case MAP_SHARED: | |
b2c56e4f ON |
1254 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1255 | return -EINVAL; | |
ce363942 TH |
1256 | /* |
1257 | * Ignore pgoff. | |
1258 | */ | |
1259 | pgoff = 0; | |
1da177e4 LT |
1260 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1261 | break; | |
1262 | case MAP_PRIVATE: | |
1263 | /* | |
1264 | * Set pgoff according to addr for anon_vma. | |
1265 | */ | |
1266 | pgoff = addr >> PAGE_SHIFT; | |
1267 | break; | |
1268 | default: | |
1269 | return -EINVAL; | |
1270 | } | |
1271 | } | |
1272 | ||
c22c0d63 ML |
1273 | /* |
1274 | * Set 'VM_NORESERVE' if we should not account for the | |
1275 | * memory use of this mapping. | |
1276 | */ | |
1277 | if (flags & MAP_NORESERVE) { | |
1278 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1279 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1280 | vm_flags |= VM_NORESERVE; | |
1281 | ||
1282 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1283 | if (file && is_file_hugepages(file)) | |
1284 | vm_flags |= VM_NORESERVE; | |
1285 | } | |
1286 | ||
1287 | addr = mmap_region(file, addr, len, vm_flags, pgoff); | |
09a9f1d2 ML |
1288 | if (!IS_ERR_VALUE(addr) && |
1289 | ((vm_flags & VM_LOCKED) || | |
1290 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1291 | *populate = len; |
bebeb3d6 | 1292 | return addr; |
0165ab44 | 1293 | } |
6be5ceb0 | 1294 | |
66f0dc48 HD |
1295 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1296 | unsigned long, prot, unsigned long, flags, | |
1297 | unsigned long, fd, unsigned long, pgoff) | |
1298 | { | |
1299 | struct file *file = NULL; | |
1e3ee14b | 1300 | unsigned long retval; |
66f0dc48 HD |
1301 | |
1302 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1303 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1304 | file = fget(fd); |
1305 | if (!file) | |
1e3ee14b | 1306 | return -EBADF; |
af73e4d9 NH |
1307 | if (is_file_hugepages(file)) |
1308 | len = ALIGN(len, huge_page_size(hstate_file(file))); | |
493af578 JE |
1309 | retval = -EINVAL; |
1310 | if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) | |
1311 | goto out_fput; | |
66f0dc48 HD |
1312 | } else if (flags & MAP_HUGETLB) { |
1313 | struct user_struct *user = NULL; | |
c103a4dc | 1314 | struct hstate *hs; |
af73e4d9 | 1315 | |
c103a4dc | 1316 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); |
091d0d55 LZ |
1317 | if (!hs) |
1318 | return -EINVAL; | |
1319 | ||
1320 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1321 | /* |
1322 | * VM_NORESERVE is used because the reservations will be | |
1323 | * taken when vm_ops->mmap() is called | |
1324 | * A dummy user value is used because we are not locking | |
1325 | * memory so no accounting is necessary | |
1326 | */ | |
af73e4d9 | 1327 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f AK |
1328 | VM_NORESERVE, |
1329 | &user, HUGETLB_ANONHUGE_INODE, | |
1330 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); | |
66f0dc48 HD |
1331 | if (IS_ERR(file)) |
1332 | return PTR_ERR(file); | |
1333 | } | |
1334 | ||
1335 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1336 | ||
9fbeb5ab | 1337 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
493af578 | 1338 | out_fput: |
66f0dc48 HD |
1339 | if (file) |
1340 | fput(file); | |
66f0dc48 HD |
1341 | return retval; |
1342 | } | |
1343 | ||
a4679373 CH |
1344 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1345 | struct mmap_arg_struct { | |
1346 | unsigned long addr; | |
1347 | unsigned long len; | |
1348 | unsigned long prot; | |
1349 | unsigned long flags; | |
1350 | unsigned long fd; | |
1351 | unsigned long offset; | |
1352 | }; | |
1353 | ||
1354 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1355 | { | |
1356 | struct mmap_arg_struct a; | |
1357 | ||
1358 | if (copy_from_user(&a, arg, sizeof(a))) | |
1359 | return -EFAULT; | |
de1741a1 | 1360 | if (offset_in_page(a.offset)) |
a4679373 CH |
1361 | return -EINVAL; |
1362 | ||
1363 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1364 | a.offset >> PAGE_SHIFT); | |
1365 | } | |
1366 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1367 | ||
4e950f6f AD |
1368 | /* |
1369 | * Some shared mappigns will want the pages marked read-only | |
1370 | * to track write events. If so, we'll downgrade vm_page_prot | |
1371 | * to the private version (using protection_map[] without the | |
1372 | * VM_SHARED bit). | |
1373 | */ | |
1374 | int vma_wants_writenotify(struct vm_area_struct *vma) | |
1375 | { | |
ca16d140 | 1376 | vm_flags_t vm_flags = vma->vm_flags; |
8a04446a | 1377 | const struct vm_operations_struct *vm_ops = vma->vm_ops; |
4e950f6f AD |
1378 | |
1379 | /* If it was private or non-writable, the write bit is already clear */ | |
1380 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1381 | return 0; | |
1382 | ||
1383 | /* The backer wishes to know when pages are first written to? */ | |
8a04446a | 1384 | if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) |
4e950f6f AD |
1385 | return 1; |
1386 | ||
64e45507 PF |
1387 | /* The open routine did something to the protections that pgprot_modify |
1388 | * won't preserve? */ | |
4e950f6f | 1389 | if (pgprot_val(vma->vm_page_prot) != |
64e45507 | 1390 | pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags))) |
4e950f6f AD |
1391 | return 0; |
1392 | ||
64e45507 PF |
1393 | /* Do we need to track softdirty? */ |
1394 | if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) | |
1395 | return 1; | |
1396 | ||
4e950f6f | 1397 | /* Specialty mapping? */ |
4b6e1e37 | 1398 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1399 | return 0; |
1400 | ||
1401 | /* Can the mapping track the dirty pages? */ | |
1402 | return vma->vm_file && vma->vm_file->f_mapping && | |
1403 | mapping_cap_account_dirty(vma->vm_file->f_mapping); | |
1404 | } | |
1405 | ||
fc8744ad LT |
1406 | /* |
1407 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1408 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1409 | */ |
ca16d140 | 1410 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1411 | { |
5a6fe125 MG |
1412 | /* |
1413 | * hugetlb has its own accounting separate from the core VM | |
1414 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1415 | */ | |
1416 | if (file && is_file_hugepages(file)) | |
1417 | return 0; | |
1418 | ||
fc8744ad LT |
1419 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1420 | } | |
1421 | ||
0165ab44 | 1422 | unsigned long mmap_region(struct file *file, unsigned long addr, |
c22c0d63 | 1423 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) |
0165ab44 MS |
1424 | { |
1425 | struct mm_struct *mm = current->mm; | |
1426 | struct vm_area_struct *vma, *prev; | |
0165ab44 MS |
1427 | int error; |
1428 | struct rb_node **rb_link, *rb_parent; | |
1429 | unsigned long charged = 0; | |
0165ab44 | 1430 | |
e8420a8e | 1431 | /* Check against address space limit. */ |
84638335 | 1432 | if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { |
e8420a8e CH |
1433 | unsigned long nr_pages; |
1434 | ||
1435 | /* | |
1436 | * MAP_FIXED may remove pages of mappings that intersects with | |
1437 | * requested mapping. Account for the pages it would unmap. | |
1438 | */ | |
e8420a8e CH |
1439 | nr_pages = count_vma_pages_range(mm, addr, addr + len); |
1440 | ||
84638335 KK |
1441 | if (!may_expand_vm(mm, vm_flags, |
1442 | (len >> PAGE_SHIFT) - nr_pages)) | |
e8420a8e CH |
1443 | return -ENOMEM; |
1444 | } | |
1445 | ||
1da177e4 | 1446 | /* Clear old maps */ |
9fcd1457 RV |
1447 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
1448 | &rb_parent)) { | |
1da177e4 LT |
1449 | if (do_munmap(mm, addr, len)) |
1450 | return -ENOMEM; | |
1da177e4 LT |
1451 | } |
1452 | ||
fc8744ad LT |
1453 | /* |
1454 | * Private writable mapping: check memory availability | |
1455 | */ | |
5a6fe125 | 1456 | if (accountable_mapping(file, vm_flags)) { |
fc8744ad | 1457 | charged = len >> PAGE_SHIFT; |
191c5424 | 1458 | if (security_vm_enough_memory_mm(mm, charged)) |
fc8744ad LT |
1459 | return -ENOMEM; |
1460 | vm_flags |= VM_ACCOUNT; | |
1da177e4 LT |
1461 | } |
1462 | ||
1463 | /* | |
de33c8db | 1464 | * Can we just expand an old mapping? |
1da177e4 | 1465 | */ |
19a809af AA |
1466 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, |
1467 | NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); | |
de33c8db LT |
1468 | if (vma) |
1469 | goto out; | |
1da177e4 LT |
1470 | |
1471 | /* | |
1472 | * Determine the object being mapped and call the appropriate | |
1473 | * specific mapper. the address has already been validated, but | |
1474 | * not unmapped, but the maps are removed from the list. | |
1475 | */ | |
c5e3b83e | 1476 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1477 | if (!vma) { |
1478 | error = -ENOMEM; | |
1479 | goto unacct_error; | |
1480 | } | |
1da177e4 LT |
1481 | |
1482 | vma->vm_mm = mm; | |
1483 | vma->vm_start = addr; | |
1484 | vma->vm_end = addr + len; | |
1485 | vma->vm_flags = vm_flags; | |
3ed75eb8 | 1486 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
1da177e4 | 1487 | vma->vm_pgoff = pgoff; |
5beb4930 | 1488 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
1489 | |
1490 | if (file) { | |
1da177e4 LT |
1491 | if (vm_flags & VM_DENYWRITE) { |
1492 | error = deny_write_access(file); | |
1493 | if (error) | |
1494 | goto free_vma; | |
1da177e4 | 1495 | } |
4bb5f5d9 DH |
1496 | if (vm_flags & VM_SHARED) { |
1497 | error = mapping_map_writable(file->f_mapping); | |
1498 | if (error) | |
1499 | goto allow_write_and_free_vma; | |
1500 | } | |
1501 | ||
1502 | /* ->mmap() can change vma->vm_file, but must guarantee that | |
1503 | * vma_link() below can deny write-access if VM_DENYWRITE is set | |
1504 | * and map writably if VM_SHARED is set. This usually means the | |
1505 | * new file must not have been exposed to user-space, yet. | |
1506 | */ | |
cb0942b8 | 1507 | vma->vm_file = get_file(file); |
1da177e4 LT |
1508 | error = file->f_op->mmap(file, vma); |
1509 | if (error) | |
1510 | goto unmap_and_free_vma; | |
f8dbf0a7 HS |
1511 | |
1512 | /* Can addr have changed?? | |
1513 | * | |
1514 | * Answer: Yes, several device drivers can do it in their | |
1515 | * f_op->mmap method. -DaveM | |
2897b4d2 JK |
1516 | * Bug: If addr is changed, prev, rb_link, rb_parent should |
1517 | * be updated for vma_link() | |
f8dbf0a7 | 1518 | */ |
2897b4d2 JK |
1519 | WARN_ON_ONCE(addr != vma->vm_start); |
1520 | ||
f8dbf0a7 | 1521 | addr = vma->vm_start; |
f8dbf0a7 | 1522 | vm_flags = vma->vm_flags; |
1da177e4 LT |
1523 | } else if (vm_flags & VM_SHARED) { |
1524 | error = shmem_zero_setup(vma); | |
1525 | if (error) | |
1526 | goto free_vma; | |
1527 | } | |
1528 | ||
de33c8db | 1529 | vma_link(mm, vma, prev, rb_link, rb_parent); |
4d3d5b41 | 1530 | /* Once vma denies write, undo our temporary denial count */ |
4bb5f5d9 DH |
1531 | if (file) { |
1532 | if (vm_flags & VM_SHARED) | |
1533 | mapping_unmap_writable(file->f_mapping); | |
1534 | if (vm_flags & VM_DENYWRITE) | |
1535 | allow_write_access(file); | |
1536 | } | |
e8686772 | 1537 | file = vma->vm_file; |
4d3d5b41 | 1538 | out: |
cdd6c482 | 1539 | perf_event_mmap(vma); |
0a4a9391 | 1540 | |
84638335 | 1541 | vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); |
1da177e4 | 1542 | if (vm_flags & VM_LOCKED) { |
bebeb3d6 ML |
1543 | if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || |
1544 | vma == get_gate_vma(current->mm))) | |
06f9d8c2 | 1545 | mm->locked_vm += (len >> PAGE_SHIFT); |
bebeb3d6 | 1546 | else |
de60f5f1 | 1547 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
bebeb3d6 | 1548 | } |
2b144498 | 1549 | |
c7a3a88c ON |
1550 | if (file) |
1551 | uprobe_mmap(vma); | |
2b144498 | 1552 | |
d9104d1c CG |
1553 | /* |
1554 | * New (or expanded) vma always get soft dirty status. | |
1555 | * Otherwise user-space soft-dirty page tracker won't | |
1556 | * be able to distinguish situation when vma area unmapped, | |
1557 | * then new mapped in-place (which must be aimed as | |
1558 | * a completely new data area). | |
1559 | */ | |
1560 | vma->vm_flags |= VM_SOFTDIRTY; | |
1561 | ||
64e45507 PF |
1562 | vma_set_page_prot(vma); |
1563 | ||
1da177e4 LT |
1564 | return addr; |
1565 | ||
1566 | unmap_and_free_vma: | |
1da177e4 LT |
1567 | vma->vm_file = NULL; |
1568 | fput(file); | |
1569 | ||
1570 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1571 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1572 | charged = 0; | |
4bb5f5d9 DH |
1573 | if (vm_flags & VM_SHARED) |
1574 | mapping_unmap_writable(file->f_mapping); | |
1575 | allow_write_and_free_vma: | |
1576 | if (vm_flags & VM_DENYWRITE) | |
1577 | allow_write_access(file); | |
1da177e4 LT |
1578 | free_vma: |
1579 | kmem_cache_free(vm_area_cachep, vma); | |
1580 | unacct_error: | |
1581 | if (charged) | |
1582 | vm_unacct_memory(charged); | |
1583 | return error; | |
1584 | } | |
1585 | ||
db4fbfb9 ML |
1586 | unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
1587 | { | |
1588 | /* | |
1589 | * We implement the search by looking for an rbtree node that | |
1590 | * immediately follows a suitable gap. That is, | |
1591 | * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; | |
1592 | * - gap_end = vma->vm_start >= info->low_limit + length; | |
1593 | * - gap_end - gap_start >= length | |
1594 | */ | |
1595 | ||
1596 | struct mm_struct *mm = current->mm; | |
1597 | struct vm_area_struct *vma; | |
1598 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1599 | ||
1600 | /* Adjust search length to account for worst case alignment overhead */ | |
1601 | length = info->length + info->align_mask; | |
1602 | if (length < info->length) | |
1603 | return -ENOMEM; | |
1604 | ||
1605 | /* Adjust search limits by the desired length */ | |
1606 | if (info->high_limit < length) | |
1607 | return -ENOMEM; | |
1608 | high_limit = info->high_limit - length; | |
1609 | ||
1610 | if (info->low_limit > high_limit) | |
1611 | return -ENOMEM; | |
1612 | low_limit = info->low_limit + length; | |
1613 | ||
1614 | /* Check if rbtree root looks promising */ | |
1615 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1616 | goto check_highest; | |
1617 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1618 | if (vma->rb_subtree_gap < length) | |
1619 | goto check_highest; | |
1620 | ||
1621 | while (true) { | |
1622 | /* Visit left subtree if it looks promising */ | |
1623 | gap_end = vma->vm_start; | |
1624 | if (gap_end >= low_limit && vma->vm_rb.rb_left) { | |
1625 | struct vm_area_struct *left = | |
1626 | rb_entry(vma->vm_rb.rb_left, | |
1627 | struct vm_area_struct, vm_rb); | |
1628 | if (left->rb_subtree_gap >= length) { | |
1629 | vma = left; | |
1630 | continue; | |
1631 | } | |
1632 | } | |
1633 | ||
1634 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1635 | check_current: | |
1636 | /* Check if current node has a suitable gap */ | |
1637 | if (gap_start > high_limit) | |
1638 | return -ENOMEM; | |
1639 | if (gap_end >= low_limit && gap_end - gap_start >= length) | |
1640 | goto found; | |
1641 | ||
1642 | /* Visit right subtree if it looks promising */ | |
1643 | if (vma->vm_rb.rb_right) { | |
1644 | struct vm_area_struct *right = | |
1645 | rb_entry(vma->vm_rb.rb_right, | |
1646 | struct vm_area_struct, vm_rb); | |
1647 | if (right->rb_subtree_gap >= length) { | |
1648 | vma = right; | |
1649 | continue; | |
1650 | } | |
1651 | } | |
1652 | ||
1653 | /* Go back up the rbtree to find next candidate node */ | |
1654 | while (true) { | |
1655 | struct rb_node *prev = &vma->vm_rb; | |
1656 | if (!rb_parent(prev)) | |
1657 | goto check_highest; | |
1658 | vma = rb_entry(rb_parent(prev), | |
1659 | struct vm_area_struct, vm_rb); | |
1660 | if (prev == vma->vm_rb.rb_left) { | |
1661 | gap_start = vma->vm_prev->vm_end; | |
1662 | gap_end = vma->vm_start; | |
1663 | goto check_current; | |
1664 | } | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | check_highest: | |
1669 | /* Check highest gap, which does not precede any rbtree node */ | |
1670 | gap_start = mm->highest_vm_end; | |
1671 | gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ | |
1672 | if (gap_start > high_limit) | |
1673 | return -ENOMEM; | |
1674 | ||
1675 | found: | |
1676 | /* We found a suitable gap. Clip it with the original low_limit. */ | |
1677 | if (gap_start < info->low_limit) | |
1678 | gap_start = info->low_limit; | |
1679 | ||
1680 | /* Adjust gap address to the desired alignment */ | |
1681 | gap_start += (info->align_offset - gap_start) & info->align_mask; | |
1682 | ||
1683 | VM_BUG_ON(gap_start + info->length > info->high_limit); | |
1684 | VM_BUG_ON(gap_start + info->length > gap_end); | |
1685 | return gap_start; | |
1686 | } | |
1687 | ||
1688 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) | |
1689 | { | |
1690 | struct mm_struct *mm = current->mm; | |
1691 | struct vm_area_struct *vma; | |
1692 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1693 | ||
1694 | /* Adjust search length to account for worst case alignment overhead */ | |
1695 | length = info->length + info->align_mask; | |
1696 | if (length < info->length) | |
1697 | return -ENOMEM; | |
1698 | ||
1699 | /* | |
1700 | * Adjust search limits by the desired length. | |
1701 | * See implementation comment at top of unmapped_area(). | |
1702 | */ | |
1703 | gap_end = info->high_limit; | |
1704 | if (gap_end < length) | |
1705 | return -ENOMEM; | |
1706 | high_limit = gap_end - length; | |
1707 | ||
1708 | if (info->low_limit > high_limit) | |
1709 | return -ENOMEM; | |
1710 | low_limit = info->low_limit + length; | |
1711 | ||
1712 | /* Check highest gap, which does not precede any rbtree node */ | |
1713 | gap_start = mm->highest_vm_end; | |
1714 | if (gap_start <= high_limit) | |
1715 | goto found_highest; | |
1716 | ||
1717 | /* Check if rbtree root looks promising */ | |
1718 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1719 | return -ENOMEM; | |
1720 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1721 | if (vma->rb_subtree_gap < length) | |
1722 | return -ENOMEM; | |
1723 | ||
1724 | while (true) { | |
1725 | /* Visit right subtree if it looks promising */ | |
1726 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1727 | if (gap_start <= high_limit && vma->vm_rb.rb_right) { | |
1728 | struct vm_area_struct *right = | |
1729 | rb_entry(vma->vm_rb.rb_right, | |
1730 | struct vm_area_struct, vm_rb); | |
1731 | if (right->rb_subtree_gap >= length) { | |
1732 | vma = right; | |
1733 | continue; | |
1734 | } | |
1735 | } | |
1736 | ||
1737 | check_current: | |
1738 | /* Check if current node has a suitable gap */ | |
1739 | gap_end = vma->vm_start; | |
1740 | if (gap_end < low_limit) | |
1741 | return -ENOMEM; | |
1742 | if (gap_start <= high_limit && gap_end - gap_start >= length) | |
1743 | goto found; | |
1744 | ||
1745 | /* Visit left subtree if it looks promising */ | |
1746 | if (vma->vm_rb.rb_left) { | |
1747 | struct vm_area_struct *left = | |
1748 | rb_entry(vma->vm_rb.rb_left, | |
1749 | struct vm_area_struct, vm_rb); | |
1750 | if (left->rb_subtree_gap >= length) { | |
1751 | vma = left; | |
1752 | continue; | |
1753 | } | |
1754 | } | |
1755 | ||
1756 | /* Go back up the rbtree to find next candidate node */ | |
1757 | while (true) { | |
1758 | struct rb_node *prev = &vma->vm_rb; | |
1759 | if (!rb_parent(prev)) | |
1760 | return -ENOMEM; | |
1761 | vma = rb_entry(rb_parent(prev), | |
1762 | struct vm_area_struct, vm_rb); | |
1763 | if (prev == vma->vm_rb.rb_right) { | |
1764 | gap_start = vma->vm_prev ? | |
1765 | vma->vm_prev->vm_end : 0; | |
1766 | goto check_current; | |
1767 | } | |
1768 | } | |
1769 | } | |
1770 | ||
1771 | found: | |
1772 | /* We found a suitable gap. Clip it with the original high_limit. */ | |
1773 | if (gap_end > info->high_limit) | |
1774 | gap_end = info->high_limit; | |
1775 | ||
1776 | found_highest: | |
1777 | /* Compute highest gap address at the desired alignment */ | |
1778 | gap_end -= info->length; | |
1779 | gap_end -= (gap_end - info->align_offset) & info->align_mask; | |
1780 | ||
1781 | VM_BUG_ON(gap_end < info->low_limit); | |
1782 | VM_BUG_ON(gap_end < gap_start); | |
1783 | return gap_end; | |
1784 | } | |
1785 | ||
1da177e4 LT |
1786 | /* Get an address range which is currently unmapped. |
1787 | * For shmat() with addr=0. | |
1788 | * | |
1789 | * Ugly calling convention alert: | |
1790 | * Return value with the low bits set means error value, | |
1791 | * ie | |
1792 | * if (ret & ~PAGE_MASK) | |
1793 | * error = ret; | |
1794 | * | |
1795 | * This function "knows" that -ENOMEM has the bits set. | |
1796 | */ | |
1797 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1798 | unsigned long | |
1799 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1800 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1801 | { | |
1802 | struct mm_struct *mm = current->mm; | |
1803 | struct vm_area_struct *vma; | |
db4fbfb9 | 1804 | struct vm_unmapped_area_info info; |
1da177e4 | 1805 | |
2afc745f | 1806 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1807 | return -ENOMEM; |
1808 | ||
06abdfb4 BH |
1809 | if (flags & MAP_FIXED) |
1810 | return addr; | |
1811 | ||
1da177e4 LT |
1812 | if (addr) { |
1813 | addr = PAGE_ALIGN(addr); | |
1814 | vma = find_vma(mm, addr); | |
2afc745f | 1815 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1816 | (!vma || addr + len <= vma->vm_start)) |
1817 | return addr; | |
1818 | } | |
1da177e4 | 1819 | |
db4fbfb9 ML |
1820 | info.flags = 0; |
1821 | info.length = len; | |
4e99b021 | 1822 | info.low_limit = mm->mmap_base; |
db4fbfb9 ML |
1823 | info.high_limit = TASK_SIZE; |
1824 | info.align_mask = 0; | |
1825 | return vm_unmapped_area(&info); | |
1da177e4 | 1826 | } |
cc71aba3 | 1827 | #endif |
1da177e4 | 1828 | |
1da177e4 LT |
1829 | /* |
1830 | * This mmap-allocator allocates new areas top-down from below the | |
1831 | * stack's low limit (the base): | |
1832 | */ | |
1833 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1834 | unsigned long | |
1835 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1836 | const unsigned long len, const unsigned long pgoff, | |
1837 | const unsigned long flags) | |
1838 | { | |
1839 | struct vm_area_struct *vma; | |
1840 | struct mm_struct *mm = current->mm; | |
db4fbfb9 ML |
1841 | unsigned long addr = addr0; |
1842 | struct vm_unmapped_area_info info; | |
1da177e4 LT |
1843 | |
1844 | /* requested length too big for entire address space */ | |
2afc745f | 1845 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1846 | return -ENOMEM; |
1847 | ||
06abdfb4 BH |
1848 | if (flags & MAP_FIXED) |
1849 | return addr; | |
1850 | ||
1da177e4 LT |
1851 | /* requesting a specific address */ |
1852 | if (addr) { | |
1853 | addr = PAGE_ALIGN(addr); | |
1854 | vma = find_vma(mm, addr); | |
2afc745f | 1855 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1856 | (!vma || addr + len <= vma->vm_start)) |
1857 | return addr; | |
1858 | } | |
1859 | ||
db4fbfb9 ML |
1860 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
1861 | info.length = len; | |
2afc745f | 1862 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); |
db4fbfb9 ML |
1863 | info.high_limit = mm->mmap_base; |
1864 | info.align_mask = 0; | |
1865 | addr = vm_unmapped_area(&info); | |
b716ad95 | 1866 | |
1da177e4 LT |
1867 | /* |
1868 | * A failed mmap() very likely causes application failure, | |
1869 | * so fall back to the bottom-up function here. This scenario | |
1870 | * can happen with large stack limits and large mmap() | |
1871 | * allocations. | |
1872 | */ | |
de1741a1 | 1873 | if (offset_in_page(addr)) { |
db4fbfb9 ML |
1874 | VM_BUG_ON(addr != -ENOMEM); |
1875 | info.flags = 0; | |
1876 | info.low_limit = TASK_UNMAPPED_BASE; | |
1877 | info.high_limit = TASK_SIZE; | |
1878 | addr = vm_unmapped_area(&info); | |
1879 | } | |
1da177e4 LT |
1880 | |
1881 | return addr; | |
1882 | } | |
1883 | #endif | |
1884 | ||
1da177e4 LT |
1885 | unsigned long |
1886 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1887 | unsigned long pgoff, unsigned long flags) | |
1888 | { | |
06abdfb4 BH |
1889 | unsigned long (*get_area)(struct file *, unsigned long, |
1890 | unsigned long, unsigned long, unsigned long); | |
1891 | ||
9206de95 AV |
1892 | unsigned long error = arch_mmap_check(addr, len, flags); |
1893 | if (error) | |
1894 | return error; | |
1895 | ||
1896 | /* Careful about overflows.. */ | |
1897 | if (len > TASK_SIZE) | |
1898 | return -ENOMEM; | |
1899 | ||
06abdfb4 | 1900 | get_area = current->mm->get_unmapped_area; |
c01d5b30 HD |
1901 | if (file) { |
1902 | if (file->f_op->get_unmapped_area) | |
1903 | get_area = file->f_op->get_unmapped_area; | |
1904 | } else if (flags & MAP_SHARED) { | |
1905 | /* | |
1906 | * mmap_region() will call shmem_zero_setup() to create a file, | |
1907 | * so use shmem's get_unmapped_area in case it can be huge. | |
1908 | * do_mmap_pgoff() will clear pgoff, so match alignment. | |
1909 | */ | |
1910 | pgoff = 0; | |
1911 | get_area = shmem_get_unmapped_area; | |
1912 | } | |
1913 | ||
06abdfb4 BH |
1914 | addr = get_area(file, addr, len, pgoff, flags); |
1915 | if (IS_ERR_VALUE(addr)) | |
1916 | return addr; | |
1da177e4 | 1917 | |
07ab67c8 LT |
1918 | if (addr > TASK_SIZE - len) |
1919 | return -ENOMEM; | |
de1741a1 | 1920 | if (offset_in_page(addr)) |
07ab67c8 | 1921 | return -EINVAL; |
06abdfb4 | 1922 | |
9ac4ed4b AV |
1923 | error = security_mmap_addr(addr); |
1924 | return error ? error : addr; | |
1da177e4 LT |
1925 | } |
1926 | ||
1927 | EXPORT_SYMBOL(get_unmapped_area); | |
1928 | ||
1929 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
48aae425 | 1930 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 1931 | { |
615d6e87 DB |
1932 | struct rb_node *rb_node; |
1933 | struct vm_area_struct *vma; | |
1da177e4 | 1934 | |
841e31e5 | 1935 | /* Check the cache first. */ |
615d6e87 DB |
1936 | vma = vmacache_find(mm, addr); |
1937 | if (likely(vma)) | |
1938 | return vma; | |
841e31e5 | 1939 | |
615d6e87 | 1940 | rb_node = mm->mm_rb.rb_node; |
841e31e5 | 1941 | |
615d6e87 DB |
1942 | while (rb_node) { |
1943 | struct vm_area_struct *tmp; | |
1944 | ||
1945 | tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
1946 | ||
1947 | if (tmp->vm_end > addr) { | |
1948 | vma = tmp; | |
1949 | if (tmp->vm_start <= addr) | |
1950 | break; | |
1951 | rb_node = rb_node->rb_left; | |
1952 | } else | |
1953 | rb_node = rb_node->rb_right; | |
1da177e4 | 1954 | } |
615d6e87 DB |
1955 | |
1956 | if (vma) | |
1957 | vmacache_update(addr, vma); | |
1da177e4 LT |
1958 | return vma; |
1959 | } | |
1960 | ||
1961 | EXPORT_SYMBOL(find_vma); | |
1962 | ||
6bd4837d KM |
1963 | /* |
1964 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. | |
6bd4837d | 1965 | */ |
1da177e4 LT |
1966 | struct vm_area_struct * |
1967 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
1968 | struct vm_area_struct **pprev) | |
1969 | { | |
6bd4837d | 1970 | struct vm_area_struct *vma; |
1da177e4 | 1971 | |
6bd4837d | 1972 | vma = find_vma(mm, addr); |
83cd904d MP |
1973 | if (vma) { |
1974 | *pprev = vma->vm_prev; | |
1975 | } else { | |
1976 | struct rb_node *rb_node = mm->mm_rb.rb_node; | |
1977 | *pprev = NULL; | |
1978 | while (rb_node) { | |
1979 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
1980 | rb_node = rb_node->rb_right; | |
1981 | } | |
1982 | } | |
6bd4837d | 1983 | return vma; |
1da177e4 LT |
1984 | } |
1985 | ||
1986 | /* | |
1987 | * Verify that the stack growth is acceptable and | |
1988 | * update accounting. This is shared with both the | |
1989 | * grow-up and grow-down cases. | |
1990 | */ | |
48aae425 | 1991 | static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) |
1da177e4 LT |
1992 | { |
1993 | struct mm_struct *mm = vma->vm_mm; | |
1994 | struct rlimit *rlim = current->signal->rlim; | |
690eac53 | 1995 | unsigned long new_start, actual_size; |
1da177e4 LT |
1996 | |
1997 | /* address space limit tests */ | |
84638335 | 1998 | if (!may_expand_vm(mm, vma->vm_flags, grow)) |
1da177e4 LT |
1999 | return -ENOMEM; |
2000 | ||
2001 | /* Stack limit test */ | |
690eac53 LT |
2002 | actual_size = size; |
2003 | if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) | |
2004 | actual_size -= PAGE_SIZE; | |
4db0c3c2 | 2005 | if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur)) |
1da177e4 LT |
2006 | return -ENOMEM; |
2007 | ||
2008 | /* mlock limit tests */ | |
2009 | if (vma->vm_flags & VM_LOCKED) { | |
2010 | unsigned long locked; | |
2011 | unsigned long limit; | |
2012 | locked = mm->locked_vm + grow; | |
4db0c3c2 | 2013 | limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); |
59e99e5b | 2014 | limit >>= PAGE_SHIFT; |
1da177e4 LT |
2015 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
2016 | return -ENOMEM; | |
2017 | } | |
2018 | ||
0d59a01b AL |
2019 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
2020 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
2021 | vma->vm_end - size; | |
2022 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
2023 | return -EFAULT; | |
2024 | ||
1da177e4 LT |
2025 | /* |
2026 | * Overcommit.. This must be the final test, as it will | |
2027 | * update security statistics. | |
2028 | */ | |
05fa199d | 2029 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
2030 | return -ENOMEM; |
2031 | ||
1da177e4 LT |
2032 | return 0; |
2033 | } | |
2034 | ||
46dea3d0 | 2035 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 2036 | /* |
46dea3d0 HD |
2037 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
2038 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 2039 | */ |
46dea3d0 | 2040 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 2041 | { |
09357814 | 2042 | struct mm_struct *mm = vma->vm_mm; |
12352d3c | 2043 | int error = 0; |
1da177e4 LT |
2044 | |
2045 | if (!(vma->vm_flags & VM_GROWSUP)) | |
2046 | return -EFAULT; | |
2047 | ||
12352d3c KK |
2048 | /* Guard against wrapping around to address 0. */ |
2049 | if (address < PAGE_ALIGN(address+4)) | |
2050 | address = PAGE_ALIGN(address+4); | |
2051 | else | |
2052 | return -ENOMEM; | |
2053 | ||
2054 | /* We must make sure the anon_vma is allocated. */ | |
1da177e4 LT |
2055 | if (unlikely(anon_vma_prepare(vma))) |
2056 | return -ENOMEM; | |
1da177e4 LT |
2057 | |
2058 | /* | |
2059 | * vma->vm_start/vm_end cannot change under us because the caller | |
2060 | * is required to hold the mmap_sem in read mode. We need the | |
2061 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2062 | */ | |
12352d3c | 2063 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2064 | |
2065 | /* Somebody else might have raced and expanded it already */ | |
2066 | if (address > vma->vm_end) { | |
2067 | unsigned long size, grow; | |
2068 | ||
2069 | size = address - vma->vm_start; | |
2070 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
2071 | ||
42c36f63 HD |
2072 | error = -ENOMEM; |
2073 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
2074 | error = acct_stack_growth(vma, size, grow); | |
2075 | if (!error) { | |
4128997b ML |
2076 | /* |
2077 | * vma_gap_update() doesn't support concurrent | |
2078 | * updates, but we only hold a shared mmap_sem | |
2079 | * lock here, so we need to protect against | |
2080 | * concurrent vma expansions. | |
12352d3c | 2081 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2082 | * we don't guarantee that all growable vmas |
2083 | * in a mm share the same root anon vma. | |
2084 | * So, we reuse mm->page_table_lock to guard | |
2085 | * against concurrent vma expansions. | |
2086 | */ | |
09357814 | 2087 | spin_lock(&mm->page_table_lock); |
87e8827b | 2088 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2089 | mm->locked_vm += grow; |
84638335 | 2090 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2091 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 2092 | vma->vm_end = address; |
bf181b9f | 2093 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 ML |
2094 | if (vma->vm_next) |
2095 | vma_gap_update(vma->vm_next); | |
2096 | else | |
09357814 ON |
2097 | mm->highest_vm_end = address; |
2098 | spin_unlock(&mm->page_table_lock); | |
4128997b | 2099 | |
42c36f63 HD |
2100 | perf_event_mmap(vma); |
2101 | } | |
3af9e859 | 2102 | } |
1da177e4 | 2103 | } |
12352d3c | 2104 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2105 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2106 | validate_mm(mm); |
1da177e4 LT |
2107 | return error; |
2108 | } | |
46dea3d0 HD |
2109 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2110 | ||
1da177e4 LT |
2111 | /* |
2112 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
2113 | */ | |
d05f3169 | 2114 | int expand_downwards(struct vm_area_struct *vma, |
b6a2fea3 | 2115 | unsigned long address) |
1da177e4 | 2116 | { |
09357814 | 2117 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 LT |
2118 | int error; |
2119 | ||
8869477a | 2120 | address &= PAGE_MASK; |
e5467859 | 2121 | error = security_mmap_addr(address); |
8869477a EP |
2122 | if (error) |
2123 | return error; | |
2124 | ||
12352d3c KK |
2125 | /* We must make sure the anon_vma is allocated. */ |
2126 | if (unlikely(anon_vma_prepare(vma))) | |
2127 | return -ENOMEM; | |
1da177e4 LT |
2128 | |
2129 | /* | |
2130 | * vma->vm_start/vm_end cannot change under us because the caller | |
2131 | * is required to hold the mmap_sem in read mode. We need the | |
2132 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2133 | */ | |
12352d3c | 2134 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2135 | |
2136 | /* Somebody else might have raced and expanded it already */ | |
2137 | if (address < vma->vm_start) { | |
2138 | unsigned long size, grow; | |
2139 | ||
2140 | size = vma->vm_end - address; | |
2141 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2142 | ||
a626ca6a LT |
2143 | error = -ENOMEM; |
2144 | if (grow <= vma->vm_pgoff) { | |
2145 | error = acct_stack_growth(vma, size, grow); | |
2146 | if (!error) { | |
4128997b ML |
2147 | /* |
2148 | * vma_gap_update() doesn't support concurrent | |
2149 | * updates, but we only hold a shared mmap_sem | |
2150 | * lock here, so we need to protect against | |
2151 | * concurrent vma expansions. | |
12352d3c | 2152 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2153 | * we don't guarantee that all growable vmas |
2154 | * in a mm share the same root anon vma. | |
2155 | * So, we reuse mm->page_table_lock to guard | |
2156 | * against concurrent vma expansions. | |
2157 | */ | |
09357814 | 2158 | spin_lock(&mm->page_table_lock); |
87e8827b | 2159 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2160 | mm->locked_vm += grow; |
84638335 | 2161 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2162 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2163 | vma->vm_start = address; |
2164 | vma->vm_pgoff -= grow; | |
bf181b9f | 2165 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 | 2166 | vma_gap_update(vma); |
09357814 | 2167 | spin_unlock(&mm->page_table_lock); |
4128997b | 2168 | |
a626ca6a LT |
2169 | perf_event_mmap(vma); |
2170 | } | |
1da177e4 LT |
2171 | } |
2172 | } | |
12352d3c | 2173 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2174 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2175 | validate_mm(mm); |
1da177e4 LT |
2176 | return error; |
2177 | } | |
2178 | ||
09884964 LT |
2179 | /* |
2180 | * Note how expand_stack() refuses to expand the stack all the way to | |
2181 | * abut the next virtual mapping, *unless* that mapping itself is also | |
2182 | * a stack mapping. We want to leave room for a guard page, after all | |
2183 | * (the guard page itself is not added here, that is done by the | |
2184 | * actual page faulting logic) | |
2185 | * | |
2186 | * This matches the behavior of the guard page logic (see mm/memory.c: | |
2187 | * check_stack_guard_page()), which only allows the guard page to be | |
2188 | * removed under these circumstances. | |
2189 | */ | |
b6a2fea3 OW |
2190 | #ifdef CONFIG_STACK_GROWSUP |
2191 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2192 | { | |
09884964 LT |
2193 | struct vm_area_struct *next; |
2194 | ||
2195 | address &= PAGE_MASK; | |
2196 | next = vma->vm_next; | |
2197 | if (next && next->vm_start == address + PAGE_SIZE) { | |
2198 | if (!(next->vm_flags & VM_GROWSUP)) | |
2199 | return -ENOMEM; | |
2200 | } | |
b6a2fea3 OW |
2201 | return expand_upwards(vma, address); |
2202 | } | |
2203 | ||
2204 | struct vm_area_struct * | |
2205 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2206 | { | |
2207 | struct vm_area_struct *vma, *prev; | |
2208 | ||
2209 | addr &= PAGE_MASK; | |
2210 | vma = find_vma_prev(mm, addr, &prev); | |
2211 | if (vma && (vma->vm_start <= addr)) | |
2212 | return vma; | |
1c127185 | 2213 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2214 | return NULL; |
cea10a19 | 2215 | if (prev->vm_flags & VM_LOCKED) |
fc05f566 | 2216 | populate_vma_page_range(prev, addr, prev->vm_end, NULL); |
b6a2fea3 OW |
2217 | return prev; |
2218 | } | |
2219 | #else | |
2220 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2221 | { | |
09884964 LT |
2222 | struct vm_area_struct *prev; |
2223 | ||
2224 | address &= PAGE_MASK; | |
2225 | prev = vma->vm_prev; | |
2226 | if (prev && prev->vm_end == address) { | |
2227 | if (!(prev->vm_flags & VM_GROWSDOWN)) | |
2228 | return -ENOMEM; | |
2229 | } | |
b6a2fea3 OW |
2230 | return expand_downwards(vma, address); |
2231 | } | |
2232 | ||
1da177e4 | 2233 | struct vm_area_struct * |
cc71aba3 | 2234 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2235 | { |
cc71aba3 | 2236 | struct vm_area_struct *vma; |
1da177e4 LT |
2237 | unsigned long start; |
2238 | ||
2239 | addr &= PAGE_MASK; | |
cc71aba3 | 2240 | vma = find_vma(mm, addr); |
1da177e4 LT |
2241 | if (!vma) |
2242 | return NULL; | |
2243 | if (vma->vm_start <= addr) | |
2244 | return vma; | |
2245 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2246 | return NULL; | |
2247 | start = vma->vm_start; | |
2248 | if (expand_stack(vma, addr)) | |
2249 | return NULL; | |
cea10a19 | 2250 | if (vma->vm_flags & VM_LOCKED) |
fc05f566 | 2251 | populate_vma_page_range(vma, addr, start, NULL); |
1da177e4 LT |
2252 | return vma; |
2253 | } | |
2254 | #endif | |
2255 | ||
e1d6d01a JB |
2256 | EXPORT_SYMBOL_GPL(find_extend_vma); |
2257 | ||
1da177e4 | 2258 | /* |
2c0b3814 | 2259 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 2260 | * so release them, and do the vma updates. |
2c0b3814 HD |
2261 | * |
2262 | * Called with the mm semaphore held. | |
1da177e4 | 2263 | */ |
2c0b3814 | 2264 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2265 | { |
4f74d2c8 LT |
2266 | unsigned long nr_accounted = 0; |
2267 | ||
365e9c87 HD |
2268 | /* Update high watermark before we lower total_vm */ |
2269 | update_hiwater_vm(mm); | |
1da177e4 | 2270 | do { |
2c0b3814 HD |
2271 | long nrpages = vma_pages(vma); |
2272 | ||
4f74d2c8 LT |
2273 | if (vma->vm_flags & VM_ACCOUNT) |
2274 | nr_accounted += nrpages; | |
84638335 | 2275 | vm_stat_account(mm, vma->vm_flags, -nrpages); |
a8fb5618 | 2276 | vma = remove_vma(vma); |
146425a3 | 2277 | } while (vma); |
4f74d2c8 | 2278 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2279 | validate_mm(mm); |
2280 | } | |
2281 | ||
2282 | /* | |
2283 | * Get rid of page table information in the indicated region. | |
2284 | * | |
f10df686 | 2285 | * Called with the mm semaphore held. |
1da177e4 LT |
2286 | */ |
2287 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
2288 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
2289 | unsigned long start, unsigned long end) | |
1da177e4 | 2290 | { |
cc71aba3 | 2291 | struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; |
d16dfc55 | 2292 | struct mmu_gather tlb; |
1da177e4 LT |
2293 | |
2294 | lru_add_drain(); | |
2b047252 | 2295 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 2296 | update_hiwater_rss(mm); |
4f74d2c8 | 2297 | unmap_vmas(&tlb, vma, start, end); |
d16dfc55 | 2298 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
6ee8630e | 2299 | next ? next->vm_start : USER_PGTABLES_CEILING); |
d16dfc55 | 2300 | tlb_finish_mmu(&tlb, start, end); |
1da177e4 LT |
2301 | } |
2302 | ||
2303 | /* | |
2304 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
2305 | * vma list as we go.. | |
2306 | */ | |
2307 | static void | |
2308 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
2309 | struct vm_area_struct *prev, unsigned long end) | |
2310 | { | |
2311 | struct vm_area_struct **insertion_point; | |
2312 | struct vm_area_struct *tail_vma = NULL; | |
2313 | ||
2314 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
297c5eee | 2315 | vma->vm_prev = NULL; |
1da177e4 | 2316 | do { |
d3737187 | 2317 | vma_rb_erase(vma, &mm->mm_rb); |
1da177e4 LT |
2318 | mm->map_count--; |
2319 | tail_vma = vma; | |
2320 | vma = vma->vm_next; | |
2321 | } while (vma && vma->vm_start < end); | |
2322 | *insertion_point = vma; | |
d3737187 | 2323 | if (vma) { |
297c5eee | 2324 | vma->vm_prev = prev; |
d3737187 ML |
2325 | vma_gap_update(vma); |
2326 | } else | |
2327 | mm->highest_vm_end = prev ? prev->vm_end : 0; | |
1da177e4 | 2328 | tail_vma->vm_next = NULL; |
615d6e87 DB |
2329 | |
2330 | /* Kill the cache */ | |
2331 | vmacache_invalidate(mm); | |
1da177e4 LT |
2332 | } |
2333 | ||
2334 | /* | |
659ace58 KM |
2335 | * __split_vma() bypasses sysctl_max_map_count checking. We use this on the |
2336 | * munmap path where it doesn't make sense to fail. | |
1da177e4 | 2337 | */ |
cc71aba3 | 2338 | static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
2339 | unsigned long addr, int new_below) |
2340 | { | |
1da177e4 | 2341 | struct vm_area_struct *new; |
e3975891 | 2342 | int err; |
1da177e4 | 2343 | |
a5516438 AK |
2344 | if (is_vm_hugetlb_page(vma) && (addr & |
2345 | ~(huge_page_mask(hstate_vma(vma))))) | |
1da177e4 LT |
2346 | return -EINVAL; |
2347 | ||
e94b1766 | 2348 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 | 2349 | if (!new) |
e3975891 | 2350 | return -ENOMEM; |
1da177e4 LT |
2351 | |
2352 | /* most fields are the same, copy all, and then fixup */ | |
2353 | *new = *vma; | |
2354 | ||
5beb4930 RR |
2355 | INIT_LIST_HEAD(&new->anon_vma_chain); |
2356 | ||
1da177e4 LT |
2357 | if (new_below) |
2358 | new->vm_end = addr; | |
2359 | else { | |
2360 | new->vm_start = addr; | |
2361 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2362 | } | |
2363 | ||
ef0855d3 ON |
2364 | err = vma_dup_policy(vma, new); |
2365 | if (err) | |
5beb4930 | 2366 | goto out_free_vma; |
1da177e4 | 2367 | |
c4ea95d7 DF |
2368 | err = anon_vma_clone(new, vma); |
2369 | if (err) | |
5beb4930 RR |
2370 | goto out_free_mpol; |
2371 | ||
e9714acf | 2372 | if (new->vm_file) |
1da177e4 LT |
2373 | get_file(new->vm_file); |
2374 | ||
2375 | if (new->vm_ops && new->vm_ops->open) | |
2376 | new->vm_ops->open(new); | |
2377 | ||
2378 | if (new_below) | |
5beb4930 | 2379 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
1da177e4 LT |
2380 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
2381 | else | |
5beb4930 | 2382 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
1da177e4 | 2383 | |
5beb4930 RR |
2384 | /* Success. */ |
2385 | if (!err) | |
2386 | return 0; | |
2387 | ||
2388 | /* Clean everything up if vma_adjust failed. */ | |
58927533 RR |
2389 | if (new->vm_ops && new->vm_ops->close) |
2390 | new->vm_ops->close(new); | |
e9714acf | 2391 | if (new->vm_file) |
5beb4930 | 2392 | fput(new->vm_file); |
2aeadc30 | 2393 | unlink_anon_vmas(new); |
5beb4930 | 2394 | out_free_mpol: |
ef0855d3 | 2395 | mpol_put(vma_policy(new)); |
5beb4930 RR |
2396 | out_free_vma: |
2397 | kmem_cache_free(vm_area_cachep, new); | |
5beb4930 | 2398 | return err; |
1da177e4 LT |
2399 | } |
2400 | ||
659ace58 KM |
2401 | /* |
2402 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2403 | * either for the first part or the tail. | |
2404 | */ | |
2405 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, | |
2406 | unsigned long addr, int new_below) | |
2407 | { | |
2408 | if (mm->map_count >= sysctl_max_map_count) | |
2409 | return -ENOMEM; | |
2410 | ||
2411 | return __split_vma(mm, vma, addr, new_below); | |
2412 | } | |
2413 | ||
1da177e4 LT |
2414 | /* Munmap is split into 2 main parts -- this part which finds |
2415 | * what needs doing, and the areas themselves, which do the | |
2416 | * work. This now handles partial unmappings. | |
2417 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
2418 | */ | |
2419 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
2420 | { | |
2421 | unsigned long end; | |
146425a3 | 2422 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 | 2423 | |
de1741a1 | 2424 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) |
1da177e4 LT |
2425 | return -EINVAL; |
2426 | ||
cc71aba3 | 2427 | len = PAGE_ALIGN(len); |
2428 | if (len == 0) | |
1da177e4 LT |
2429 | return -EINVAL; |
2430 | ||
2431 | /* Find the first overlapping VMA */ | |
9be34c9d | 2432 | vma = find_vma(mm, start); |
146425a3 | 2433 | if (!vma) |
1da177e4 | 2434 | return 0; |
9be34c9d | 2435 | prev = vma->vm_prev; |
146425a3 | 2436 | /* we have start < vma->vm_end */ |
1da177e4 LT |
2437 | |
2438 | /* if it doesn't overlap, we have nothing.. */ | |
2439 | end = start + len; | |
146425a3 | 2440 | if (vma->vm_start >= end) |
1da177e4 LT |
2441 | return 0; |
2442 | ||
2443 | /* | |
2444 | * If we need to split any vma, do it now to save pain later. | |
2445 | * | |
2446 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2447 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2448 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2449 | */ | |
146425a3 | 2450 | if (start > vma->vm_start) { |
659ace58 KM |
2451 | int error; |
2452 | ||
2453 | /* | |
2454 | * Make sure that map_count on return from munmap() will | |
2455 | * not exceed its limit; but let map_count go just above | |
2456 | * its limit temporarily, to help free resources as expected. | |
2457 | */ | |
2458 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
2459 | return -ENOMEM; | |
2460 | ||
2461 | error = __split_vma(mm, vma, start, 0); | |
1da177e4 LT |
2462 | if (error) |
2463 | return error; | |
146425a3 | 2464 | prev = vma; |
1da177e4 LT |
2465 | } |
2466 | ||
2467 | /* Does it split the last one? */ | |
2468 | last = find_vma(mm, end); | |
2469 | if (last && end > last->vm_start) { | |
659ace58 | 2470 | int error = __split_vma(mm, last, end, 1); |
1da177e4 LT |
2471 | if (error) |
2472 | return error; | |
2473 | } | |
cc71aba3 | 2474 | vma = prev ? prev->vm_next : mm->mmap; |
1da177e4 | 2475 | |
ba470de4 RR |
2476 | /* |
2477 | * unlock any mlock()ed ranges before detaching vmas | |
2478 | */ | |
2479 | if (mm->locked_vm) { | |
2480 | struct vm_area_struct *tmp = vma; | |
2481 | while (tmp && tmp->vm_start < end) { | |
2482 | if (tmp->vm_flags & VM_LOCKED) { | |
2483 | mm->locked_vm -= vma_pages(tmp); | |
2484 | munlock_vma_pages_all(tmp); | |
2485 | } | |
2486 | tmp = tmp->vm_next; | |
2487 | } | |
2488 | } | |
2489 | ||
1da177e4 LT |
2490 | /* |
2491 | * Remove the vma's, and unmap the actual pages | |
2492 | */ | |
146425a3 HD |
2493 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
2494 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 | 2495 | |
1de4fa14 DH |
2496 | arch_unmap(mm, vma, start, end); |
2497 | ||
1da177e4 | 2498 | /* Fix up all other VM information */ |
2c0b3814 | 2499 | remove_vma_list(mm, vma); |
1da177e4 LT |
2500 | |
2501 | return 0; | |
2502 | } | |
1da177e4 | 2503 | |
bfce281c | 2504 | int vm_munmap(unsigned long start, size_t len) |
1da177e4 LT |
2505 | { |
2506 | int ret; | |
bfce281c | 2507 | struct mm_struct *mm = current->mm; |
1da177e4 | 2508 | |
ae798783 MH |
2509 | if (down_write_killable(&mm->mmap_sem)) |
2510 | return -EINTR; | |
2511 | ||
a46ef99d | 2512 | ret = do_munmap(mm, start, len); |
1da177e4 LT |
2513 | up_write(&mm->mmap_sem); |
2514 | return ret; | |
2515 | } | |
a46ef99d LT |
2516 | EXPORT_SYMBOL(vm_munmap); |
2517 | ||
2518 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2519 | { | |
dc0ef0df MH |
2520 | int ret; |
2521 | struct mm_struct *mm = current->mm; | |
2522 | ||
a46ef99d | 2523 | profile_munmap(addr); |
dc0ef0df MH |
2524 | if (down_write_killable(&mm->mmap_sem)) |
2525 | return -EINTR; | |
2526 | ret = do_munmap(mm, addr, len); | |
2527 | up_write(&mm->mmap_sem); | |
2528 | return ret; | |
a46ef99d | 2529 | } |
1da177e4 | 2530 | |
c8d78c18 KS |
2531 | |
2532 | /* | |
2533 | * Emulation of deprecated remap_file_pages() syscall. | |
2534 | */ | |
2535 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, | |
2536 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) | |
2537 | { | |
2538 | ||
2539 | struct mm_struct *mm = current->mm; | |
2540 | struct vm_area_struct *vma; | |
2541 | unsigned long populate = 0; | |
2542 | unsigned long ret = -EINVAL; | |
2543 | struct file *file; | |
2544 | ||
756a025f JP |
2545 | pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n", |
2546 | current->comm, current->pid); | |
c8d78c18 KS |
2547 | |
2548 | if (prot) | |
2549 | return ret; | |
2550 | start = start & PAGE_MASK; | |
2551 | size = size & PAGE_MASK; | |
2552 | ||
2553 | if (start + size <= start) | |
2554 | return ret; | |
2555 | ||
2556 | /* Does pgoff wrap? */ | |
2557 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) | |
2558 | return ret; | |
2559 | ||
dc0ef0df MH |
2560 | if (down_write_killable(&mm->mmap_sem)) |
2561 | return -EINTR; | |
2562 | ||
c8d78c18 KS |
2563 | vma = find_vma(mm, start); |
2564 | ||
2565 | if (!vma || !(vma->vm_flags & VM_SHARED)) | |
2566 | goto out; | |
2567 | ||
48f7df32 | 2568 | if (start < vma->vm_start) |
c8d78c18 KS |
2569 | goto out; |
2570 | ||
48f7df32 KS |
2571 | if (start + size > vma->vm_end) { |
2572 | struct vm_area_struct *next; | |
2573 | ||
2574 | for (next = vma->vm_next; next; next = next->vm_next) { | |
2575 | /* hole between vmas ? */ | |
2576 | if (next->vm_start != next->vm_prev->vm_end) | |
2577 | goto out; | |
2578 | ||
2579 | if (next->vm_file != vma->vm_file) | |
2580 | goto out; | |
2581 | ||
2582 | if (next->vm_flags != vma->vm_flags) | |
2583 | goto out; | |
2584 | ||
2585 | if (start + size <= next->vm_end) | |
2586 | break; | |
2587 | } | |
2588 | ||
2589 | if (!next) | |
2590 | goto out; | |
c8d78c18 KS |
2591 | } |
2592 | ||
2593 | prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; | |
2594 | prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; | |
2595 | prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; | |
2596 | ||
2597 | flags &= MAP_NONBLOCK; | |
2598 | flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; | |
2599 | if (vma->vm_flags & VM_LOCKED) { | |
48f7df32 | 2600 | struct vm_area_struct *tmp; |
c8d78c18 | 2601 | flags |= MAP_LOCKED; |
48f7df32 | 2602 | |
c8d78c18 | 2603 | /* drop PG_Mlocked flag for over-mapped range */ |
48f7df32 KS |
2604 | for (tmp = vma; tmp->vm_start >= start + size; |
2605 | tmp = tmp->vm_next) { | |
9a73f61b KS |
2606 | /* |
2607 | * Split pmd and munlock page on the border | |
2608 | * of the range. | |
2609 | */ | |
2610 | vma_adjust_trans_huge(tmp, start, start + size, 0); | |
2611 | ||
48f7df32 KS |
2612 | munlock_vma_pages_range(tmp, |
2613 | max(tmp->vm_start, start), | |
2614 | min(tmp->vm_end, start + size)); | |
2615 | } | |
c8d78c18 KS |
2616 | } |
2617 | ||
2618 | file = get_file(vma->vm_file); | |
2619 | ret = do_mmap_pgoff(vma->vm_file, start, size, | |
2620 | prot, flags, pgoff, &populate); | |
2621 | fput(file); | |
2622 | out: | |
2623 | up_write(&mm->mmap_sem); | |
2624 | if (populate) | |
2625 | mm_populate(ret, populate); | |
2626 | if (!IS_ERR_VALUE(ret)) | |
2627 | ret = 0; | |
2628 | return ret; | |
2629 | } | |
2630 | ||
1da177e4 LT |
2631 | static inline void verify_mm_writelocked(struct mm_struct *mm) |
2632 | { | |
a241ec65 | 2633 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
2634 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
2635 | WARN_ON(1); | |
2636 | up_read(&mm->mmap_sem); | |
2637 | } | |
2638 | #endif | |
2639 | } | |
2640 | ||
2641 | /* | |
2642 | * this is really a simplified "do_mmap". it only handles | |
2643 | * anonymous maps. eventually we may be able to do some | |
2644 | * brk-specific accounting here. | |
2645 | */ | |
5d22fc25 | 2646 | static int do_brk(unsigned long addr, unsigned long len) |
1da177e4 | 2647 | { |
cc71aba3 | 2648 | struct mm_struct *mm = current->mm; |
2649 | struct vm_area_struct *vma, *prev; | |
1da177e4 | 2650 | unsigned long flags; |
cc71aba3 | 2651 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 2652 | pgoff_t pgoff = addr >> PAGE_SHIFT; |
3a459756 | 2653 | int error; |
1da177e4 LT |
2654 | |
2655 | len = PAGE_ALIGN(len); | |
2656 | if (!len) | |
5d22fc25 | 2657 | return 0; |
1da177e4 | 2658 | |
3a459756 KK |
2659 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; |
2660 | ||
2c6a1016 | 2661 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
de1741a1 | 2662 | if (offset_in_page(error)) |
3a459756 KK |
2663 | return error; |
2664 | ||
363ee17f DB |
2665 | error = mlock_future_check(mm, mm->def_flags, len); |
2666 | if (error) | |
2667 | return error; | |
1da177e4 LT |
2668 | |
2669 | /* | |
2670 | * mm->mmap_sem is required to protect against another thread | |
2671 | * changing the mappings in case we sleep. | |
2672 | */ | |
2673 | verify_mm_writelocked(mm); | |
2674 | ||
2675 | /* | |
2676 | * Clear old maps. this also does some error checking for us | |
2677 | */ | |
9fcd1457 RV |
2678 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
2679 | &rb_parent)) { | |
1da177e4 LT |
2680 | if (do_munmap(mm, addr, len)) |
2681 | return -ENOMEM; | |
1da177e4 LT |
2682 | } |
2683 | ||
2684 | /* Check against address space limits *after* clearing old maps... */ | |
84638335 | 2685 | if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) |
1da177e4 LT |
2686 | return -ENOMEM; |
2687 | ||
2688 | if (mm->map_count > sysctl_max_map_count) | |
2689 | return -ENOMEM; | |
2690 | ||
191c5424 | 2691 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2692 | return -ENOMEM; |
2693 | ||
1da177e4 | 2694 | /* Can we just expand an old private anonymous mapping? */ |
ba470de4 | 2695 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
19a809af | 2696 | NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); |
ba470de4 | 2697 | if (vma) |
1da177e4 LT |
2698 | goto out; |
2699 | ||
2700 | /* | |
2701 | * create a vma struct for an anonymous mapping | |
2702 | */ | |
c5e3b83e | 2703 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2704 | if (!vma) { |
2705 | vm_unacct_memory(len >> PAGE_SHIFT); | |
2706 | return -ENOMEM; | |
2707 | } | |
1da177e4 | 2708 | |
5beb4930 | 2709 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
2710 | vma->vm_mm = mm; |
2711 | vma->vm_start = addr; | |
2712 | vma->vm_end = addr + len; | |
2713 | vma->vm_pgoff = pgoff; | |
2714 | vma->vm_flags = flags; | |
3ed75eb8 | 2715 | vma->vm_page_prot = vm_get_page_prot(flags); |
1da177e4 LT |
2716 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2717 | out: | |
3af9e859 | 2718 | perf_event_mmap(vma); |
1da177e4 | 2719 | mm->total_vm += len >> PAGE_SHIFT; |
84638335 | 2720 | mm->data_vm += len >> PAGE_SHIFT; |
128557ff ML |
2721 | if (flags & VM_LOCKED) |
2722 | mm->locked_vm += (len >> PAGE_SHIFT); | |
d9104d1c | 2723 | vma->vm_flags |= VM_SOFTDIRTY; |
5d22fc25 | 2724 | return 0; |
1da177e4 LT |
2725 | } |
2726 | ||
5d22fc25 | 2727 | int vm_brk(unsigned long addr, unsigned long len) |
e4eb1ff6 LT |
2728 | { |
2729 | struct mm_struct *mm = current->mm; | |
5d22fc25 | 2730 | int ret; |
128557ff | 2731 | bool populate; |
e4eb1ff6 | 2732 | |
2d6c9282 MH |
2733 | if (down_write_killable(&mm->mmap_sem)) |
2734 | return -EINTR; | |
2735 | ||
e4eb1ff6 | 2736 | ret = do_brk(addr, len); |
128557ff | 2737 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
e4eb1ff6 | 2738 | up_write(&mm->mmap_sem); |
5d22fc25 | 2739 | if (populate && !ret) |
128557ff | 2740 | mm_populate(addr, len); |
e4eb1ff6 LT |
2741 | return ret; |
2742 | } | |
2743 | EXPORT_SYMBOL(vm_brk); | |
1da177e4 LT |
2744 | |
2745 | /* Release all mmaps. */ | |
2746 | void exit_mmap(struct mm_struct *mm) | |
2747 | { | |
d16dfc55 | 2748 | struct mmu_gather tlb; |
ba470de4 | 2749 | struct vm_area_struct *vma; |
1da177e4 LT |
2750 | unsigned long nr_accounted = 0; |
2751 | ||
d6dd61c8 | 2752 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 2753 | mmu_notifier_release(mm); |
d6dd61c8 | 2754 | |
ba470de4 RR |
2755 | if (mm->locked_vm) { |
2756 | vma = mm->mmap; | |
2757 | while (vma) { | |
2758 | if (vma->vm_flags & VM_LOCKED) | |
2759 | munlock_vma_pages_all(vma); | |
2760 | vma = vma->vm_next; | |
2761 | } | |
2762 | } | |
9480c53e JF |
2763 | |
2764 | arch_exit_mmap(mm); | |
2765 | ||
ba470de4 | 2766 | vma = mm->mmap; |
9480c53e JF |
2767 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
2768 | return; | |
2769 | ||
1da177e4 | 2770 | lru_add_drain(); |
1da177e4 | 2771 | flush_cache_mm(mm); |
2b047252 | 2772 | tlb_gather_mmu(&tlb, mm, 0, -1); |
901608d9 | 2773 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
e0da382c | 2774 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
4f74d2c8 | 2775 | unmap_vmas(&tlb, vma, 0, -1); |
9ba69294 | 2776 | |
6ee8630e | 2777 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); |
853f5e26 | 2778 | tlb_finish_mmu(&tlb, 0, -1); |
1da177e4 | 2779 | |
1da177e4 | 2780 | /* |
8f4f8c16 HD |
2781 | * Walk the list again, actually closing and freeing it, |
2782 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 2783 | */ |
4f74d2c8 LT |
2784 | while (vma) { |
2785 | if (vma->vm_flags & VM_ACCOUNT) | |
2786 | nr_accounted += vma_pages(vma); | |
a8fb5618 | 2787 | vma = remove_vma(vma); |
4f74d2c8 LT |
2788 | } |
2789 | vm_unacct_memory(nr_accounted); | |
1da177e4 LT |
2790 | } |
2791 | ||
2792 | /* Insert vm structure into process list sorted by address | |
2793 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
c8c06efa | 2794 | * then i_mmap_rwsem is taken here. |
1da177e4 | 2795 | */ |
6597d783 | 2796 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2797 | { |
6597d783 HD |
2798 | struct vm_area_struct *prev; |
2799 | struct rb_node **rb_link, *rb_parent; | |
1da177e4 | 2800 | |
c9d13f5f CG |
2801 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
2802 | &prev, &rb_link, &rb_parent)) | |
2803 | return -ENOMEM; | |
2804 | if ((vma->vm_flags & VM_ACCOUNT) && | |
2805 | security_vm_enough_memory_mm(mm, vma_pages(vma))) | |
2806 | return -ENOMEM; | |
2807 | ||
1da177e4 LT |
2808 | /* |
2809 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
2810 | * until its first write fault, when page's anon_vma and index | |
2811 | * are set. But now set the vm_pgoff it will almost certainly | |
2812 | * end up with (unless mremap moves it elsewhere before that | |
2813 | * first wfault), so /proc/pid/maps tells a consistent story. | |
2814 | * | |
2815 | * By setting it to reflect the virtual start address of the | |
2816 | * vma, merges and splits can happen in a seamless way, just | |
2817 | * using the existing file pgoff checks and manipulations. | |
2818 | * Similarly in do_mmap_pgoff and in do_brk. | |
2819 | */ | |
8a9cc3b5 | 2820 | if (vma_is_anonymous(vma)) { |
1da177e4 LT |
2821 | BUG_ON(vma->anon_vma); |
2822 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
2823 | } | |
2b144498 | 2824 | |
1da177e4 LT |
2825 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2826 | return 0; | |
2827 | } | |
2828 | ||
2829 | /* | |
2830 | * Copy the vma structure to a new location in the same mm, | |
2831 | * prior to moving page table entries, to effect an mremap move. | |
2832 | */ | |
2833 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
2834 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
2835 | bool *need_rmap_locks) | |
1da177e4 LT |
2836 | { |
2837 | struct vm_area_struct *vma = *vmap; | |
2838 | unsigned long vma_start = vma->vm_start; | |
2839 | struct mm_struct *mm = vma->vm_mm; | |
2840 | struct vm_area_struct *new_vma, *prev; | |
2841 | struct rb_node **rb_link, *rb_parent; | |
948f017b | 2842 | bool faulted_in_anon_vma = true; |
1da177e4 LT |
2843 | |
2844 | /* | |
2845 | * If anonymous vma has not yet been faulted, update new pgoff | |
2846 | * to match new location, to increase its chance of merging. | |
2847 | */ | |
ce75799b | 2848 | if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { |
1da177e4 | 2849 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
2850 | faulted_in_anon_vma = false; |
2851 | } | |
1da177e4 | 2852 | |
6597d783 HD |
2853 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
2854 | return NULL; /* should never get here */ | |
1da177e4 | 2855 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
19a809af AA |
2856 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
2857 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
2858 | if (new_vma) { |
2859 | /* | |
2860 | * Source vma may have been merged into new_vma | |
2861 | */ | |
948f017b AA |
2862 | if (unlikely(vma_start >= new_vma->vm_start && |
2863 | vma_start < new_vma->vm_end)) { | |
2864 | /* | |
2865 | * The only way we can get a vma_merge with | |
2866 | * self during an mremap is if the vma hasn't | |
2867 | * been faulted in yet and we were allowed to | |
2868 | * reset the dst vma->vm_pgoff to the | |
2869 | * destination address of the mremap to allow | |
2870 | * the merge to happen. mremap must change the | |
2871 | * vm_pgoff linearity between src and dst vmas | |
2872 | * (in turn preventing a vma_merge) to be | |
2873 | * safe. It is only safe to keep the vm_pgoff | |
2874 | * linear if there are no pages mapped yet. | |
2875 | */ | |
81d1b09c | 2876 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
38a76013 | 2877 | *vmap = vma = new_vma; |
108d6642 | 2878 | } |
38a76013 | 2879 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 2880 | } else { |
e94b1766 | 2881 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
e3975891 CG |
2882 | if (!new_vma) |
2883 | goto out; | |
2884 | *new_vma = *vma; | |
2885 | new_vma->vm_start = addr; | |
2886 | new_vma->vm_end = addr + len; | |
2887 | new_vma->vm_pgoff = pgoff; | |
2888 | if (vma_dup_policy(vma, new_vma)) | |
2889 | goto out_free_vma; | |
2890 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); | |
2891 | if (anon_vma_clone(new_vma, vma)) | |
2892 | goto out_free_mempol; | |
2893 | if (new_vma->vm_file) | |
2894 | get_file(new_vma->vm_file); | |
2895 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
2896 | new_vma->vm_ops->open(new_vma); | |
2897 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
2898 | *need_rmap_locks = false; | |
1da177e4 LT |
2899 | } |
2900 | return new_vma; | |
5beb4930 | 2901 | |
e3975891 | 2902 | out_free_mempol: |
ef0855d3 | 2903 | mpol_put(vma_policy(new_vma)); |
e3975891 | 2904 | out_free_vma: |
5beb4930 | 2905 | kmem_cache_free(vm_area_cachep, new_vma); |
e3975891 | 2906 | out: |
5beb4930 | 2907 | return NULL; |
1da177e4 | 2908 | } |
119f657c | 2909 | |
2910 | /* | |
2911 | * Return true if the calling process may expand its vm space by the passed | |
2912 | * number of pages | |
2913 | */ | |
84638335 | 2914 | bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) |
119f657c | 2915 | { |
84638335 KK |
2916 | if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) |
2917 | return false; | |
119f657c | 2918 | |
d977d56c KK |
2919 | if (is_data_mapping(flags) && |
2920 | mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { | |
f4fcd558 KK |
2921 | /* Workaround for Valgrind */ |
2922 | if (rlimit(RLIMIT_DATA) == 0 && | |
2923 | mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) | |
2924 | return true; | |
2925 | if (!ignore_rlimit_data) { | |
2926 | pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n", | |
d977d56c KK |
2927 | current->comm, current->pid, |
2928 | (mm->data_vm + npages) << PAGE_SHIFT, | |
2929 | rlimit(RLIMIT_DATA)); | |
d977d56c | 2930 | return false; |
f4fcd558 | 2931 | } |
d977d56c | 2932 | } |
119f657c | 2933 | |
84638335 KK |
2934 | return true; |
2935 | } | |
2936 | ||
2937 | void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) | |
2938 | { | |
2939 | mm->total_vm += npages; | |
2940 | ||
d977d56c | 2941 | if (is_exec_mapping(flags)) |
84638335 | 2942 | mm->exec_vm += npages; |
d977d56c | 2943 | else if (is_stack_mapping(flags)) |
84638335 | 2944 | mm->stack_vm += npages; |
d977d56c | 2945 | else if (is_data_mapping(flags)) |
84638335 | 2946 | mm->data_vm += npages; |
119f657c | 2947 | } |
fa5dc22f | 2948 | |
a62c34bd AL |
2949 | static int special_mapping_fault(struct vm_area_struct *vma, |
2950 | struct vm_fault *vmf); | |
2951 | ||
2952 | /* | |
2953 | * Having a close hook prevents vma merging regardless of flags. | |
2954 | */ | |
2955 | static void special_mapping_close(struct vm_area_struct *vma) | |
2956 | { | |
2957 | } | |
2958 | ||
2959 | static const char *special_mapping_name(struct vm_area_struct *vma) | |
2960 | { | |
2961 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; | |
2962 | } | |
2963 | ||
b059a453 DS |
2964 | static int special_mapping_mremap(struct vm_area_struct *new_vma) |
2965 | { | |
2966 | struct vm_special_mapping *sm = new_vma->vm_private_data; | |
2967 | ||
2968 | if (sm->mremap) | |
2969 | return sm->mremap(sm, new_vma); | |
2970 | return 0; | |
2971 | } | |
2972 | ||
a62c34bd AL |
2973 | static const struct vm_operations_struct special_mapping_vmops = { |
2974 | .close = special_mapping_close, | |
2975 | .fault = special_mapping_fault, | |
b059a453 | 2976 | .mremap = special_mapping_mremap, |
a62c34bd AL |
2977 | .name = special_mapping_name, |
2978 | }; | |
2979 | ||
2980 | static const struct vm_operations_struct legacy_special_mapping_vmops = { | |
2981 | .close = special_mapping_close, | |
2982 | .fault = special_mapping_fault, | |
2983 | }; | |
fa5dc22f | 2984 | |
b1d0e4f5 NP |
2985 | static int special_mapping_fault(struct vm_area_struct *vma, |
2986 | struct vm_fault *vmf) | |
fa5dc22f | 2987 | { |
b1d0e4f5 | 2988 | pgoff_t pgoff; |
fa5dc22f RM |
2989 | struct page **pages; |
2990 | ||
f872f540 | 2991 | if (vma->vm_ops == &legacy_special_mapping_vmops) { |
a62c34bd | 2992 | pages = vma->vm_private_data; |
f872f540 AL |
2993 | } else { |
2994 | struct vm_special_mapping *sm = vma->vm_private_data; | |
2995 | ||
2996 | if (sm->fault) | |
2997 | return sm->fault(sm, vma, vmf); | |
2998 | ||
2999 | pages = sm->pages; | |
3000 | } | |
a62c34bd | 3001 | |
8a9cc3b5 | 3002 | for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) |
b1d0e4f5 | 3003 | pgoff--; |
fa5dc22f RM |
3004 | |
3005 | if (*pages) { | |
3006 | struct page *page = *pages; | |
3007 | get_page(page); | |
b1d0e4f5 NP |
3008 | vmf->page = page; |
3009 | return 0; | |
fa5dc22f RM |
3010 | } |
3011 | ||
b1d0e4f5 | 3012 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
3013 | } |
3014 | ||
a62c34bd AL |
3015 | static struct vm_area_struct *__install_special_mapping( |
3016 | struct mm_struct *mm, | |
3017 | unsigned long addr, unsigned long len, | |
27f28b97 CG |
3018 | unsigned long vm_flags, void *priv, |
3019 | const struct vm_operations_struct *ops) | |
fa5dc22f | 3020 | { |
462e635e | 3021 | int ret; |
fa5dc22f RM |
3022 | struct vm_area_struct *vma; |
3023 | ||
3024 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
3025 | if (unlikely(vma == NULL)) | |
3935ed6a | 3026 | return ERR_PTR(-ENOMEM); |
fa5dc22f | 3027 | |
5beb4930 | 3028 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
fa5dc22f RM |
3029 | vma->vm_mm = mm; |
3030 | vma->vm_start = addr; | |
3031 | vma->vm_end = addr + len; | |
3032 | ||
d9104d1c | 3033 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; |
3ed75eb8 | 3034 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f | 3035 | |
a62c34bd AL |
3036 | vma->vm_ops = ops; |
3037 | vma->vm_private_data = priv; | |
fa5dc22f | 3038 | |
462e635e TO |
3039 | ret = insert_vm_struct(mm, vma); |
3040 | if (ret) | |
3041 | goto out; | |
fa5dc22f | 3042 | |
84638335 | 3043 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); |
fa5dc22f | 3044 | |
cdd6c482 | 3045 | perf_event_mmap(vma); |
089dd79d | 3046 | |
3935ed6a | 3047 | return vma; |
462e635e TO |
3048 | |
3049 | out: | |
3050 | kmem_cache_free(vm_area_cachep, vma); | |
3935ed6a SS |
3051 | return ERR_PTR(ret); |
3052 | } | |
3053 | ||
a62c34bd AL |
3054 | /* |
3055 | * Called with mm->mmap_sem held for writing. | |
3056 | * Insert a new vma covering the given region, with the given flags. | |
3057 | * Its pages are supplied by the given array of struct page *. | |
3058 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
3059 | * The region past the last page supplied will always produce SIGBUS. | |
3060 | * The array pointer and the pages it points to are assumed to stay alive | |
3061 | * for as long as this mapping might exist. | |
3062 | */ | |
3063 | struct vm_area_struct *_install_special_mapping( | |
3064 | struct mm_struct *mm, | |
3065 | unsigned long addr, unsigned long len, | |
3066 | unsigned long vm_flags, const struct vm_special_mapping *spec) | |
3067 | { | |
27f28b97 CG |
3068 | return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, |
3069 | &special_mapping_vmops); | |
a62c34bd AL |
3070 | } |
3071 | ||
3935ed6a SS |
3072 | int install_special_mapping(struct mm_struct *mm, |
3073 | unsigned long addr, unsigned long len, | |
3074 | unsigned long vm_flags, struct page **pages) | |
3075 | { | |
a62c34bd | 3076 | struct vm_area_struct *vma = __install_special_mapping( |
27f28b97 CG |
3077 | mm, addr, len, vm_flags, (void *)pages, |
3078 | &legacy_special_mapping_vmops); | |
3935ed6a | 3079 | |
14bd5b45 | 3080 | return PTR_ERR_OR_ZERO(vma); |
fa5dc22f | 3081 | } |
7906d00c AA |
3082 | |
3083 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3084 | ||
454ed842 | 3085 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3086 | { |
bf181b9f | 3087 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3088 | /* |
3089 | * The LSB of head.next can't change from under us | |
3090 | * because we hold the mm_all_locks_mutex. | |
3091 | */ | |
572043c9 | 3092 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); |
7906d00c AA |
3093 | /* |
3094 | * We can safely modify head.next after taking the | |
5a505085 | 3095 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3096 | * the same anon_vma we won't take it again. |
3097 | * | |
3098 | * No need of atomic instructions here, head.next | |
3099 | * can't change from under us thanks to the | |
5a505085 | 3100 | * anon_vma->root->rwsem. |
7906d00c AA |
3101 | */ |
3102 | if (__test_and_set_bit(0, (unsigned long *) | |
bf181b9f | 3103 | &anon_vma->root->rb_root.rb_node)) |
7906d00c AA |
3104 | BUG(); |
3105 | } | |
3106 | } | |
3107 | ||
454ed842 | 3108 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3109 | { |
3110 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3111 | /* | |
3112 | * AS_MM_ALL_LOCKS can't change from under us because | |
3113 | * we hold the mm_all_locks_mutex. | |
3114 | * | |
3115 | * Operations on ->flags have to be atomic because | |
3116 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3117 | * mm_all_locks_mutex, there may be other cpus | |
3118 | * changing other bitflags in parallel to us. | |
3119 | */ | |
3120 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3121 | BUG(); | |
c8c06efa | 3122 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); |
7906d00c AA |
3123 | } |
3124 | } | |
3125 | ||
3126 | /* | |
3127 | * This operation locks against the VM for all pte/vma/mm related | |
3128 | * operations that could ever happen on a certain mm. This includes | |
3129 | * vmtruncate, try_to_unmap, and all page faults. | |
3130 | * | |
3131 | * The caller must take the mmap_sem in write mode before calling | |
3132 | * mm_take_all_locks(). The caller isn't allowed to release the | |
3133 | * mmap_sem until mm_drop_all_locks() returns. | |
3134 | * | |
3135 | * mmap_sem in write mode is required in order to block all operations | |
3136 | * that could modify pagetables and free pages without need of | |
27ba0644 | 3137 | * altering the vma layout. It's also needed in write mode to avoid new |
7906d00c AA |
3138 | * anon_vmas to be associated with existing vmas. |
3139 | * | |
3140 | * A single task can't take more than one mm_take_all_locks() in a row | |
3141 | * or it would deadlock. | |
3142 | * | |
bf181b9f | 3143 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3144 | * mapping->flags avoid to take the same lock twice, if more than one |
3145 | * vma in this mm is backed by the same anon_vma or address_space. | |
3146 | * | |
88f306b6 KS |
3147 | * We take locks in following order, accordingly to comment at beginning |
3148 | * of mm/rmap.c: | |
3149 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for | |
3150 | * hugetlb mapping); | |
3151 | * - all i_mmap_rwsem locks; | |
3152 | * - all anon_vma->rwseml | |
3153 | * | |
3154 | * We can take all locks within these types randomly because the VM code | |
3155 | * doesn't nest them and we protected from parallel mm_take_all_locks() by | |
3156 | * mm_all_locks_mutex. | |
7906d00c AA |
3157 | * |
3158 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3159 | * that may have to take thousand of locks. | |
3160 | * | |
3161 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3162 | */ | |
3163 | int mm_take_all_locks(struct mm_struct *mm) | |
3164 | { | |
3165 | struct vm_area_struct *vma; | |
5beb4930 | 3166 | struct anon_vma_chain *avc; |
7906d00c AA |
3167 | |
3168 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3169 | ||
3170 | mutex_lock(&mm_all_locks_mutex); | |
3171 | ||
3172 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3173 | if (signal_pending(current)) | |
3174 | goto out_unlock; | |
88f306b6 KS |
3175 | if (vma->vm_file && vma->vm_file->f_mapping && |
3176 | is_vm_hugetlb_page(vma)) | |
3177 | vm_lock_mapping(mm, vma->vm_file->f_mapping); | |
3178 | } | |
3179 | ||
3180 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3181 | if (signal_pending(current)) | |
3182 | goto out_unlock; | |
3183 | if (vma->vm_file && vma->vm_file->f_mapping && | |
3184 | !is_vm_hugetlb_page(vma)) | |
454ed842 | 3185 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3186 | } |
7cd5a02f PZ |
3187 | |
3188 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3189 | if (signal_pending(current)) | |
3190 | goto out_unlock; | |
3191 | if (vma->anon_vma) | |
5beb4930 RR |
3192 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3193 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3194 | } |
7cd5a02f | 3195 | |
584cff54 | 3196 | return 0; |
7906d00c AA |
3197 | |
3198 | out_unlock: | |
584cff54 KC |
3199 | mm_drop_all_locks(mm); |
3200 | return -EINTR; | |
7906d00c AA |
3201 | } |
3202 | ||
3203 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3204 | { | |
bf181b9f | 3205 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3206 | /* |
3207 | * The LSB of head.next can't change to 0 from under | |
3208 | * us because we hold the mm_all_locks_mutex. | |
3209 | * | |
3210 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3211 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3212 | * never see our bitflag. |
3213 | * | |
3214 | * No need of atomic instructions here, head.next | |
3215 | * can't change from under us until we release the | |
5a505085 | 3216 | * anon_vma->root->rwsem. |
7906d00c AA |
3217 | */ |
3218 | if (!__test_and_clear_bit(0, (unsigned long *) | |
bf181b9f | 3219 | &anon_vma->root->rb_root.rb_node)) |
7906d00c | 3220 | BUG(); |
08b52706 | 3221 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3222 | } |
3223 | } | |
3224 | ||
3225 | static void vm_unlock_mapping(struct address_space *mapping) | |
3226 | { | |
3227 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3228 | /* | |
3229 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3230 | * because we hold the mm_all_locks_mutex. | |
3231 | */ | |
83cde9e8 | 3232 | i_mmap_unlock_write(mapping); |
7906d00c AA |
3233 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3234 | &mapping->flags)) | |
3235 | BUG(); | |
3236 | } | |
3237 | } | |
3238 | ||
3239 | /* | |
3240 | * The mmap_sem cannot be released by the caller until | |
3241 | * mm_drop_all_locks() returns. | |
3242 | */ | |
3243 | void mm_drop_all_locks(struct mm_struct *mm) | |
3244 | { | |
3245 | struct vm_area_struct *vma; | |
5beb4930 | 3246 | struct anon_vma_chain *avc; |
7906d00c AA |
3247 | |
3248 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3249 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); | |
3250 | ||
3251 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3252 | if (vma->anon_vma) | |
5beb4930 RR |
3253 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3254 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3255 | if (vma->vm_file && vma->vm_file->f_mapping) |
3256 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3257 | } | |
3258 | ||
3259 | mutex_unlock(&mm_all_locks_mutex); | |
3260 | } | |
8feae131 DH |
3261 | |
3262 | /* | |
3263 | * initialise the VMA slab | |
3264 | */ | |
3265 | void __init mmap_init(void) | |
3266 | { | |
00a62ce9 KM |
3267 | int ret; |
3268 | ||
908c7f19 | 3269 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 3270 | VM_BUG_ON(ret); |
8feae131 | 3271 | } |
c9b1d098 AS |
3272 | |
3273 | /* | |
3274 | * Initialise sysctl_user_reserve_kbytes. | |
3275 | * | |
3276 | * This is intended to prevent a user from starting a single memory hogging | |
3277 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3278 | * mode. | |
3279 | * | |
3280 | * The default value is min(3% of free memory, 128MB) | |
3281 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3282 | */ | |
1640879a | 3283 | static int init_user_reserve(void) |
c9b1d098 AS |
3284 | { |
3285 | unsigned long free_kbytes; | |
3286 | ||
3287 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3288 | ||
3289 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3290 | return 0; | |
3291 | } | |
a64fb3cd | 3292 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
3293 | |
3294 | /* | |
3295 | * Initialise sysctl_admin_reserve_kbytes. | |
3296 | * | |
3297 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3298 | * to log in and kill a memory hogging process. | |
3299 | * | |
3300 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3301 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3302 | * only reserve 3% of free pages by default. | |
3303 | */ | |
1640879a | 3304 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3305 | { |
3306 | unsigned long free_kbytes; | |
3307 | ||
3308 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3309 | ||
3310 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3311 | return 0; | |
3312 | } | |
a64fb3cd | 3313 | subsys_initcall(init_admin_reserve); |
1640879a AS |
3314 | |
3315 | /* | |
3316 | * Reinititalise user and admin reserves if memory is added or removed. | |
3317 | * | |
3318 | * The default user reserve max is 128MB, and the default max for the | |
3319 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3320 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3321 | * and tools like top. It may make sense to increase or even disable the | |
3322 | * reserve depending on the existence of swap or variations in the recovery | |
3323 | * tools. So, the admin may have changed them. | |
3324 | * | |
3325 | * If memory is added and the reserves have been eliminated or increased above | |
3326 | * the default max, then we'll trust the admin. | |
3327 | * | |
3328 | * If memory is removed and there isn't enough free memory, then we | |
3329 | * need to reset the reserves. | |
3330 | * | |
3331 | * Otherwise keep the reserve set by the admin. | |
3332 | */ | |
3333 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3334 | unsigned long action, void *data) | |
3335 | { | |
3336 | unsigned long tmp, free_kbytes; | |
3337 | ||
3338 | switch (action) { | |
3339 | case MEM_ONLINE: | |
3340 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3341 | tmp = sysctl_user_reserve_kbytes; | |
3342 | if (0 < tmp && tmp < (1UL << 17)) | |
3343 | init_user_reserve(); | |
3344 | ||
3345 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3346 | tmp = sysctl_admin_reserve_kbytes; | |
3347 | if (0 < tmp && tmp < (1UL << 13)) | |
3348 | init_admin_reserve(); | |
3349 | ||
3350 | break; | |
3351 | case MEM_OFFLINE: | |
3352 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3353 | ||
3354 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3355 | init_user_reserve(); | |
3356 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3357 | sysctl_user_reserve_kbytes); | |
3358 | } | |
3359 | ||
3360 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3361 | init_admin_reserve(); | |
3362 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3363 | sysctl_admin_reserve_kbytes); | |
3364 | } | |
3365 | break; | |
3366 | default: | |
3367 | break; | |
3368 | } | |
3369 | return NOTIFY_OK; | |
3370 | } | |
3371 | ||
3372 | static struct notifier_block reserve_mem_nb = { | |
3373 | .notifier_call = reserve_mem_notifier, | |
3374 | }; | |
3375 | ||
3376 | static int __meminit init_reserve_notifier(void) | |
3377 | { | |
3378 | if (register_hotmemory_notifier(&reserve_mem_nb)) | |
b1de0d13 | 3379 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
1640879a AS |
3380 | |
3381 | return 0; | |
3382 | } | |
a64fb3cd | 3383 | subsys_initcall(init_reserve_notifier); |