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 | * | |
cab15ce6 CM |
91 | * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and |
92 | * MAP_PRIVATE: | |
93 | * r: (no) no | |
94 | * w: (no) no | |
95 | * x: (yes) yes | |
1da177e4 LT |
96 | */ |
97 | pgprot_t protection_map[16] = { | |
98 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, | |
99 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
100 | }; | |
101 | ||
804af2cf HD |
102 | pgprot_t vm_get_page_prot(unsigned long vm_flags) |
103 | { | |
b845f313 DK |
104 | return __pgprot(pgprot_val(protection_map[vm_flags & |
105 | (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | | |
106 | pgprot_val(arch_vm_get_page_prot(vm_flags))); | |
804af2cf HD |
107 | } |
108 | EXPORT_SYMBOL(vm_get_page_prot); | |
109 | ||
64e45507 PF |
110 | static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) |
111 | { | |
112 | return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); | |
113 | } | |
114 | ||
115 | /* Update vma->vm_page_prot to reflect vma->vm_flags. */ | |
116 | void vma_set_page_prot(struct vm_area_struct *vma) | |
117 | { | |
118 | unsigned long vm_flags = vma->vm_flags; | |
119 | ||
120 | vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); | |
121 | if (vma_wants_writenotify(vma)) { | |
122 | vm_flags &= ~VM_SHARED; | |
123 | vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, | |
124 | vm_flags); | |
125 | } | |
126 | } | |
127 | ||
1da177e4 | 128 | /* |
c8c06efa | 129 | * Requires inode->i_mapping->i_mmap_rwsem |
1da177e4 LT |
130 | */ |
131 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
132 | struct file *file, struct address_space *mapping) | |
133 | { | |
134 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 135 | atomic_inc(&file_inode(file)->i_writecount); |
1da177e4 | 136 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 137 | mapping_unmap_writable(mapping); |
1da177e4 LT |
138 | |
139 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 140 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
141 | flush_dcache_mmap_unlock(mapping); |
142 | } | |
143 | ||
144 | /* | |
6b2dbba8 | 145 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 146 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 147 | */ |
a8fb5618 | 148 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
149 | { |
150 | struct file *file = vma->vm_file; | |
151 | ||
1da177e4 LT |
152 | if (file) { |
153 | struct address_space *mapping = file->f_mapping; | |
83cde9e8 | 154 | i_mmap_lock_write(mapping); |
1da177e4 | 155 | __remove_shared_vm_struct(vma, file, mapping); |
83cde9e8 | 156 | i_mmap_unlock_write(mapping); |
1da177e4 | 157 | } |
a8fb5618 HD |
158 | } |
159 | ||
160 | /* | |
161 | * Close a vm structure and free it, returning the next. | |
162 | */ | |
163 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
164 | { | |
165 | struct vm_area_struct *next = vma->vm_next; | |
166 | ||
a8fb5618 | 167 | might_sleep(); |
1da177e4 LT |
168 | if (vma->vm_ops && vma->vm_ops->close) |
169 | vma->vm_ops->close(vma); | |
e9714acf | 170 | if (vma->vm_file) |
a8fb5618 | 171 | fput(vma->vm_file); |
f0be3d32 | 172 | mpol_put(vma_policy(vma)); |
1da177e4 | 173 | kmem_cache_free(vm_area_cachep, vma); |
a8fb5618 | 174 | return next; |
1da177e4 LT |
175 | } |
176 | ||
5d22fc25 | 177 | static int do_brk(unsigned long addr, unsigned long len); |
e4eb1ff6 | 178 | |
6a6160a7 | 179 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 | 180 | { |
8764b338 | 181 | unsigned long retval; |
1da177e4 LT |
182 | unsigned long newbrk, oldbrk; |
183 | struct mm_struct *mm = current->mm; | |
a5b4592c | 184 | unsigned long min_brk; |
128557ff | 185 | bool populate; |
1da177e4 | 186 | |
dc0ef0df MH |
187 | if (down_write_killable(&mm->mmap_sem)) |
188 | return -EINTR; | |
1da177e4 | 189 | |
a5b4592c | 190 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
191 | /* |
192 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
193 | * randomize_va_space to 2, which will still cause mm->start_brk | |
194 | * to be arbitrarily shifted | |
195 | */ | |
4471a675 | 196 | if (current->brk_randomized) |
5520e894 JK |
197 | min_brk = mm->start_brk; |
198 | else | |
199 | min_brk = mm->end_data; | |
a5b4592c JK |
200 | #else |
201 | min_brk = mm->start_brk; | |
202 | #endif | |
203 | if (brk < min_brk) | |
1da177e4 | 204 | goto out; |
1e624196 RG |
205 | |
206 | /* | |
207 | * Check against rlimit here. If this check is done later after the test | |
208 | * of oldbrk with newbrk then it can escape the test and let the data | |
209 | * segment grow beyond its set limit the in case where the limit is | |
210 | * not page aligned -Ram Gupta | |
211 | */ | |
8764b338 CG |
212 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
213 | mm->end_data, mm->start_data)) | |
1e624196 RG |
214 | goto out; |
215 | ||
1da177e4 LT |
216 | newbrk = PAGE_ALIGN(brk); |
217 | oldbrk = PAGE_ALIGN(mm->brk); | |
218 | if (oldbrk == newbrk) | |
219 | goto set_brk; | |
220 | ||
221 | /* Always allow shrinking brk. */ | |
222 | if (brk <= mm->brk) { | |
223 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
224 | goto set_brk; | |
225 | goto out; | |
226 | } | |
227 | ||
1da177e4 LT |
228 | /* Check against existing mmap mappings. */ |
229 | if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) | |
230 | goto out; | |
231 | ||
232 | /* Ok, looks good - let it rip. */ | |
5d22fc25 | 233 | if (do_brk(oldbrk, newbrk-oldbrk) < 0) |
1da177e4 | 234 | goto out; |
128557ff | 235 | |
1da177e4 LT |
236 | set_brk: |
237 | mm->brk = brk; | |
128557ff ML |
238 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
239 | up_write(&mm->mmap_sem); | |
240 | if (populate) | |
241 | mm_populate(oldbrk, newbrk - oldbrk); | |
242 | return brk; | |
243 | ||
1da177e4 LT |
244 | out: |
245 | retval = mm->brk; | |
246 | up_write(&mm->mmap_sem); | |
247 | return retval; | |
248 | } | |
249 | ||
d3737187 ML |
250 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
251 | { | |
252 | unsigned long max, subtree_gap; | |
253 | max = vma->vm_start; | |
254 | if (vma->vm_prev) | |
255 | max -= vma->vm_prev->vm_end; | |
256 | if (vma->vm_rb.rb_left) { | |
257 | subtree_gap = rb_entry(vma->vm_rb.rb_left, | |
258 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
259 | if (subtree_gap > max) | |
260 | max = subtree_gap; | |
261 | } | |
262 | if (vma->vm_rb.rb_right) { | |
263 | subtree_gap = rb_entry(vma->vm_rb.rb_right, | |
264 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
265 | if (subtree_gap > max) | |
266 | max = subtree_gap; | |
267 | } | |
268 | return max; | |
269 | } | |
270 | ||
ed8ea815 | 271 | #ifdef CONFIG_DEBUG_VM_RB |
acf128d0 | 272 | static int browse_rb(struct mm_struct *mm) |
1da177e4 | 273 | { |
acf128d0 | 274 | struct rb_root *root = &mm->mm_rb; |
5a0768f6 | 275 | int i = 0, j, bug = 0; |
1da177e4 LT |
276 | struct rb_node *nd, *pn = NULL; |
277 | unsigned long prev = 0, pend = 0; | |
278 | ||
279 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
280 | struct vm_area_struct *vma; | |
281 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
5a0768f6 | 282 | if (vma->vm_start < prev) { |
ff26f70f AM |
283 | pr_emerg("vm_start %lx < prev %lx\n", |
284 | vma->vm_start, prev); | |
5a0768f6 ML |
285 | bug = 1; |
286 | } | |
287 | if (vma->vm_start < pend) { | |
ff26f70f AM |
288 | pr_emerg("vm_start %lx < pend %lx\n", |
289 | vma->vm_start, pend); | |
5a0768f6 ML |
290 | bug = 1; |
291 | } | |
292 | if (vma->vm_start > vma->vm_end) { | |
ff26f70f AM |
293 | pr_emerg("vm_start %lx > vm_end %lx\n", |
294 | vma->vm_start, vma->vm_end); | |
5a0768f6 ML |
295 | bug = 1; |
296 | } | |
acf128d0 | 297 | spin_lock(&mm->page_table_lock); |
5a0768f6 | 298 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { |
8542bdfc | 299 | pr_emerg("free gap %lx, correct %lx\n", |
5a0768f6 ML |
300 | vma->rb_subtree_gap, |
301 | vma_compute_subtree_gap(vma)); | |
302 | bug = 1; | |
303 | } | |
acf128d0 | 304 | spin_unlock(&mm->page_table_lock); |
1da177e4 LT |
305 | i++; |
306 | pn = nd; | |
d1af65d1 DM |
307 | prev = vma->vm_start; |
308 | pend = vma->vm_end; | |
1da177e4 LT |
309 | } |
310 | j = 0; | |
5a0768f6 | 311 | for (nd = pn; nd; nd = rb_prev(nd)) |
1da177e4 | 312 | j++; |
5a0768f6 | 313 | if (i != j) { |
8542bdfc | 314 | pr_emerg("backwards %d, forwards %d\n", j, i); |
5a0768f6 | 315 | bug = 1; |
1da177e4 | 316 | } |
5a0768f6 | 317 | return bug ? -1 : i; |
1da177e4 LT |
318 | } |
319 | ||
d3737187 ML |
320 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
321 | { | |
322 | struct rb_node *nd; | |
323 | ||
324 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
325 | struct vm_area_struct *vma; | |
326 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
96dad67f SL |
327 | VM_BUG_ON_VMA(vma != ignore && |
328 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma), | |
329 | vma); | |
1da177e4 | 330 | } |
1da177e4 LT |
331 | } |
332 | ||
eafd4dc4 | 333 | static void validate_mm(struct mm_struct *mm) |
1da177e4 LT |
334 | { |
335 | int bug = 0; | |
336 | int i = 0; | |
5a0768f6 | 337 | unsigned long highest_address = 0; |
ed8ea815 | 338 | struct vm_area_struct *vma = mm->mmap; |
ff26f70f | 339 | |
ed8ea815 | 340 | while (vma) { |
12352d3c | 341 | struct anon_vma *anon_vma = vma->anon_vma; |
ed8ea815 | 342 | struct anon_vma_chain *avc; |
ff26f70f | 343 | |
12352d3c KK |
344 | if (anon_vma) { |
345 | anon_vma_lock_read(anon_vma); | |
346 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
347 | anon_vma_interval_tree_verify(avc); | |
348 | anon_vma_unlock_read(anon_vma); | |
349 | } | |
350 | ||
5a0768f6 | 351 | highest_address = vma->vm_end; |
ed8ea815 | 352 | vma = vma->vm_next; |
1da177e4 LT |
353 | i++; |
354 | } | |
5a0768f6 | 355 | if (i != mm->map_count) { |
8542bdfc | 356 | pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); |
5a0768f6 ML |
357 | bug = 1; |
358 | } | |
359 | if (highest_address != mm->highest_vm_end) { | |
8542bdfc | 360 | pr_emerg("mm->highest_vm_end %lx, found %lx\n", |
ff26f70f | 361 | mm->highest_vm_end, highest_address); |
5a0768f6 ML |
362 | bug = 1; |
363 | } | |
acf128d0 | 364 | i = browse_rb(mm); |
5a0768f6 | 365 | if (i != mm->map_count) { |
ff26f70f AM |
366 | if (i != -1) |
367 | pr_emerg("map_count %d rb %d\n", mm->map_count, i); | |
5a0768f6 ML |
368 | bug = 1; |
369 | } | |
96dad67f | 370 | VM_BUG_ON_MM(bug, mm); |
1da177e4 LT |
371 | } |
372 | #else | |
d3737187 | 373 | #define validate_mm_rb(root, ignore) do { } while (0) |
1da177e4 LT |
374 | #define validate_mm(mm) do { } while (0) |
375 | #endif | |
376 | ||
d3737187 ML |
377 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
378 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) | |
379 | ||
380 | /* | |
381 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or | |
382 | * vma->vm_prev->vm_end values changed, without modifying the vma's position | |
383 | * in the rbtree. | |
384 | */ | |
385 | static void vma_gap_update(struct vm_area_struct *vma) | |
386 | { | |
387 | /* | |
388 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback | |
389 | * function that does exacltly what we want. | |
390 | */ | |
391 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); | |
392 | } | |
393 | ||
394 | static inline void vma_rb_insert(struct vm_area_struct *vma, | |
395 | struct rb_root *root) | |
396 | { | |
397 | /* All rb_subtree_gap values must be consistent prior to insertion */ | |
398 | validate_mm_rb(root, NULL); | |
399 | ||
400 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
401 | } | |
402 | ||
403 | static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) | |
404 | { | |
405 | /* | |
406 | * All rb_subtree_gap values must be consistent prior to erase, | |
407 | * with the possible exception of the vma being erased. | |
408 | */ | |
409 | validate_mm_rb(root, vma); | |
410 | ||
411 | /* | |
412 | * Note rb_erase_augmented is a fairly large inline function, | |
413 | * so make sure we instantiate it only once with our desired | |
414 | * augmented rbtree callbacks. | |
415 | */ | |
416 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
417 | } | |
418 | ||
bf181b9f ML |
419 | /* |
420 | * vma has some anon_vma assigned, and is already inserted on that | |
421 | * anon_vma's interval trees. | |
422 | * | |
423 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
424 | * vma must be removed from the anon_vma's interval trees using | |
425 | * anon_vma_interval_tree_pre_update_vma(). | |
426 | * | |
427 | * After the update, the vma will be reinserted using | |
428 | * anon_vma_interval_tree_post_update_vma(). | |
429 | * | |
430 | * The entire update must be protected by exclusive mmap_sem and by | |
431 | * the root anon_vma's mutex. | |
432 | */ | |
433 | static inline void | |
434 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
435 | { | |
436 | struct anon_vma_chain *avc; | |
437 | ||
438 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
439 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
440 | } | |
441 | ||
442 | static inline void | |
443 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
444 | { | |
445 | struct anon_vma_chain *avc; | |
446 | ||
447 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
448 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
449 | } | |
450 | ||
6597d783 HD |
451 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
452 | unsigned long end, struct vm_area_struct **pprev, | |
453 | struct rb_node ***rb_link, struct rb_node **rb_parent) | |
1da177e4 | 454 | { |
6597d783 | 455 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
1da177e4 LT |
456 | |
457 | __rb_link = &mm->mm_rb.rb_node; | |
458 | rb_prev = __rb_parent = NULL; | |
1da177e4 LT |
459 | |
460 | while (*__rb_link) { | |
461 | struct vm_area_struct *vma_tmp; | |
462 | ||
463 | __rb_parent = *__rb_link; | |
464 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
465 | ||
466 | if (vma_tmp->vm_end > addr) { | |
6597d783 HD |
467 | /* Fail if an existing vma overlaps the area */ |
468 | if (vma_tmp->vm_start < end) | |
469 | return -ENOMEM; | |
1da177e4 LT |
470 | __rb_link = &__rb_parent->rb_left; |
471 | } else { | |
472 | rb_prev = __rb_parent; | |
473 | __rb_link = &__rb_parent->rb_right; | |
474 | } | |
475 | } | |
476 | ||
477 | *pprev = NULL; | |
478 | if (rb_prev) | |
479 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
480 | *rb_link = __rb_link; | |
481 | *rb_parent = __rb_parent; | |
6597d783 | 482 | return 0; |
1da177e4 LT |
483 | } |
484 | ||
e8420a8e CH |
485 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
486 | unsigned long addr, unsigned long end) | |
487 | { | |
488 | unsigned long nr_pages = 0; | |
489 | struct vm_area_struct *vma; | |
490 | ||
491 | /* Find first overlaping mapping */ | |
492 | vma = find_vma_intersection(mm, addr, end); | |
493 | if (!vma) | |
494 | return 0; | |
495 | ||
496 | nr_pages = (min(end, vma->vm_end) - | |
497 | max(addr, vma->vm_start)) >> PAGE_SHIFT; | |
498 | ||
499 | /* Iterate over the rest of the overlaps */ | |
500 | for (vma = vma->vm_next; vma; vma = vma->vm_next) { | |
501 | unsigned long overlap_len; | |
502 | ||
503 | if (vma->vm_start > end) | |
504 | break; | |
505 | ||
506 | overlap_len = min(end, vma->vm_end) - vma->vm_start; | |
507 | nr_pages += overlap_len >> PAGE_SHIFT; | |
508 | } | |
509 | ||
510 | return nr_pages; | |
511 | } | |
512 | ||
1da177e4 LT |
513 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
514 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
515 | { | |
d3737187 ML |
516 | /* Update tracking information for the gap following the new vma. */ |
517 | if (vma->vm_next) | |
518 | vma_gap_update(vma->vm_next); | |
519 | else | |
520 | mm->highest_vm_end = vma->vm_end; | |
521 | ||
522 | /* | |
523 | * vma->vm_prev wasn't known when we followed the rbtree to find the | |
524 | * correct insertion point for that vma. As a result, we could not | |
525 | * update the vma vm_rb parents rb_subtree_gap values on the way down. | |
526 | * So, we first insert the vma with a zero rb_subtree_gap value | |
527 | * (to be consistent with what we did on the way down), and then | |
528 | * immediately update the gap to the correct value. Finally we | |
529 | * rebalance the rbtree after all augmented values have been set. | |
530 | */ | |
1da177e4 | 531 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
d3737187 ML |
532 | vma->rb_subtree_gap = 0; |
533 | vma_gap_update(vma); | |
534 | vma_rb_insert(vma, &mm->mm_rb); | |
1da177e4 LT |
535 | } |
536 | ||
cb8f488c | 537 | static void __vma_link_file(struct vm_area_struct *vma) |
1da177e4 | 538 | { |
48aae425 | 539 | struct file *file; |
1da177e4 LT |
540 | |
541 | file = vma->vm_file; | |
542 | if (file) { | |
543 | struct address_space *mapping = file->f_mapping; | |
544 | ||
545 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 546 | atomic_dec(&file_inode(file)->i_writecount); |
1da177e4 | 547 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 548 | atomic_inc(&mapping->i_mmap_writable); |
1da177e4 LT |
549 | |
550 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 551 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 LT |
552 | flush_dcache_mmap_unlock(mapping); |
553 | } | |
554 | } | |
555 | ||
556 | static void | |
557 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
558 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
559 | struct rb_node *rb_parent) | |
560 | { | |
561 | __vma_link_list(mm, vma, prev, rb_parent); | |
562 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
1da177e4 LT |
563 | } |
564 | ||
565 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
566 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
567 | struct rb_node *rb_parent) | |
568 | { | |
569 | struct address_space *mapping = NULL; | |
570 | ||
64ac4940 | 571 | if (vma->vm_file) { |
1da177e4 | 572 | mapping = vma->vm_file->f_mapping; |
83cde9e8 | 573 | i_mmap_lock_write(mapping); |
64ac4940 | 574 | } |
1da177e4 LT |
575 | |
576 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
577 | __vma_link_file(vma); | |
578 | ||
1da177e4 | 579 | if (mapping) |
83cde9e8 | 580 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
581 | |
582 | mm->map_count++; | |
583 | validate_mm(mm); | |
584 | } | |
585 | ||
586 | /* | |
88f6b4c3 | 587 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
6b2dbba8 | 588 | * mm's list and rbtree. It has already been inserted into the interval tree. |
1da177e4 | 589 | */ |
48aae425 | 590 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 591 | { |
6597d783 | 592 | struct vm_area_struct *prev; |
48aae425 | 593 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 594 | |
6597d783 HD |
595 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
596 | &prev, &rb_link, &rb_parent)) | |
597 | BUG(); | |
1da177e4 LT |
598 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
599 | mm->map_count++; | |
600 | } | |
601 | ||
602 | static inline void | |
603 | __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, | |
604 | struct vm_area_struct *prev) | |
605 | { | |
d3737187 | 606 | struct vm_area_struct *next; |
297c5eee | 607 | |
d3737187 ML |
608 | vma_rb_erase(vma, &mm->mm_rb); |
609 | prev->vm_next = next = vma->vm_next; | |
297c5eee LT |
610 | if (next) |
611 | next->vm_prev = prev; | |
615d6e87 DB |
612 | |
613 | /* Kill the cache */ | |
614 | vmacache_invalidate(mm); | |
1da177e4 LT |
615 | } |
616 | ||
617 | /* | |
618 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
619 | * is already present in an i_mmap tree without adjusting the tree. | |
620 | * The following helper function should be used when such adjustments | |
621 | * are necessary. The "insert" vma (if any) is to be inserted | |
622 | * before we drop the necessary locks. | |
623 | */ | |
5beb4930 | 624 | int vma_adjust(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
625 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) |
626 | { | |
627 | struct mm_struct *mm = vma->vm_mm; | |
628 | struct vm_area_struct *next = vma->vm_next; | |
1da177e4 | 629 | struct address_space *mapping = NULL; |
6b2dbba8 | 630 | struct rb_root *root = NULL; |
012f1800 | 631 | struct anon_vma *anon_vma = NULL; |
1da177e4 | 632 | struct file *file = vma->vm_file; |
d3737187 | 633 | bool start_changed = false, end_changed = false; |
1da177e4 LT |
634 | long adjust_next = 0; |
635 | int remove_next = 0; | |
636 | ||
637 | if (next && !insert) { | |
734537c9 | 638 | struct vm_area_struct *exporter = NULL, *importer = NULL; |
287d97ac | 639 | |
1da177e4 LT |
640 | if (end >= next->vm_end) { |
641 | /* | |
642 | * vma expands, overlapping all the next, and | |
643 | * perhaps the one after too (mprotect case 6). | |
644 | */ | |
734537c9 | 645 | remove_next = 1 + (end > next->vm_end); |
1da177e4 | 646 | end = next->vm_end; |
287d97ac | 647 | exporter = next; |
1da177e4 | 648 | importer = vma; |
734537c9 KS |
649 | |
650 | /* | |
651 | * If next doesn't have anon_vma, import from vma after | |
652 | * next, if the vma overlaps with it. | |
653 | */ | |
654 | if (remove_next == 2 && next && !next->anon_vma) | |
655 | exporter = next->vm_next; | |
656 | ||
1da177e4 LT |
657 | } else if (end > next->vm_start) { |
658 | /* | |
659 | * vma expands, overlapping part of the next: | |
660 | * mprotect case 5 shifting the boundary up. | |
661 | */ | |
662 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
287d97ac | 663 | exporter = next; |
1da177e4 LT |
664 | importer = vma; |
665 | } else if (end < vma->vm_end) { | |
666 | /* | |
667 | * vma shrinks, and !insert tells it's not | |
668 | * split_vma inserting another: so it must be | |
669 | * mprotect case 4 shifting the boundary down. | |
670 | */ | |
cc71aba3 | 671 | adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); |
287d97ac | 672 | exporter = vma; |
1da177e4 LT |
673 | importer = next; |
674 | } | |
1da177e4 | 675 | |
5beb4930 RR |
676 | /* |
677 | * Easily overlooked: when mprotect shifts the boundary, | |
678 | * make sure the expanding vma has anon_vma set if the | |
679 | * shrinking vma had, to cover any anon pages imported. | |
680 | */ | |
287d97ac | 681 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
c4ea95d7 DF |
682 | int error; |
683 | ||
b800c91a | 684 | importer->anon_vma = exporter->anon_vma; |
c4ea95d7 | 685 | error = anon_vma_clone(importer, exporter); |
3fe89b3e | 686 | if (error) |
c4ea95d7 | 687 | return error; |
5beb4930 RR |
688 | } |
689 | } | |
734537c9 | 690 | again: |
37f9f559 KS |
691 | vma_adjust_trans_huge(vma, start, end, adjust_next); |
692 | ||
1da177e4 LT |
693 | if (file) { |
694 | mapping = file->f_mapping; | |
27ba0644 KS |
695 | root = &mapping->i_mmap; |
696 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); | |
682968e0 | 697 | |
27ba0644 KS |
698 | if (adjust_next) |
699 | uprobe_munmap(next, next->vm_start, next->vm_end); | |
682968e0 | 700 | |
83cde9e8 | 701 | i_mmap_lock_write(mapping); |
1da177e4 | 702 | if (insert) { |
1da177e4 | 703 | /* |
6b2dbba8 | 704 | * Put into interval tree now, so instantiated pages |
1da177e4 LT |
705 | * are visible to arm/parisc __flush_dcache_page |
706 | * throughout; but we cannot insert into address | |
707 | * space until vma start or end is updated. | |
708 | */ | |
709 | __vma_link_file(insert); | |
710 | } | |
711 | } | |
712 | ||
bf181b9f ML |
713 | anon_vma = vma->anon_vma; |
714 | if (!anon_vma && adjust_next) | |
715 | anon_vma = next->anon_vma; | |
716 | if (anon_vma) { | |
81d1b09c SL |
717 | VM_BUG_ON_VMA(adjust_next && next->anon_vma && |
718 | anon_vma != next->anon_vma, next); | |
4fc3f1d6 | 719 | anon_vma_lock_write(anon_vma); |
bf181b9f ML |
720 | anon_vma_interval_tree_pre_update_vma(vma); |
721 | if (adjust_next) | |
722 | anon_vma_interval_tree_pre_update_vma(next); | |
723 | } | |
012f1800 | 724 | |
1da177e4 LT |
725 | if (root) { |
726 | flush_dcache_mmap_lock(mapping); | |
6b2dbba8 | 727 | vma_interval_tree_remove(vma, root); |
1da177e4 | 728 | if (adjust_next) |
6b2dbba8 | 729 | vma_interval_tree_remove(next, root); |
1da177e4 LT |
730 | } |
731 | ||
d3737187 ML |
732 | if (start != vma->vm_start) { |
733 | vma->vm_start = start; | |
734 | start_changed = true; | |
735 | } | |
736 | if (end != vma->vm_end) { | |
737 | vma->vm_end = end; | |
738 | end_changed = true; | |
739 | } | |
1da177e4 LT |
740 | vma->vm_pgoff = pgoff; |
741 | if (adjust_next) { | |
742 | next->vm_start += adjust_next << PAGE_SHIFT; | |
743 | next->vm_pgoff += adjust_next; | |
744 | } | |
745 | ||
746 | if (root) { | |
747 | if (adjust_next) | |
6b2dbba8 ML |
748 | vma_interval_tree_insert(next, root); |
749 | vma_interval_tree_insert(vma, root); | |
1da177e4 LT |
750 | flush_dcache_mmap_unlock(mapping); |
751 | } | |
752 | ||
753 | if (remove_next) { | |
754 | /* | |
755 | * vma_merge has merged next into vma, and needs | |
756 | * us to remove next before dropping the locks. | |
757 | */ | |
758 | __vma_unlink(mm, next, vma); | |
759 | if (file) | |
760 | __remove_shared_vm_struct(next, file, mapping); | |
1da177e4 LT |
761 | } else if (insert) { |
762 | /* | |
763 | * split_vma has split insert from vma, and needs | |
764 | * us to insert it before dropping the locks | |
765 | * (it may either follow vma or precede it). | |
766 | */ | |
767 | __insert_vm_struct(mm, insert); | |
d3737187 ML |
768 | } else { |
769 | if (start_changed) | |
770 | vma_gap_update(vma); | |
771 | if (end_changed) { | |
772 | if (!next) | |
773 | mm->highest_vm_end = end; | |
774 | else if (!adjust_next) | |
775 | vma_gap_update(next); | |
776 | } | |
1da177e4 LT |
777 | } |
778 | ||
bf181b9f ML |
779 | if (anon_vma) { |
780 | anon_vma_interval_tree_post_update_vma(vma); | |
781 | if (adjust_next) | |
782 | anon_vma_interval_tree_post_update_vma(next); | |
08b52706 | 783 | anon_vma_unlock_write(anon_vma); |
bf181b9f | 784 | } |
1da177e4 | 785 | if (mapping) |
83cde9e8 | 786 | i_mmap_unlock_write(mapping); |
1da177e4 | 787 | |
2b144498 | 788 | if (root) { |
7b2d81d4 | 789 | uprobe_mmap(vma); |
2b144498 SD |
790 | |
791 | if (adjust_next) | |
7b2d81d4 | 792 | uprobe_mmap(next); |
2b144498 SD |
793 | } |
794 | ||
1da177e4 | 795 | if (remove_next) { |
925d1c40 | 796 | if (file) { |
cbc91f71 | 797 | uprobe_munmap(next, next->vm_start, next->vm_end); |
1da177e4 | 798 | fput(file); |
925d1c40 | 799 | } |
5beb4930 RR |
800 | if (next->anon_vma) |
801 | anon_vma_merge(vma, next); | |
1da177e4 | 802 | mm->map_count--; |
3964acd0 | 803 | mpol_put(vma_policy(next)); |
1da177e4 LT |
804 | kmem_cache_free(vm_area_cachep, next); |
805 | /* | |
806 | * In mprotect's case 6 (see comments on vma_merge), | |
807 | * we must remove another next too. It would clutter | |
808 | * up the code too much to do both in one go. | |
809 | */ | |
d3737187 | 810 | next = vma->vm_next; |
734537c9 KS |
811 | if (remove_next == 2) { |
812 | remove_next = 1; | |
813 | end = next->vm_end; | |
1da177e4 | 814 | goto again; |
734537c9 | 815 | } |
d3737187 ML |
816 | else if (next) |
817 | vma_gap_update(next); | |
818 | else | |
819 | mm->highest_vm_end = end; | |
1da177e4 | 820 | } |
2b144498 | 821 | if (insert && file) |
7b2d81d4 | 822 | uprobe_mmap(insert); |
1da177e4 LT |
823 | |
824 | validate_mm(mm); | |
5beb4930 RR |
825 | |
826 | return 0; | |
1da177e4 LT |
827 | } |
828 | ||
829 | /* | |
830 | * If the vma has a ->close operation then the driver probably needs to release | |
831 | * per-vma resources, so we don't attempt to merge those. | |
832 | */ | |
1da177e4 | 833 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
19a809af AA |
834 | struct file *file, unsigned long vm_flags, |
835 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 836 | { |
34228d47 CG |
837 | /* |
838 | * VM_SOFTDIRTY should not prevent from VMA merging, if we | |
839 | * match the flags but dirty bit -- the caller should mark | |
840 | * merged VMA as dirty. If dirty bit won't be excluded from | |
841 | * comparison, we increase pressue on the memory system forcing | |
842 | * the kernel to generate new VMAs when old one could be | |
843 | * extended instead. | |
844 | */ | |
845 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) | |
1da177e4 LT |
846 | return 0; |
847 | if (vma->vm_file != file) | |
848 | return 0; | |
849 | if (vma->vm_ops && vma->vm_ops->close) | |
850 | return 0; | |
19a809af AA |
851 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) |
852 | return 0; | |
1da177e4 LT |
853 | return 1; |
854 | } | |
855 | ||
856 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
965f55de SL |
857 | struct anon_vma *anon_vma2, |
858 | struct vm_area_struct *vma) | |
1da177e4 | 859 | { |
965f55de SL |
860 | /* |
861 | * The list_is_singular() test is to avoid merging VMA cloned from | |
862 | * parents. This can improve scalability caused by anon_vma lock. | |
863 | */ | |
864 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
865 | list_is_singular(&vma->anon_vma_chain))) | |
866 | return 1; | |
867 | return anon_vma1 == anon_vma2; | |
1da177e4 LT |
868 | } |
869 | ||
870 | /* | |
871 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
872 | * in front of (at a lower virtual address and file offset than) the vma. | |
873 | * | |
874 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
875 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
876 | * | |
877 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
878 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
879 | * wrap, nor mmaps which cover the final page at index -1UL. | |
880 | */ | |
881 | static int | |
882 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
883 | struct anon_vma *anon_vma, struct file *file, |
884 | pgoff_t vm_pgoff, | |
885 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 886 | { |
19a809af | 887 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 888 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
889 | if (vma->vm_pgoff == vm_pgoff) |
890 | return 1; | |
891 | } | |
892 | return 0; | |
893 | } | |
894 | ||
895 | /* | |
896 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
897 | * beyond (at a higher virtual address and file offset than) the vma. | |
898 | * | |
899 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
900 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
901 | */ | |
902 | static int | |
903 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
904 | struct anon_vma *anon_vma, struct file *file, |
905 | pgoff_t vm_pgoff, | |
906 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 907 | { |
19a809af | 908 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 909 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 910 | pgoff_t vm_pglen; |
d6e93217 | 911 | vm_pglen = vma_pages(vma); |
1da177e4 LT |
912 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
913 | return 1; | |
914 | } | |
915 | return 0; | |
916 | } | |
917 | ||
918 | /* | |
919 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
920 | * whether that can be merged with its predecessor or its successor. | |
921 | * Or both (it neatly fills a hole). | |
922 | * | |
923 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
924 | * certain not to be mapped by the time vma_merge is called; but when | |
925 | * called for mprotect, it is certain to be already mapped (either at | |
926 | * an offset within prev, or at the start of next), and the flags of | |
927 | * this area are about to be changed to vm_flags - and the no-change | |
928 | * case has already been eliminated. | |
929 | * | |
930 | * The following mprotect cases have to be considered, where AAAA is | |
931 | * the area passed down from mprotect_fixup, never extending beyond one | |
932 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
933 | * | |
934 | * AAAA AAAA AAAA AAAA | |
935 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
936 | * cannot merge might become might become might become | |
937 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
938 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
939 | * mremap move: PPPPNNNNNNNN 8 | |
940 | * AAAA | |
941 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
942 | * might become case 1 below case 2 below case 3 below | |
943 | * | |
944 | * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: | |
945 | * mprotect_fixup updates vm_flags & vm_page_prot on successful return. | |
946 | */ | |
947 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
948 | struct vm_area_struct *prev, unsigned long addr, | |
949 | unsigned long end, unsigned long vm_flags, | |
cc71aba3 | 950 | struct anon_vma *anon_vma, struct file *file, |
19a809af AA |
951 | pgoff_t pgoff, struct mempolicy *policy, |
952 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 LT |
953 | { |
954 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
955 | struct vm_area_struct *area, *next; | |
5beb4930 | 956 | int err; |
1da177e4 LT |
957 | |
958 | /* | |
959 | * We later require that vma->vm_flags == vm_flags, | |
960 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
961 | */ | |
962 | if (vm_flags & VM_SPECIAL) | |
963 | return NULL; | |
964 | ||
965 | if (prev) | |
966 | next = prev->vm_next; | |
967 | else | |
968 | next = mm->mmap; | |
969 | area = next; | |
970 | if (next && next->vm_end == end) /* cases 6, 7, 8 */ | |
971 | next = next->vm_next; | |
972 | ||
973 | /* | |
974 | * Can it merge with the predecessor? | |
975 | */ | |
976 | if (prev && prev->vm_end == addr && | |
cc71aba3 | 977 | mpol_equal(vma_policy(prev), policy) && |
1da177e4 | 978 | can_vma_merge_after(prev, vm_flags, |
19a809af AA |
979 | anon_vma, file, pgoff, |
980 | vm_userfaultfd_ctx)) { | |
1da177e4 LT |
981 | /* |
982 | * OK, it can. Can we now merge in the successor as well? | |
983 | */ | |
984 | if (next && end == next->vm_start && | |
985 | mpol_equal(policy, vma_policy(next)) && | |
986 | can_vma_merge_before(next, vm_flags, | |
19a809af AA |
987 | anon_vma, file, |
988 | pgoff+pglen, | |
989 | vm_userfaultfd_ctx) && | |
1da177e4 | 990 | is_mergeable_anon_vma(prev->anon_vma, |
965f55de | 991 | next->anon_vma, NULL)) { |
1da177e4 | 992 | /* cases 1, 6 */ |
5beb4930 | 993 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
994 | next->vm_end, prev->vm_pgoff, NULL); |
995 | } else /* cases 2, 5, 7 */ | |
5beb4930 | 996 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 | 997 | end, prev->vm_pgoff, NULL); |
5beb4930 RR |
998 | if (err) |
999 | return NULL; | |
6d50e60c | 1000 | khugepaged_enter_vma_merge(prev, vm_flags); |
1da177e4 LT |
1001 | return prev; |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * Can this new request be merged in front of next? | |
1006 | */ | |
1007 | if (next && end == next->vm_start && | |
cc71aba3 | 1008 | mpol_equal(policy, vma_policy(next)) && |
1da177e4 | 1009 | can_vma_merge_before(next, vm_flags, |
19a809af AA |
1010 | anon_vma, file, pgoff+pglen, |
1011 | vm_userfaultfd_ctx)) { | |
1da177e4 | 1012 | if (prev && addr < prev->vm_end) /* case 4 */ |
5beb4930 | 1013 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
1014 | addr, prev->vm_pgoff, NULL); |
1015 | else /* cases 3, 8 */ | |
5beb4930 | 1016 | err = vma_adjust(area, addr, next->vm_end, |
1da177e4 | 1017 | next->vm_pgoff - pglen, NULL); |
5beb4930 RR |
1018 | if (err) |
1019 | return NULL; | |
6d50e60c | 1020 | khugepaged_enter_vma_merge(area, vm_flags); |
1da177e4 LT |
1021 | return area; |
1022 | } | |
1023 | ||
1024 | return NULL; | |
1025 | } | |
1026 | ||
d0e9fe17 LT |
1027 | /* |
1028 | * Rough compatbility check to quickly see if it's even worth looking | |
1029 | * at sharing an anon_vma. | |
1030 | * | |
1031 | * They need to have the same vm_file, and the flags can only differ | |
1032 | * in things that mprotect may change. | |
1033 | * | |
1034 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1035 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1036 | * there is a vm_ops->close() function, because that indicates that the | |
1037 | * driver is doing some kind of reference counting. But that doesn't | |
1038 | * really matter for the anon_vma sharing case. | |
1039 | */ | |
1040 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1041 | { | |
1042 | return a->vm_end == b->vm_start && | |
1043 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1044 | a->vm_file == b->vm_file && | |
34228d47 | 1045 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && |
d0e9fe17 LT |
1046 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1051 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1052 | * the same as 'old', the other will be the new one that is trying | |
1053 | * to share the anon_vma. | |
1054 | * | |
1055 | * NOTE! This runs with mm_sem held for reading, so it is possible that | |
1056 | * the anon_vma of 'old' is concurrently in the process of being set up | |
1057 | * by another page fault trying to merge _that_. But that's ok: if it | |
1058 | * is being set up, that automatically means that it will be a singleton | |
1059 | * acceptable for merging, so we can do all of this optimistically. But | |
4db0c3c2 | 1060 | * we do that READ_ONCE() to make sure that we never re-load the pointer. |
d0e9fe17 LT |
1061 | * |
1062 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1063 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1064 | * is to return an anon_vma that is "complex" due to having gone through | |
1065 | * a fork). | |
1066 | * | |
1067 | * We also make sure that the two vma's are compatible (adjacent, | |
1068 | * and with the same memory policies). That's all stable, even with just | |
1069 | * a read lock on the mm_sem. | |
1070 | */ | |
1071 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1072 | { | |
1073 | if (anon_vma_compatible(a, b)) { | |
4db0c3c2 | 1074 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); |
d0e9fe17 LT |
1075 | |
1076 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1077 | return anon_vma; | |
1078 | } | |
1079 | return NULL; | |
1080 | } | |
1081 | ||
1da177e4 LT |
1082 | /* |
1083 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1084 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1085 | * to allocate a new anon_vma. It checks because a repetitive | |
1086 | * sequence of mprotects and faults may otherwise lead to distinct | |
1087 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1088 | * mprotect. | |
1089 | */ | |
1090 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1091 | { | |
d0e9fe17 | 1092 | struct anon_vma *anon_vma; |
1da177e4 | 1093 | struct vm_area_struct *near; |
1da177e4 LT |
1094 | |
1095 | near = vma->vm_next; | |
1096 | if (!near) | |
1097 | goto try_prev; | |
1098 | ||
d0e9fe17 LT |
1099 | anon_vma = reusable_anon_vma(near, vma, near); |
1100 | if (anon_vma) | |
1101 | return anon_vma; | |
1da177e4 | 1102 | try_prev: |
9be34c9d | 1103 | near = vma->vm_prev; |
1da177e4 LT |
1104 | if (!near) |
1105 | goto none; | |
1106 | ||
d0e9fe17 LT |
1107 | anon_vma = reusable_anon_vma(near, near, vma); |
1108 | if (anon_vma) | |
1109 | return anon_vma; | |
1da177e4 LT |
1110 | none: |
1111 | /* | |
1112 | * There's no absolute need to look only at touching neighbours: | |
1113 | * we could search further afield for "compatible" anon_vmas. | |
1114 | * But it would probably just be a waste of time searching, | |
1115 | * or lead to too many vmas hanging off the same anon_vma. | |
1116 | * We're trying to allow mprotect remerging later on, | |
1117 | * not trying to minimize memory used for anon_vmas. | |
1118 | */ | |
1119 | return NULL; | |
1120 | } | |
1121 | ||
40401530 AV |
1122 | /* |
1123 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1124 | * low as possible but still greater than mmap_min_addr | |
1125 | */ | |
1126 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1127 | { | |
1128 | hint &= PAGE_MASK; | |
1129 | if (((void *)hint != NULL) && | |
1130 | (hint < mmap_min_addr)) | |
1131 | return PAGE_ALIGN(mmap_min_addr); | |
1132 | return hint; | |
1133 | } | |
1134 | ||
363ee17f DB |
1135 | static inline int mlock_future_check(struct mm_struct *mm, |
1136 | unsigned long flags, | |
1137 | unsigned long len) | |
1138 | { | |
1139 | unsigned long locked, lock_limit; | |
1140 | ||
1141 | /* mlock MCL_FUTURE? */ | |
1142 | if (flags & VM_LOCKED) { | |
1143 | locked = len >> PAGE_SHIFT; | |
1144 | locked += mm->locked_vm; | |
1145 | lock_limit = rlimit(RLIMIT_MEMLOCK); | |
1146 | lock_limit >>= PAGE_SHIFT; | |
1147 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
1148 | return -EAGAIN; | |
1149 | } | |
1150 | return 0; | |
1151 | } | |
1152 | ||
1da177e4 | 1153 | /* |
27f5de79 | 1154 | * The caller must hold down_write(¤t->mm->mmap_sem). |
1da177e4 | 1155 | */ |
1fcfd8db | 1156 | unsigned long do_mmap(struct file *file, unsigned long addr, |
1da177e4 | 1157 | unsigned long len, unsigned long prot, |
1fcfd8db ON |
1158 | unsigned long flags, vm_flags_t vm_flags, |
1159 | unsigned long pgoff, unsigned long *populate) | |
1da177e4 | 1160 | { |
cc71aba3 | 1161 | struct mm_struct *mm = current->mm; |
62b5f7d0 | 1162 | int pkey = 0; |
1da177e4 | 1163 | |
41badc15 | 1164 | *populate = 0; |
bebeb3d6 | 1165 | |
e37609bb PK |
1166 | if (!len) |
1167 | return -EINVAL; | |
1168 | ||
1da177e4 LT |
1169 | /* |
1170 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1171 | * | |
1172 | * (the exception is when the underlying filesystem is noexec | |
1173 | * mounted, in which case we dont add PROT_EXEC.) | |
1174 | */ | |
1175 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
90f8572b | 1176 | if (!(file && path_noexec(&file->f_path))) |
1da177e4 LT |
1177 | prot |= PROT_EXEC; |
1178 | ||
7cd94146 EP |
1179 | if (!(flags & MAP_FIXED)) |
1180 | addr = round_hint_to_min(addr); | |
1181 | ||
1da177e4 LT |
1182 | /* Careful about overflows.. */ |
1183 | len = PAGE_ALIGN(len); | |
9206de95 | 1184 | if (!len) |
1da177e4 LT |
1185 | return -ENOMEM; |
1186 | ||
1187 | /* offset overflow? */ | |
1188 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
cc71aba3 | 1189 | return -EOVERFLOW; |
1da177e4 LT |
1190 | |
1191 | /* Too many mappings? */ | |
1192 | if (mm->map_count > sysctl_max_map_count) | |
1193 | return -ENOMEM; | |
1194 | ||
1195 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1196 | * that it represents a valid section of the address space. | |
1197 | */ | |
1198 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
de1741a1 | 1199 | if (offset_in_page(addr)) |
1da177e4 LT |
1200 | return addr; |
1201 | ||
62b5f7d0 DH |
1202 | if (prot == PROT_EXEC) { |
1203 | pkey = execute_only_pkey(mm); | |
1204 | if (pkey < 0) | |
1205 | pkey = 0; | |
1206 | } | |
1207 | ||
1da177e4 LT |
1208 | /* Do simple checking here so the lower-level routines won't have |
1209 | * to. we assume access permissions have been handled by the open | |
1210 | * of the memory object, so we don't do any here. | |
1211 | */ | |
62b5f7d0 | 1212 | vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | |
1da177e4 LT |
1213 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1214 | ||
cdf7b341 | 1215 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1216 | if (!can_do_mlock()) |
1217 | return -EPERM; | |
ba470de4 | 1218 | |
363ee17f DB |
1219 | if (mlock_future_check(mm, vm_flags, len)) |
1220 | return -EAGAIN; | |
1da177e4 | 1221 | |
1da177e4 | 1222 | if (file) { |
077bf22b ON |
1223 | struct inode *inode = file_inode(file); |
1224 | ||
1da177e4 LT |
1225 | switch (flags & MAP_TYPE) { |
1226 | case MAP_SHARED: | |
1227 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
1228 | return -EACCES; | |
1229 | ||
1230 | /* | |
1231 | * Make sure we don't allow writing to an append-only | |
1232 | * file.. | |
1233 | */ | |
1234 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1235 | return -EACCES; | |
1236 | ||
1237 | /* | |
1238 | * Make sure there are no mandatory locks on the file. | |
1239 | */ | |
d7a06983 | 1240 | if (locks_verify_locked(file)) |
1da177e4 LT |
1241 | return -EAGAIN; |
1242 | ||
1243 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
1244 | if (!(file->f_mode & FMODE_WRITE)) | |
1245 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
1246 | ||
1247 | /* fall through */ | |
1248 | case MAP_PRIVATE: | |
1249 | if (!(file->f_mode & FMODE_READ)) | |
1250 | return -EACCES; | |
90f8572b | 1251 | if (path_noexec(&file->f_path)) { |
80c5606c LT |
1252 | if (vm_flags & VM_EXEC) |
1253 | return -EPERM; | |
1254 | vm_flags &= ~VM_MAYEXEC; | |
1255 | } | |
80c5606c | 1256 | |
72c2d531 | 1257 | if (!file->f_op->mmap) |
80c5606c | 1258 | return -ENODEV; |
b2c56e4f ON |
1259 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1260 | return -EINVAL; | |
1da177e4 LT |
1261 | break; |
1262 | ||
1263 | default: | |
1264 | return -EINVAL; | |
1265 | } | |
1266 | } else { | |
1267 | switch (flags & MAP_TYPE) { | |
1268 | case MAP_SHARED: | |
b2c56e4f ON |
1269 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1270 | return -EINVAL; | |
ce363942 TH |
1271 | /* |
1272 | * Ignore pgoff. | |
1273 | */ | |
1274 | pgoff = 0; | |
1da177e4 LT |
1275 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1276 | break; | |
1277 | case MAP_PRIVATE: | |
1278 | /* | |
1279 | * Set pgoff according to addr for anon_vma. | |
1280 | */ | |
1281 | pgoff = addr >> PAGE_SHIFT; | |
1282 | break; | |
1283 | default: | |
1284 | return -EINVAL; | |
1285 | } | |
1286 | } | |
1287 | ||
c22c0d63 ML |
1288 | /* |
1289 | * Set 'VM_NORESERVE' if we should not account for the | |
1290 | * memory use of this mapping. | |
1291 | */ | |
1292 | if (flags & MAP_NORESERVE) { | |
1293 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1294 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1295 | vm_flags |= VM_NORESERVE; | |
1296 | ||
1297 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1298 | if (file && is_file_hugepages(file)) | |
1299 | vm_flags |= VM_NORESERVE; | |
1300 | } | |
1301 | ||
1302 | addr = mmap_region(file, addr, len, vm_flags, pgoff); | |
09a9f1d2 ML |
1303 | if (!IS_ERR_VALUE(addr) && |
1304 | ((vm_flags & VM_LOCKED) || | |
1305 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1306 | *populate = len; |
bebeb3d6 | 1307 | return addr; |
0165ab44 | 1308 | } |
6be5ceb0 | 1309 | |
66f0dc48 HD |
1310 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1311 | unsigned long, prot, unsigned long, flags, | |
1312 | unsigned long, fd, unsigned long, pgoff) | |
1313 | { | |
1314 | struct file *file = NULL; | |
1e3ee14b | 1315 | unsigned long retval; |
66f0dc48 HD |
1316 | |
1317 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1318 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1319 | file = fget(fd); |
1320 | if (!file) | |
1e3ee14b | 1321 | return -EBADF; |
af73e4d9 NH |
1322 | if (is_file_hugepages(file)) |
1323 | len = ALIGN(len, huge_page_size(hstate_file(file))); | |
493af578 JE |
1324 | retval = -EINVAL; |
1325 | if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) | |
1326 | goto out_fput; | |
66f0dc48 HD |
1327 | } else if (flags & MAP_HUGETLB) { |
1328 | struct user_struct *user = NULL; | |
c103a4dc | 1329 | struct hstate *hs; |
af73e4d9 | 1330 | |
c103a4dc | 1331 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); |
091d0d55 LZ |
1332 | if (!hs) |
1333 | return -EINVAL; | |
1334 | ||
1335 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1336 | /* |
1337 | * VM_NORESERVE is used because the reservations will be | |
1338 | * taken when vm_ops->mmap() is called | |
1339 | * A dummy user value is used because we are not locking | |
1340 | * memory so no accounting is necessary | |
1341 | */ | |
af73e4d9 | 1342 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f AK |
1343 | VM_NORESERVE, |
1344 | &user, HUGETLB_ANONHUGE_INODE, | |
1345 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); | |
66f0dc48 HD |
1346 | if (IS_ERR(file)) |
1347 | return PTR_ERR(file); | |
1348 | } | |
1349 | ||
1350 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1351 | ||
9fbeb5ab | 1352 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
493af578 | 1353 | out_fput: |
66f0dc48 HD |
1354 | if (file) |
1355 | fput(file); | |
66f0dc48 HD |
1356 | return retval; |
1357 | } | |
1358 | ||
a4679373 CH |
1359 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1360 | struct mmap_arg_struct { | |
1361 | unsigned long addr; | |
1362 | unsigned long len; | |
1363 | unsigned long prot; | |
1364 | unsigned long flags; | |
1365 | unsigned long fd; | |
1366 | unsigned long offset; | |
1367 | }; | |
1368 | ||
1369 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1370 | { | |
1371 | struct mmap_arg_struct a; | |
1372 | ||
1373 | if (copy_from_user(&a, arg, sizeof(a))) | |
1374 | return -EFAULT; | |
de1741a1 | 1375 | if (offset_in_page(a.offset)) |
a4679373 CH |
1376 | return -EINVAL; |
1377 | ||
1378 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1379 | a.offset >> PAGE_SHIFT); | |
1380 | } | |
1381 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1382 | ||
4e950f6f AD |
1383 | /* |
1384 | * Some shared mappigns will want the pages marked read-only | |
1385 | * to track write events. If so, we'll downgrade vm_page_prot | |
1386 | * to the private version (using protection_map[] without the | |
1387 | * VM_SHARED bit). | |
1388 | */ | |
1389 | int vma_wants_writenotify(struct vm_area_struct *vma) | |
1390 | { | |
ca16d140 | 1391 | vm_flags_t vm_flags = vma->vm_flags; |
8a04446a | 1392 | const struct vm_operations_struct *vm_ops = vma->vm_ops; |
4e950f6f AD |
1393 | |
1394 | /* If it was private or non-writable, the write bit is already clear */ | |
1395 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1396 | return 0; | |
1397 | ||
1398 | /* The backer wishes to know when pages are first written to? */ | |
8a04446a | 1399 | if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) |
4e950f6f AD |
1400 | return 1; |
1401 | ||
64e45507 PF |
1402 | /* The open routine did something to the protections that pgprot_modify |
1403 | * won't preserve? */ | |
4e950f6f | 1404 | if (pgprot_val(vma->vm_page_prot) != |
64e45507 | 1405 | pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags))) |
4e950f6f AD |
1406 | return 0; |
1407 | ||
64e45507 PF |
1408 | /* Do we need to track softdirty? */ |
1409 | if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) | |
1410 | return 1; | |
1411 | ||
4e950f6f | 1412 | /* Specialty mapping? */ |
4b6e1e37 | 1413 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1414 | return 0; |
1415 | ||
1416 | /* Can the mapping track the dirty pages? */ | |
1417 | return vma->vm_file && vma->vm_file->f_mapping && | |
1418 | mapping_cap_account_dirty(vma->vm_file->f_mapping); | |
1419 | } | |
1420 | ||
fc8744ad LT |
1421 | /* |
1422 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1423 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1424 | */ |
ca16d140 | 1425 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1426 | { |
5a6fe125 MG |
1427 | /* |
1428 | * hugetlb has its own accounting separate from the core VM | |
1429 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1430 | */ | |
1431 | if (file && is_file_hugepages(file)) | |
1432 | return 0; | |
1433 | ||
fc8744ad LT |
1434 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1435 | } | |
1436 | ||
0165ab44 | 1437 | unsigned long mmap_region(struct file *file, unsigned long addr, |
c22c0d63 | 1438 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) |
0165ab44 MS |
1439 | { |
1440 | struct mm_struct *mm = current->mm; | |
1441 | struct vm_area_struct *vma, *prev; | |
0165ab44 MS |
1442 | int error; |
1443 | struct rb_node **rb_link, *rb_parent; | |
1444 | unsigned long charged = 0; | |
0165ab44 | 1445 | |
e8420a8e | 1446 | /* Check against address space limit. */ |
84638335 | 1447 | if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { |
e8420a8e CH |
1448 | unsigned long nr_pages; |
1449 | ||
1450 | /* | |
1451 | * MAP_FIXED may remove pages of mappings that intersects with | |
1452 | * requested mapping. Account for the pages it would unmap. | |
1453 | */ | |
e8420a8e CH |
1454 | nr_pages = count_vma_pages_range(mm, addr, addr + len); |
1455 | ||
84638335 KK |
1456 | if (!may_expand_vm(mm, vm_flags, |
1457 | (len >> PAGE_SHIFT) - nr_pages)) | |
e8420a8e CH |
1458 | return -ENOMEM; |
1459 | } | |
1460 | ||
1da177e4 | 1461 | /* Clear old maps */ |
9fcd1457 RV |
1462 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
1463 | &rb_parent)) { | |
1da177e4 LT |
1464 | if (do_munmap(mm, addr, len)) |
1465 | return -ENOMEM; | |
1da177e4 LT |
1466 | } |
1467 | ||
fc8744ad LT |
1468 | /* |
1469 | * Private writable mapping: check memory availability | |
1470 | */ | |
5a6fe125 | 1471 | if (accountable_mapping(file, vm_flags)) { |
fc8744ad | 1472 | charged = len >> PAGE_SHIFT; |
191c5424 | 1473 | if (security_vm_enough_memory_mm(mm, charged)) |
fc8744ad LT |
1474 | return -ENOMEM; |
1475 | vm_flags |= VM_ACCOUNT; | |
1da177e4 LT |
1476 | } |
1477 | ||
1478 | /* | |
de33c8db | 1479 | * Can we just expand an old mapping? |
1da177e4 | 1480 | */ |
19a809af AA |
1481 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, |
1482 | NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); | |
de33c8db LT |
1483 | if (vma) |
1484 | goto out; | |
1da177e4 LT |
1485 | |
1486 | /* | |
1487 | * Determine the object being mapped and call the appropriate | |
1488 | * specific mapper. the address has already been validated, but | |
1489 | * not unmapped, but the maps are removed from the list. | |
1490 | */ | |
c5e3b83e | 1491 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1492 | if (!vma) { |
1493 | error = -ENOMEM; | |
1494 | goto unacct_error; | |
1495 | } | |
1da177e4 LT |
1496 | |
1497 | vma->vm_mm = mm; | |
1498 | vma->vm_start = addr; | |
1499 | vma->vm_end = addr + len; | |
1500 | vma->vm_flags = vm_flags; | |
3ed75eb8 | 1501 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
1da177e4 | 1502 | vma->vm_pgoff = pgoff; |
5beb4930 | 1503 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
1504 | |
1505 | if (file) { | |
1da177e4 LT |
1506 | if (vm_flags & VM_DENYWRITE) { |
1507 | error = deny_write_access(file); | |
1508 | if (error) | |
1509 | goto free_vma; | |
1da177e4 | 1510 | } |
4bb5f5d9 DH |
1511 | if (vm_flags & VM_SHARED) { |
1512 | error = mapping_map_writable(file->f_mapping); | |
1513 | if (error) | |
1514 | goto allow_write_and_free_vma; | |
1515 | } | |
1516 | ||
1517 | /* ->mmap() can change vma->vm_file, but must guarantee that | |
1518 | * vma_link() below can deny write-access if VM_DENYWRITE is set | |
1519 | * and map writably if VM_SHARED is set. This usually means the | |
1520 | * new file must not have been exposed to user-space, yet. | |
1521 | */ | |
cb0942b8 | 1522 | vma->vm_file = get_file(file); |
1da177e4 LT |
1523 | error = file->f_op->mmap(file, vma); |
1524 | if (error) | |
1525 | goto unmap_and_free_vma; | |
f8dbf0a7 HS |
1526 | |
1527 | /* Can addr have changed?? | |
1528 | * | |
1529 | * Answer: Yes, several device drivers can do it in their | |
1530 | * f_op->mmap method. -DaveM | |
2897b4d2 JK |
1531 | * Bug: If addr is changed, prev, rb_link, rb_parent should |
1532 | * be updated for vma_link() | |
f8dbf0a7 | 1533 | */ |
2897b4d2 JK |
1534 | WARN_ON_ONCE(addr != vma->vm_start); |
1535 | ||
f8dbf0a7 | 1536 | addr = vma->vm_start; |
f8dbf0a7 | 1537 | vm_flags = vma->vm_flags; |
1da177e4 LT |
1538 | } else if (vm_flags & VM_SHARED) { |
1539 | error = shmem_zero_setup(vma); | |
1540 | if (error) | |
1541 | goto free_vma; | |
1542 | } | |
1543 | ||
de33c8db | 1544 | vma_link(mm, vma, prev, rb_link, rb_parent); |
4d3d5b41 | 1545 | /* Once vma denies write, undo our temporary denial count */ |
4bb5f5d9 DH |
1546 | if (file) { |
1547 | if (vm_flags & VM_SHARED) | |
1548 | mapping_unmap_writable(file->f_mapping); | |
1549 | if (vm_flags & VM_DENYWRITE) | |
1550 | allow_write_access(file); | |
1551 | } | |
e8686772 | 1552 | file = vma->vm_file; |
4d3d5b41 | 1553 | out: |
cdd6c482 | 1554 | perf_event_mmap(vma); |
0a4a9391 | 1555 | |
84638335 | 1556 | vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); |
1da177e4 | 1557 | if (vm_flags & VM_LOCKED) { |
bebeb3d6 ML |
1558 | if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || |
1559 | vma == get_gate_vma(current->mm))) | |
06f9d8c2 | 1560 | mm->locked_vm += (len >> PAGE_SHIFT); |
bebeb3d6 | 1561 | else |
de60f5f1 | 1562 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
bebeb3d6 | 1563 | } |
2b144498 | 1564 | |
c7a3a88c ON |
1565 | if (file) |
1566 | uprobe_mmap(vma); | |
2b144498 | 1567 | |
d9104d1c CG |
1568 | /* |
1569 | * New (or expanded) vma always get soft dirty status. | |
1570 | * Otherwise user-space soft-dirty page tracker won't | |
1571 | * be able to distinguish situation when vma area unmapped, | |
1572 | * then new mapped in-place (which must be aimed as | |
1573 | * a completely new data area). | |
1574 | */ | |
1575 | vma->vm_flags |= VM_SOFTDIRTY; | |
1576 | ||
64e45507 PF |
1577 | vma_set_page_prot(vma); |
1578 | ||
1da177e4 LT |
1579 | return addr; |
1580 | ||
1581 | unmap_and_free_vma: | |
1da177e4 LT |
1582 | vma->vm_file = NULL; |
1583 | fput(file); | |
1584 | ||
1585 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1586 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1587 | charged = 0; | |
4bb5f5d9 DH |
1588 | if (vm_flags & VM_SHARED) |
1589 | mapping_unmap_writable(file->f_mapping); | |
1590 | allow_write_and_free_vma: | |
1591 | if (vm_flags & VM_DENYWRITE) | |
1592 | allow_write_access(file); | |
1da177e4 LT |
1593 | free_vma: |
1594 | kmem_cache_free(vm_area_cachep, vma); | |
1595 | unacct_error: | |
1596 | if (charged) | |
1597 | vm_unacct_memory(charged); | |
1598 | return error; | |
1599 | } | |
1600 | ||
db4fbfb9 ML |
1601 | unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
1602 | { | |
1603 | /* | |
1604 | * We implement the search by looking for an rbtree node that | |
1605 | * immediately follows a suitable gap. That is, | |
1606 | * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; | |
1607 | * - gap_end = vma->vm_start >= info->low_limit + length; | |
1608 | * - gap_end - gap_start >= length | |
1609 | */ | |
1610 | ||
1611 | struct mm_struct *mm = current->mm; | |
1612 | struct vm_area_struct *vma; | |
1613 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1614 | ||
1615 | /* Adjust search length to account for worst case alignment overhead */ | |
1616 | length = info->length + info->align_mask; | |
1617 | if (length < info->length) | |
1618 | return -ENOMEM; | |
1619 | ||
1620 | /* Adjust search limits by the desired length */ | |
1621 | if (info->high_limit < length) | |
1622 | return -ENOMEM; | |
1623 | high_limit = info->high_limit - length; | |
1624 | ||
1625 | if (info->low_limit > high_limit) | |
1626 | return -ENOMEM; | |
1627 | low_limit = info->low_limit + length; | |
1628 | ||
1629 | /* Check if rbtree root looks promising */ | |
1630 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1631 | goto check_highest; | |
1632 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1633 | if (vma->rb_subtree_gap < length) | |
1634 | goto check_highest; | |
1635 | ||
1636 | while (true) { | |
1637 | /* Visit left subtree if it looks promising */ | |
1638 | gap_end = vma->vm_start; | |
1639 | if (gap_end >= low_limit && vma->vm_rb.rb_left) { | |
1640 | struct vm_area_struct *left = | |
1641 | rb_entry(vma->vm_rb.rb_left, | |
1642 | struct vm_area_struct, vm_rb); | |
1643 | if (left->rb_subtree_gap >= length) { | |
1644 | vma = left; | |
1645 | continue; | |
1646 | } | |
1647 | } | |
1648 | ||
1649 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1650 | check_current: | |
1651 | /* Check if current node has a suitable gap */ | |
1652 | if (gap_start > high_limit) | |
1653 | return -ENOMEM; | |
1654 | if (gap_end >= low_limit && gap_end - gap_start >= length) | |
1655 | goto found; | |
1656 | ||
1657 | /* Visit right subtree if it looks promising */ | |
1658 | if (vma->vm_rb.rb_right) { | |
1659 | struct vm_area_struct *right = | |
1660 | rb_entry(vma->vm_rb.rb_right, | |
1661 | struct vm_area_struct, vm_rb); | |
1662 | if (right->rb_subtree_gap >= length) { | |
1663 | vma = right; | |
1664 | continue; | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | /* Go back up the rbtree to find next candidate node */ | |
1669 | while (true) { | |
1670 | struct rb_node *prev = &vma->vm_rb; | |
1671 | if (!rb_parent(prev)) | |
1672 | goto check_highest; | |
1673 | vma = rb_entry(rb_parent(prev), | |
1674 | struct vm_area_struct, vm_rb); | |
1675 | if (prev == vma->vm_rb.rb_left) { | |
1676 | gap_start = vma->vm_prev->vm_end; | |
1677 | gap_end = vma->vm_start; | |
1678 | goto check_current; | |
1679 | } | |
1680 | } | |
1681 | } | |
1682 | ||
1683 | check_highest: | |
1684 | /* Check highest gap, which does not precede any rbtree node */ | |
1685 | gap_start = mm->highest_vm_end; | |
1686 | gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ | |
1687 | if (gap_start > high_limit) | |
1688 | return -ENOMEM; | |
1689 | ||
1690 | found: | |
1691 | /* We found a suitable gap. Clip it with the original low_limit. */ | |
1692 | if (gap_start < info->low_limit) | |
1693 | gap_start = info->low_limit; | |
1694 | ||
1695 | /* Adjust gap address to the desired alignment */ | |
1696 | gap_start += (info->align_offset - gap_start) & info->align_mask; | |
1697 | ||
1698 | VM_BUG_ON(gap_start + info->length > info->high_limit); | |
1699 | VM_BUG_ON(gap_start + info->length > gap_end); | |
1700 | return gap_start; | |
1701 | } | |
1702 | ||
1703 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) | |
1704 | { | |
1705 | struct mm_struct *mm = current->mm; | |
1706 | struct vm_area_struct *vma; | |
1707 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1708 | ||
1709 | /* Adjust search length to account for worst case alignment overhead */ | |
1710 | length = info->length + info->align_mask; | |
1711 | if (length < info->length) | |
1712 | return -ENOMEM; | |
1713 | ||
1714 | /* | |
1715 | * Adjust search limits by the desired length. | |
1716 | * See implementation comment at top of unmapped_area(). | |
1717 | */ | |
1718 | gap_end = info->high_limit; | |
1719 | if (gap_end < length) | |
1720 | return -ENOMEM; | |
1721 | high_limit = gap_end - length; | |
1722 | ||
1723 | if (info->low_limit > high_limit) | |
1724 | return -ENOMEM; | |
1725 | low_limit = info->low_limit + length; | |
1726 | ||
1727 | /* Check highest gap, which does not precede any rbtree node */ | |
1728 | gap_start = mm->highest_vm_end; | |
1729 | if (gap_start <= high_limit) | |
1730 | goto found_highest; | |
1731 | ||
1732 | /* Check if rbtree root looks promising */ | |
1733 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1734 | return -ENOMEM; | |
1735 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1736 | if (vma->rb_subtree_gap < length) | |
1737 | return -ENOMEM; | |
1738 | ||
1739 | while (true) { | |
1740 | /* Visit right subtree if it looks promising */ | |
1741 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1742 | if (gap_start <= high_limit && vma->vm_rb.rb_right) { | |
1743 | struct vm_area_struct *right = | |
1744 | rb_entry(vma->vm_rb.rb_right, | |
1745 | struct vm_area_struct, vm_rb); | |
1746 | if (right->rb_subtree_gap >= length) { | |
1747 | vma = right; | |
1748 | continue; | |
1749 | } | |
1750 | } | |
1751 | ||
1752 | check_current: | |
1753 | /* Check if current node has a suitable gap */ | |
1754 | gap_end = vma->vm_start; | |
1755 | if (gap_end < low_limit) | |
1756 | return -ENOMEM; | |
1757 | if (gap_start <= high_limit && gap_end - gap_start >= length) | |
1758 | goto found; | |
1759 | ||
1760 | /* Visit left subtree if it looks promising */ | |
1761 | if (vma->vm_rb.rb_left) { | |
1762 | struct vm_area_struct *left = | |
1763 | rb_entry(vma->vm_rb.rb_left, | |
1764 | struct vm_area_struct, vm_rb); | |
1765 | if (left->rb_subtree_gap >= length) { | |
1766 | vma = left; | |
1767 | continue; | |
1768 | } | |
1769 | } | |
1770 | ||
1771 | /* Go back up the rbtree to find next candidate node */ | |
1772 | while (true) { | |
1773 | struct rb_node *prev = &vma->vm_rb; | |
1774 | if (!rb_parent(prev)) | |
1775 | return -ENOMEM; | |
1776 | vma = rb_entry(rb_parent(prev), | |
1777 | struct vm_area_struct, vm_rb); | |
1778 | if (prev == vma->vm_rb.rb_right) { | |
1779 | gap_start = vma->vm_prev ? | |
1780 | vma->vm_prev->vm_end : 0; | |
1781 | goto check_current; | |
1782 | } | |
1783 | } | |
1784 | } | |
1785 | ||
1786 | found: | |
1787 | /* We found a suitable gap. Clip it with the original high_limit. */ | |
1788 | if (gap_end > info->high_limit) | |
1789 | gap_end = info->high_limit; | |
1790 | ||
1791 | found_highest: | |
1792 | /* Compute highest gap address at the desired alignment */ | |
1793 | gap_end -= info->length; | |
1794 | gap_end -= (gap_end - info->align_offset) & info->align_mask; | |
1795 | ||
1796 | VM_BUG_ON(gap_end < info->low_limit); | |
1797 | VM_BUG_ON(gap_end < gap_start); | |
1798 | return gap_end; | |
1799 | } | |
1800 | ||
1da177e4 LT |
1801 | /* Get an address range which is currently unmapped. |
1802 | * For shmat() with addr=0. | |
1803 | * | |
1804 | * Ugly calling convention alert: | |
1805 | * Return value with the low bits set means error value, | |
1806 | * ie | |
1807 | * if (ret & ~PAGE_MASK) | |
1808 | * error = ret; | |
1809 | * | |
1810 | * This function "knows" that -ENOMEM has the bits set. | |
1811 | */ | |
1812 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1813 | unsigned long | |
1814 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1815 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1816 | { | |
1817 | struct mm_struct *mm = current->mm; | |
1818 | struct vm_area_struct *vma; | |
db4fbfb9 | 1819 | struct vm_unmapped_area_info info; |
1da177e4 | 1820 | |
2afc745f | 1821 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1822 | return -ENOMEM; |
1823 | ||
06abdfb4 BH |
1824 | if (flags & MAP_FIXED) |
1825 | return addr; | |
1826 | ||
1da177e4 LT |
1827 | if (addr) { |
1828 | addr = PAGE_ALIGN(addr); | |
1829 | vma = find_vma(mm, addr); | |
2afc745f | 1830 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1831 | (!vma || addr + len <= vma->vm_start)) |
1832 | return addr; | |
1833 | } | |
1da177e4 | 1834 | |
db4fbfb9 ML |
1835 | info.flags = 0; |
1836 | info.length = len; | |
4e99b021 | 1837 | info.low_limit = mm->mmap_base; |
db4fbfb9 ML |
1838 | info.high_limit = TASK_SIZE; |
1839 | info.align_mask = 0; | |
1840 | return vm_unmapped_area(&info); | |
1da177e4 | 1841 | } |
cc71aba3 | 1842 | #endif |
1da177e4 | 1843 | |
1da177e4 LT |
1844 | /* |
1845 | * This mmap-allocator allocates new areas top-down from below the | |
1846 | * stack's low limit (the base): | |
1847 | */ | |
1848 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1849 | unsigned long | |
1850 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1851 | const unsigned long len, const unsigned long pgoff, | |
1852 | const unsigned long flags) | |
1853 | { | |
1854 | struct vm_area_struct *vma; | |
1855 | struct mm_struct *mm = current->mm; | |
db4fbfb9 ML |
1856 | unsigned long addr = addr0; |
1857 | struct vm_unmapped_area_info info; | |
1da177e4 LT |
1858 | |
1859 | /* requested length too big for entire address space */ | |
2afc745f | 1860 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1861 | return -ENOMEM; |
1862 | ||
06abdfb4 BH |
1863 | if (flags & MAP_FIXED) |
1864 | return addr; | |
1865 | ||
1da177e4 LT |
1866 | /* requesting a specific address */ |
1867 | if (addr) { | |
1868 | addr = PAGE_ALIGN(addr); | |
1869 | vma = find_vma(mm, addr); | |
2afc745f | 1870 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1871 | (!vma || addr + len <= vma->vm_start)) |
1872 | return addr; | |
1873 | } | |
1874 | ||
db4fbfb9 ML |
1875 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
1876 | info.length = len; | |
2afc745f | 1877 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); |
db4fbfb9 ML |
1878 | info.high_limit = mm->mmap_base; |
1879 | info.align_mask = 0; | |
1880 | addr = vm_unmapped_area(&info); | |
b716ad95 | 1881 | |
1da177e4 LT |
1882 | /* |
1883 | * A failed mmap() very likely causes application failure, | |
1884 | * so fall back to the bottom-up function here. This scenario | |
1885 | * can happen with large stack limits and large mmap() | |
1886 | * allocations. | |
1887 | */ | |
de1741a1 | 1888 | if (offset_in_page(addr)) { |
db4fbfb9 ML |
1889 | VM_BUG_ON(addr != -ENOMEM); |
1890 | info.flags = 0; | |
1891 | info.low_limit = TASK_UNMAPPED_BASE; | |
1892 | info.high_limit = TASK_SIZE; | |
1893 | addr = vm_unmapped_area(&info); | |
1894 | } | |
1da177e4 LT |
1895 | |
1896 | return addr; | |
1897 | } | |
1898 | #endif | |
1899 | ||
1da177e4 LT |
1900 | unsigned long |
1901 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1902 | unsigned long pgoff, unsigned long flags) | |
1903 | { | |
06abdfb4 BH |
1904 | unsigned long (*get_area)(struct file *, unsigned long, |
1905 | unsigned long, unsigned long, unsigned long); | |
1906 | ||
9206de95 AV |
1907 | unsigned long error = arch_mmap_check(addr, len, flags); |
1908 | if (error) | |
1909 | return error; | |
1910 | ||
1911 | /* Careful about overflows.. */ | |
1912 | if (len > TASK_SIZE) | |
1913 | return -ENOMEM; | |
1914 | ||
06abdfb4 | 1915 | get_area = current->mm->get_unmapped_area; |
c01d5b30 HD |
1916 | if (file) { |
1917 | if (file->f_op->get_unmapped_area) | |
1918 | get_area = file->f_op->get_unmapped_area; | |
1919 | } else if (flags & MAP_SHARED) { | |
1920 | /* | |
1921 | * mmap_region() will call shmem_zero_setup() to create a file, | |
1922 | * so use shmem's get_unmapped_area in case it can be huge. | |
1923 | * do_mmap_pgoff() will clear pgoff, so match alignment. | |
1924 | */ | |
1925 | pgoff = 0; | |
1926 | get_area = shmem_get_unmapped_area; | |
1927 | } | |
1928 | ||
06abdfb4 BH |
1929 | addr = get_area(file, addr, len, pgoff, flags); |
1930 | if (IS_ERR_VALUE(addr)) | |
1931 | return addr; | |
1da177e4 | 1932 | |
07ab67c8 LT |
1933 | if (addr > TASK_SIZE - len) |
1934 | return -ENOMEM; | |
de1741a1 | 1935 | if (offset_in_page(addr)) |
07ab67c8 | 1936 | return -EINVAL; |
06abdfb4 | 1937 | |
9ac4ed4b AV |
1938 | error = security_mmap_addr(addr); |
1939 | return error ? error : addr; | |
1da177e4 LT |
1940 | } |
1941 | ||
1942 | EXPORT_SYMBOL(get_unmapped_area); | |
1943 | ||
1944 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
48aae425 | 1945 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 1946 | { |
615d6e87 DB |
1947 | struct rb_node *rb_node; |
1948 | struct vm_area_struct *vma; | |
1da177e4 | 1949 | |
841e31e5 | 1950 | /* Check the cache first. */ |
615d6e87 DB |
1951 | vma = vmacache_find(mm, addr); |
1952 | if (likely(vma)) | |
1953 | return vma; | |
841e31e5 | 1954 | |
615d6e87 | 1955 | rb_node = mm->mm_rb.rb_node; |
841e31e5 | 1956 | |
615d6e87 DB |
1957 | while (rb_node) { |
1958 | struct vm_area_struct *tmp; | |
1959 | ||
1960 | tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
1961 | ||
1962 | if (tmp->vm_end > addr) { | |
1963 | vma = tmp; | |
1964 | if (tmp->vm_start <= addr) | |
1965 | break; | |
1966 | rb_node = rb_node->rb_left; | |
1967 | } else | |
1968 | rb_node = rb_node->rb_right; | |
1da177e4 | 1969 | } |
615d6e87 DB |
1970 | |
1971 | if (vma) | |
1972 | vmacache_update(addr, vma); | |
1da177e4 LT |
1973 | return vma; |
1974 | } | |
1975 | ||
1976 | EXPORT_SYMBOL(find_vma); | |
1977 | ||
6bd4837d KM |
1978 | /* |
1979 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. | |
6bd4837d | 1980 | */ |
1da177e4 LT |
1981 | struct vm_area_struct * |
1982 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
1983 | struct vm_area_struct **pprev) | |
1984 | { | |
6bd4837d | 1985 | struct vm_area_struct *vma; |
1da177e4 | 1986 | |
6bd4837d | 1987 | vma = find_vma(mm, addr); |
83cd904d MP |
1988 | if (vma) { |
1989 | *pprev = vma->vm_prev; | |
1990 | } else { | |
1991 | struct rb_node *rb_node = mm->mm_rb.rb_node; | |
1992 | *pprev = NULL; | |
1993 | while (rb_node) { | |
1994 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
1995 | rb_node = rb_node->rb_right; | |
1996 | } | |
1997 | } | |
6bd4837d | 1998 | return vma; |
1da177e4 LT |
1999 | } |
2000 | ||
2001 | /* | |
2002 | * Verify that the stack growth is acceptable and | |
2003 | * update accounting. This is shared with both the | |
2004 | * grow-up and grow-down cases. | |
2005 | */ | |
48aae425 | 2006 | static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) |
1da177e4 LT |
2007 | { |
2008 | struct mm_struct *mm = vma->vm_mm; | |
2009 | struct rlimit *rlim = current->signal->rlim; | |
690eac53 | 2010 | unsigned long new_start, actual_size; |
1da177e4 LT |
2011 | |
2012 | /* address space limit tests */ | |
84638335 | 2013 | if (!may_expand_vm(mm, vma->vm_flags, grow)) |
1da177e4 LT |
2014 | return -ENOMEM; |
2015 | ||
2016 | /* Stack limit test */ | |
690eac53 LT |
2017 | actual_size = size; |
2018 | if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) | |
2019 | actual_size -= PAGE_SIZE; | |
4db0c3c2 | 2020 | if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur)) |
1da177e4 LT |
2021 | return -ENOMEM; |
2022 | ||
2023 | /* mlock limit tests */ | |
2024 | if (vma->vm_flags & VM_LOCKED) { | |
2025 | unsigned long locked; | |
2026 | unsigned long limit; | |
2027 | locked = mm->locked_vm + grow; | |
4db0c3c2 | 2028 | limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); |
59e99e5b | 2029 | limit >>= PAGE_SHIFT; |
1da177e4 LT |
2030 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
2031 | return -ENOMEM; | |
2032 | } | |
2033 | ||
0d59a01b AL |
2034 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
2035 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
2036 | vma->vm_end - size; | |
2037 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
2038 | return -EFAULT; | |
2039 | ||
1da177e4 LT |
2040 | /* |
2041 | * Overcommit.. This must be the final test, as it will | |
2042 | * update security statistics. | |
2043 | */ | |
05fa199d | 2044 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
2045 | return -ENOMEM; |
2046 | ||
1da177e4 LT |
2047 | return 0; |
2048 | } | |
2049 | ||
46dea3d0 | 2050 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 2051 | /* |
46dea3d0 HD |
2052 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
2053 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 2054 | */ |
46dea3d0 | 2055 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 2056 | { |
09357814 | 2057 | struct mm_struct *mm = vma->vm_mm; |
12352d3c | 2058 | int error = 0; |
1da177e4 LT |
2059 | |
2060 | if (!(vma->vm_flags & VM_GROWSUP)) | |
2061 | return -EFAULT; | |
2062 | ||
12352d3c KK |
2063 | /* Guard against wrapping around to address 0. */ |
2064 | if (address < PAGE_ALIGN(address+4)) | |
2065 | address = PAGE_ALIGN(address+4); | |
2066 | else | |
2067 | return -ENOMEM; | |
2068 | ||
2069 | /* We must make sure the anon_vma is allocated. */ | |
1da177e4 LT |
2070 | if (unlikely(anon_vma_prepare(vma))) |
2071 | return -ENOMEM; | |
1da177e4 LT |
2072 | |
2073 | /* | |
2074 | * vma->vm_start/vm_end cannot change under us because the caller | |
2075 | * is required to hold the mmap_sem in read mode. We need the | |
2076 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2077 | */ | |
12352d3c | 2078 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2079 | |
2080 | /* Somebody else might have raced and expanded it already */ | |
2081 | if (address > vma->vm_end) { | |
2082 | unsigned long size, grow; | |
2083 | ||
2084 | size = address - vma->vm_start; | |
2085 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
2086 | ||
42c36f63 HD |
2087 | error = -ENOMEM; |
2088 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
2089 | error = acct_stack_growth(vma, size, grow); | |
2090 | if (!error) { | |
4128997b ML |
2091 | /* |
2092 | * vma_gap_update() doesn't support concurrent | |
2093 | * updates, but we only hold a shared mmap_sem | |
2094 | * lock here, so we need to protect against | |
2095 | * concurrent vma expansions. | |
12352d3c | 2096 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2097 | * we don't guarantee that all growable vmas |
2098 | * in a mm share the same root anon vma. | |
2099 | * So, we reuse mm->page_table_lock to guard | |
2100 | * against concurrent vma expansions. | |
2101 | */ | |
09357814 | 2102 | spin_lock(&mm->page_table_lock); |
87e8827b | 2103 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2104 | mm->locked_vm += grow; |
84638335 | 2105 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2106 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 2107 | vma->vm_end = address; |
bf181b9f | 2108 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 ML |
2109 | if (vma->vm_next) |
2110 | vma_gap_update(vma->vm_next); | |
2111 | else | |
09357814 ON |
2112 | mm->highest_vm_end = address; |
2113 | spin_unlock(&mm->page_table_lock); | |
4128997b | 2114 | |
42c36f63 HD |
2115 | perf_event_mmap(vma); |
2116 | } | |
3af9e859 | 2117 | } |
1da177e4 | 2118 | } |
12352d3c | 2119 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2120 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2121 | validate_mm(mm); |
1da177e4 LT |
2122 | return error; |
2123 | } | |
46dea3d0 HD |
2124 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2125 | ||
1da177e4 LT |
2126 | /* |
2127 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
2128 | */ | |
d05f3169 | 2129 | int expand_downwards(struct vm_area_struct *vma, |
b6a2fea3 | 2130 | unsigned long address) |
1da177e4 | 2131 | { |
09357814 | 2132 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 LT |
2133 | int error; |
2134 | ||
8869477a | 2135 | address &= PAGE_MASK; |
e5467859 | 2136 | error = security_mmap_addr(address); |
8869477a EP |
2137 | if (error) |
2138 | return error; | |
2139 | ||
12352d3c KK |
2140 | /* We must make sure the anon_vma is allocated. */ |
2141 | if (unlikely(anon_vma_prepare(vma))) | |
2142 | return -ENOMEM; | |
1da177e4 LT |
2143 | |
2144 | /* | |
2145 | * vma->vm_start/vm_end cannot change under us because the caller | |
2146 | * is required to hold the mmap_sem in read mode. We need the | |
2147 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2148 | */ | |
12352d3c | 2149 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2150 | |
2151 | /* Somebody else might have raced and expanded it already */ | |
2152 | if (address < vma->vm_start) { | |
2153 | unsigned long size, grow; | |
2154 | ||
2155 | size = vma->vm_end - address; | |
2156 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2157 | ||
a626ca6a LT |
2158 | error = -ENOMEM; |
2159 | if (grow <= vma->vm_pgoff) { | |
2160 | error = acct_stack_growth(vma, size, grow); | |
2161 | if (!error) { | |
4128997b ML |
2162 | /* |
2163 | * vma_gap_update() doesn't support concurrent | |
2164 | * updates, but we only hold a shared mmap_sem | |
2165 | * lock here, so we need to protect against | |
2166 | * concurrent vma expansions. | |
12352d3c | 2167 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2168 | * we don't guarantee that all growable vmas |
2169 | * in a mm share the same root anon vma. | |
2170 | * So, we reuse mm->page_table_lock to guard | |
2171 | * against concurrent vma expansions. | |
2172 | */ | |
09357814 | 2173 | spin_lock(&mm->page_table_lock); |
87e8827b | 2174 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2175 | mm->locked_vm += grow; |
84638335 | 2176 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2177 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2178 | vma->vm_start = address; |
2179 | vma->vm_pgoff -= grow; | |
bf181b9f | 2180 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 | 2181 | vma_gap_update(vma); |
09357814 | 2182 | spin_unlock(&mm->page_table_lock); |
4128997b | 2183 | |
a626ca6a LT |
2184 | perf_event_mmap(vma); |
2185 | } | |
1da177e4 LT |
2186 | } |
2187 | } | |
12352d3c | 2188 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2189 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2190 | validate_mm(mm); |
1da177e4 LT |
2191 | return error; |
2192 | } | |
2193 | ||
09884964 LT |
2194 | /* |
2195 | * Note how expand_stack() refuses to expand the stack all the way to | |
2196 | * abut the next virtual mapping, *unless* that mapping itself is also | |
2197 | * a stack mapping. We want to leave room for a guard page, after all | |
2198 | * (the guard page itself is not added here, that is done by the | |
2199 | * actual page faulting logic) | |
2200 | * | |
2201 | * This matches the behavior of the guard page logic (see mm/memory.c: | |
2202 | * check_stack_guard_page()), which only allows the guard page to be | |
2203 | * removed under these circumstances. | |
2204 | */ | |
b6a2fea3 OW |
2205 | #ifdef CONFIG_STACK_GROWSUP |
2206 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2207 | { | |
09884964 LT |
2208 | struct vm_area_struct *next; |
2209 | ||
2210 | address &= PAGE_MASK; | |
2211 | next = vma->vm_next; | |
2212 | if (next && next->vm_start == address + PAGE_SIZE) { | |
2213 | if (!(next->vm_flags & VM_GROWSUP)) | |
2214 | return -ENOMEM; | |
2215 | } | |
b6a2fea3 OW |
2216 | return expand_upwards(vma, address); |
2217 | } | |
2218 | ||
2219 | struct vm_area_struct * | |
2220 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2221 | { | |
2222 | struct vm_area_struct *vma, *prev; | |
2223 | ||
2224 | addr &= PAGE_MASK; | |
2225 | vma = find_vma_prev(mm, addr, &prev); | |
2226 | if (vma && (vma->vm_start <= addr)) | |
2227 | return vma; | |
1c127185 | 2228 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2229 | return NULL; |
cea10a19 | 2230 | if (prev->vm_flags & VM_LOCKED) |
fc05f566 | 2231 | populate_vma_page_range(prev, addr, prev->vm_end, NULL); |
b6a2fea3 OW |
2232 | return prev; |
2233 | } | |
2234 | #else | |
2235 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2236 | { | |
09884964 LT |
2237 | struct vm_area_struct *prev; |
2238 | ||
2239 | address &= PAGE_MASK; | |
2240 | prev = vma->vm_prev; | |
2241 | if (prev && prev->vm_end == address) { | |
2242 | if (!(prev->vm_flags & VM_GROWSDOWN)) | |
2243 | return -ENOMEM; | |
2244 | } | |
b6a2fea3 OW |
2245 | return expand_downwards(vma, address); |
2246 | } | |
2247 | ||
1da177e4 | 2248 | struct vm_area_struct * |
cc71aba3 | 2249 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2250 | { |
cc71aba3 | 2251 | struct vm_area_struct *vma; |
1da177e4 LT |
2252 | unsigned long start; |
2253 | ||
2254 | addr &= PAGE_MASK; | |
cc71aba3 | 2255 | vma = find_vma(mm, addr); |
1da177e4 LT |
2256 | if (!vma) |
2257 | return NULL; | |
2258 | if (vma->vm_start <= addr) | |
2259 | return vma; | |
2260 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2261 | return NULL; | |
2262 | start = vma->vm_start; | |
2263 | if (expand_stack(vma, addr)) | |
2264 | return NULL; | |
cea10a19 | 2265 | if (vma->vm_flags & VM_LOCKED) |
fc05f566 | 2266 | populate_vma_page_range(vma, addr, start, NULL); |
1da177e4 LT |
2267 | return vma; |
2268 | } | |
2269 | #endif | |
2270 | ||
e1d6d01a JB |
2271 | EXPORT_SYMBOL_GPL(find_extend_vma); |
2272 | ||
1da177e4 | 2273 | /* |
2c0b3814 | 2274 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 2275 | * so release them, and do the vma updates. |
2c0b3814 HD |
2276 | * |
2277 | * Called with the mm semaphore held. | |
1da177e4 | 2278 | */ |
2c0b3814 | 2279 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2280 | { |
4f74d2c8 LT |
2281 | unsigned long nr_accounted = 0; |
2282 | ||
365e9c87 HD |
2283 | /* Update high watermark before we lower total_vm */ |
2284 | update_hiwater_vm(mm); | |
1da177e4 | 2285 | do { |
2c0b3814 HD |
2286 | long nrpages = vma_pages(vma); |
2287 | ||
4f74d2c8 LT |
2288 | if (vma->vm_flags & VM_ACCOUNT) |
2289 | nr_accounted += nrpages; | |
84638335 | 2290 | vm_stat_account(mm, vma->vm_flags, -nrpages); |
a8fb5618 | 2291 | vma = remove_vma(vma); |
146425a3 | 2292 | } while (vma); |
4f74d2c8 | 2293 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2294 | validate_mm(mm); |
2295 | } | |
2296 | ||
2297 | /* | |
2298 | * Get rid of page table information in the indicated region. | |
2299 | * | |
f10df686 | 2300 | * Called with the mm semaphore held. |
1da177e4 LT |
2301 | */ |
2302 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
2303 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
2304 | unsigned long start, unsigned long end) | |
1da177e4 | 2305 | { |
cc71aba3 | 2306 | struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; |
d16dfc55 | 2307 | struct mmu_gather tlb; |
1da177e4 LT |
2308 | |
2309 | lru_add_drain(); | |
2b047252 | 2310 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 2311 | update_hiwater_rss(mm); |
4f74d2c8 | 2312 | unmap_vmas(&tlb, vma, start, end); |
d16dfc55 | 2313 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
6ee8630e | 2314 | next ? next->vm_start : USER_PGTABLES_CEILING); |
d16dfc55 | 2315 | tlb_finish_mmu(&tlb, start, end); |
1da177e4 LT |
2316 | } |
2317 | ||
2318 | /* | |
2319 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
2320 | * vma list as we go.. | |
2321 | */ | |
2322 | static void | |
2323 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
2324 | struct vm_area_struct *prev, unsigned long end) | |
2325 | { | |
2326 | struct vm_area_struct **insertion_point; | |
2327 | struct vm_area_struct *tail_vma = NULL; | |
2328 | ||
2329 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
297c5eee | 2330 | vma->vm_prev = NULL; |
1da177e4 | 2331 | do { |
d3737187 | 2332 | vma_rb_erase(vma, &mm->mm_rb); |
1da177e4 LT |
2333 | mm->map_count--; |
2334 | tail_vma = vma; | |
2335 | vma = vma->vm_next; | |
2336 | } while (vma && vma->vm_start < end); | |
2337 | *insertion_point = vma; | |
d3737187 | 2338 | if (vma) { |
297c5eee | 2339 | vma->vm_prev = prev; |
d3737187 ML |
2340 | vma_gap_update(vma); |
2341 | } else | |
2342 | mm->highest_vm_end = prev ? prev->vm_end : 0; | |
1da177e4 | 2343 | tail_vma->vm_next = NULL; |
615d6e87 DB |
2344 | |
2345 | /* Kill the cache */ | |
2346 | vmacache_invalidate(mm); | |
1da177e4 LT |
2347 | } |
2348 | ||
2349 | /* | |
659ace58 KM |
2350 | * __split_vma() bypasses sysctl_max_map_count checking. We use this on the |
2351 | * munmap path where it doesn't make sense to fail. | |
1da177e4 | 2352 | */ |
cc71aba3 | 2353 | static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
2354 | unsigned long addr, int new_below) |
2355 | { | |
1da177e4 | 2356 | struct vm_area_struct *new; |
e3975891 | 2357 | int err; |
1da177e4 | 2358 | |
a5516438 AK |
2359 | if (is_vm_hugetlb_page(vma) && (addr & |
2360 | ~(huge_page_mask(hstate_vma(vma))))) | |
1da177e4 LT |
2361 | return -EINVAL; |
2362 | ||
e94b1766 | 2363 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 | 2364 | if (!new) |
e3975891 | 2365 | return -ENOMEM; |
1da177e4 LT |
2366 | |
2367 | /* most fields are the same, copy all, and then fixup */ | |
2368 | *new = *vma; | |
2369 | ||
5beb4930 RR |
2370 | INIT_LIST_HEAD(&new->anon_vma_chain); |
2371 | ||
1da177e4 LT |
2372 | if (new_below) |
2373 | new->vm_end = addr; | |
2374 | else { | |
2375 | new->vm_start = addr; | |
2376 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2377 | } | |
2378 | ||
ef0855d3 ON |
2379 | err = vma_dup_policy(vma, new); |
2380 | if (err) | |
5beb4930 | 2381 | goto out_free_vma; |
1da177e4 | 2382 | |
c4ea95d7 DF |
2383 | err = anon_vma_clone(new, vma); |
2384 | if (err) | |
5beb4930 RR |
2385 | goto out_free_mpol; |
2386 | ||
e9714acf | 2387 | if (new->vm_file) |
1da177e4 LT |
2388 | get_file(new->vm_file); |
2389 | ||
2390 | if (new->vm_ops && new->vm_ops->open) | |
2391 | new->vm_ops->open(new); | |
2392 | ||
2393 | if (new_below) | |
5beb4930 | 2394 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
1da177e4 LT |
2395 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
2396 | else | |
5beb4930 | 2397 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
1da177e4 | 2398 | |
5beb4930 RR |
2399 | /* Success. */ |
2400 | if (!err) | |
2401 | return 0; | |
2402 | ||
2403 | /* Clean everything up if vma_adjust failed. */ | |
58927533 RR |
2404 | if (new->vm_ops && new->vm_ops->close) |
2405 | new->vm_ops->close(new); | |
e9714acf | 2406 | if (new->vm_file) |
5beb4930 | 2407 | fput(new->vm_file); |
2aeadc30 | 2408 | unlink_anon_vmas(new); |
5beb4930 | 2409 | out_free_mpol: |
ef0855d3 | 2410 | mpol_put(vma_policy(new)); |
5beb4930 RR |
2411 | out_free_vma: |
2412 | kmem_cache_free(vm_area_cachep, new); | |
5beb4930 | 2413 | return err; |
1da177e4 LT |
2414 | } |
2415 | ||
659ace58 KM |
2416 | /* |
2417 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2418 | * either for the first part or the tail. | |
2419 | */ | |
2420 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, | |
2421 | unsigned long addr, int new_below) | |
2422 | { | |
2423 | if (mm->map_count >= sysctl_max_map_count) | |
2424 | return -ENOMEM; | |
2425 | ||
2426 | return __split_vma(mm, vma, addr, new_below); | |
2427 | } | |
2428 | ||
1da177e4 LT |
2429 | /* Munmap is split into 2 main parts -- this part which finds |
2430 | * what needs doing, and the areas themselves, which do the | |
2431 | * work. This now handles partial unmappings. | |
2432 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
2433 | */ | |
2434 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
2435 | { | |
2436 | unsigned long end; | |
146425a3 | 2437 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 | 2438 | |
de1741a1 | 2439 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) |
1da177e4 LT |
2440 | return -EINVAL; |
2441 | ||
cc71aba3 | 2442 | len = PAGE_ALIGN(len); |
2443 | if (len == 0) | |
1da177e4 LT |
2444 | return -EINVAL; |
2445 | ||
2446 | /* Find the first overlapping VMA */ | |
9be34c9d | 2447 | vma = find_vma(mm, start); |
146425a3 | 2448 | if (!vma) |
1da177e4 | 2449 | return 0; |
9be34c9d | 2450 | prev = vma->vm_prev; |
146425a3 | 2451 | /* we have start < vma->vm_end */ |
1da177e4 LT |
2452 | |
2453 | /* if it doesn't overlap, we have nothing.. */ | |
2454 | end = start + len; | |
146425a3 | 2455 | if (vma->vm_start >= end) |
1da177e4 LT |
2456 | return 0; |
2457 | ||
2458 | /* | |
2459 | * If we need to split any vma, do it now to save pain later. | |
2460 | * | |
2461 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2462 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2463 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2464 | */ | |
146425a3 | 2465 | if (start > vma->vm_start) { |
659ace58 KM |
2466 | int error; |
2467 | ||
2468 | /* | |
2469 | * Make sure that map_count on return from munmap() will | |
2470 | * not exceed its limit; but let map_count go just above | |
2471 | * its limit temporarily, to help free resources as expected. | |
2472 | */ | |
2473 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
2474 | return -ENOMEM; | |
2475 | ||
2476 | error = __split_vma(mm, vma, start, 0); | |
1da177e4 LT |
2477 | if (error) |
2478 | return error; | |
146425a3 | 2479 | prev = vma; |
1da177e4 LT |
2480 | } |
2481 | ||
2482 | /* Does it split the last one? */ | |
2483 | last = find_vma(mm, end); | |
2484 | if (last && end > last->vm_start) { | |
659ace58 | 2485 | int error = __split_vma(mm, last, end, 1); |
1da177e4 LT |
2486 | if (error) |
2487 | return error; | |
2488 | } | |
cc71aba3 | 2489 | vma = prev ? prev->vm_next : mm->mmap; |
1da177e4 | 2490 | |
ba470de4 RR |
2491 | /* |
2492 | * unlock any mlock()ed ranges before detaching vmas | |
2493 | */ | |
2494 | if (mm->locked_vm) { | |
2495 | struct vm_area_struct *tmp = vma; | |
2496 | while (tmp && tmp->vm_start < end) { | |
2497 | if (tmp->vm_flags & VM_LOCKED) { | |
2498 | mm->locked_vm -= vma_pages(tmp); | |
2499 | munlock_vma_pages_all(tmp); | |
2500 | } | |
2501 | tmp = tmp->vm_next; | |
2502 | } | |
2503 | } | |
2504 | ||
1da177e4 LT |
2505 | /* |
2506 | * Remove the vma's, and unmap the actual pages | |
2507 | */ | |
146425a3 HD |
2508 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
2509 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 | 2510 | |
1de4fa14 DH |
2511 | arch_unmap(mm, vma, start, end); |
2512 | ||
1da177e4 | 2513 | /* Fix up all other VM information */ |
2c0b3814 | 2514 | remove_vma_list(mm, vma); |
1da177e4 LT |
2515 | |
2516 | return 0; | |
2517 | } | |
1da177e4 | 2518 | |
bfce281c | 2519 | int vm_munmap(unsigned long start, size_t len) |
1da177e4 LT |
2520 | { |
2521 | int ret; | |
bfce281c | 2522 | struct mm_struct *mm = current->mm; |
1da177e4 | 2523 | |
ae798783 MH |
2524 | if (down_write_killable(&mm->mmap_sem)) |
2525 | return -EINTR; | |
2526 | ||
a46ef99d | 2527 | ret = do_munmap(mm, start, len); |
1da177e4 LT |
2528 | up_write(&mm->mmap_sem); |
2529 | return ret; | |
2530 | } | |
a46ef99d LT |
2531 | EXPORT_SYMBOL(vm_munmap); |
2532 | ||
2533 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2534 | { | |
dc0ef0df MH |
2535 | int ret; |
2536 | struct mm_struct *mm = current->mm; | |
2537 | ||
a46ef99d | 2538 | profile_munmap(addr); |
dc0ef0df MH |
2539 | if (down_write_killable(&mm->mmap_sem)) |
2540 | return -EINTR; | |
2541 | ret = do_munmap(mm, addr, len); | |
2542 | up_write(&mm->mmap_sem); | |
2543 | return ret; | |
a46ef99d | 2544 | } |
1da177e4 | 2545 | |
c8d78c18 KS |
2546 | |
2547 | /* | |
2548 | * Emulation of deprecated remap_file_pages() syscall. | |
2549 | */ | |
2550 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, | |
2551 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) | |
2552 | { | |
2553 | ||
2554 | struct mm_struct *mm = current->mm; | |
2555 | struct vm_area_struct *vma; | |
2556 | unsigned long populate = 0; | |
2557 | unsigned long ret = -EINVAL; | |
2558 | struct file *file; | |
2559 | ||
756a025f JP |
2560 | pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n", |
2561 | current->comm, current->pid); | |
c8d78c18 KS |
2562 | |
2563 | if (prot) | |
2564 | return ret; | |
2565 | start = start & PAGE_MASK; | |
2566 | size = size & PAGE_MASK; | |
2567 | ||
2568 | if (start + size <= start) | |
2569 | return ret; | |
2570 | ||
2571 | /* Does pgoff wrap? */ | |
2572 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) | |
2573 | return ret; | |
2574 | ||
dc0ef0df MH |
2575 | if (down_write_killable(&mm->mmap_sem)) |
2576 | return -EINTR; | |
2577 | ||
c8d78c18 KS |
2578 | vma = find_vma(mm, start); |
2579 | ||
2580 | if (!vma || !(vma->vm_flags & VM_SHARED)) | |
2581 | goto out; | |
2582 | ||
48f7df32 | 2583 | if (start < vma->vm_start) |
c8d78c18 KS |
2584 | goto out; |
2585 | ||
48f7df32 KS |
2586 | if (start + size > vma->vm_end) { |
2587 | struct vm_area_struct *next; | |
2588 | ||
2589 | for (next = vma->vm_next; next; next = next->vm_next) { | |
2590 | /* hole between vmas ? */ | |
2591 | if (next->vm_start != next->vm_prev->vm_end) | |
2592 | goto out; | |
2593 | ||
2594 | if (next->vm_file != vma->vm_file) | |
2595 | goto out; | |
2596 | ||
2597 | if (next->vm_flags != vma->vm_flags) | |
2598 | goto out; | |
2599 | ||
2600 | if (start + size <= next->vm_end) | |
2601 | break; | |
2602 | } | |
2603 | ||
2604 | if (!next) | |
2605 | goto out; | |
c8d78c18 KS |
2606 | } |
2607 | ||
2608 | prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; | |
2609 | prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; | |
2610 | prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; | |
2611 | ||
2612 | flags &= MAP_NONBLOCK; | |
2613 | flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; | |
2614 | if (vma->vm_flags & VM_LOCKED) { | |
48f7df32 | 2615 | struct vm_area_struct *tmp; |
c8d78c18 | 2616 | flags |= MAP_LOCKED; |
48f7df32 | 2617 | |
c8d78c18 | 2618 | /* drop PG_Mlocked flag for over-mapped range */ |
48f7df32 KS |
2619 | for (tmp = vma; tmp->vm_start >= start + size; |
2620 | tmp = tmp->vm_next) { | |
9a73f61b KS |
2621 | /* |
2622 | * Split pmd and munlock page on the border | |
2623 | * of the range. | |
2624 | */ | |
2625 | vma_adjust_trans_huge(tmp, start, start + size, 0); | |
2626 | ||
48f7df32 KS |
2627 | munlock_vma_pages_range(tmp, |
2628 | max(tmp->vm_start, start), | |
2629 | min(tmp->vm_end, start + size)); | |
2630 | } | |
c8d78c18 KS |
2631 | } |
2632 | ||
2633 | file = get_file(vma->vm_file); | |
2634 | ret = do_mmap_pgoff(vma->vm_file, start, size, | |
2635 | prot, flags, pgoff, &populate); | |
2636 | fput(file); | |
2637 | out: | |
2638 | up_write(&mm->mmap_sem); | |
2639 | if (populate) | |
2640 | mm_populate(ret, populate); | |
2641 | if (!IS_ERR_VALUE(ret)) | |
2642 | ret = 0; | |
2643 | return ret; | |
2644 | } | |
2645 | ||
1da177e4 LT |
2646 | static inline void verify_mm_writelocked(struct mm_struct *mm) |
2647 | { | |
a241ec65 | 2648 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
2649 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
2650 | WARN_ON(1); | |
2651 | up_read(&mm->mmap_sem); | |
2652 | } | |
2653 | #endif | |
2654 | } | |
2655 | ||
2656 | /* | |
2657 | * this is really a simplified "do_mmap". it only handles | |
2658 | * anonymous maps. eventually we may be able to do some | |
2659 | * brk-specific accounting here. | |
2660 | */ | |
ba093a6d | 2661 | static int do_brk(unsigned long addr, unsigned long request) |
1da177e4 | 2662 | { |
cc71aba3 | 2663 | struct mm_struct *mm = current->mm; |
2664 | struct vm_area_struct *vma, *prev; | |
ba093a6d | 2665 | unsigned long flags, len; |
cc71aba3 | 2666 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 2667 | pgoff_t pgoff = addr >> PAGE_SHIFT; |
3a459756 | 2668 | int error; |
1da177e4 | 2669 | |
ba093a6d KC |
2670 | len = PAGE_ALIGN(request); |
2671 | if (len < request) | |
2672 | return -ENOMEM; | |
1da177e4 | 2673 | if (!len) |
5d22fc25 | 2674 | return 0; |
1da177e4 | 2675 | |
3a459756 KK |
2676 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; |
2677 | ||
2c6a1016 | 2678 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
de1741a1 | 2679 | if (offset_in_page(error)) |
3a459756 KK |
2680 | return error; |
2681 | ||
363ee17f DB |
2682 | error = mlock_future_check(mm, mm->def_flags, len); |
2683 | if (error) | |
2684 | return error; | |
1da177e4 LT |
2685 | |
2686 | /* | |
2687 | * mm->mmap_sem is required to protect against another thread | |
2688 | * changing the mappings in case we sleep. | |
2689 | */ | |
2690 | verify_mm_writelocked(mm); | |
2691 | ||
2692 | /* | |
2693 | * Clear old maps. this also does some error checking for us | |
2694 | */ | |
9fcd1457 RV |
2695 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
2696 | &rb_parent)) { | |
1da177e4 LT |
2697 | if (do_munmap(mm, addr, len)) |
2698 | return -ENOMEM; | |
1da177e4 LT |
2699 | } |
2700 | ||
2701 | /* Check against address space limits *after* clearing old maps... */ | |
84638335 | 2702 | if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) |
1da177e4 LT |
2703 | return -ENOMEM; |
2704 | ||
2705 | if (mm->map_count > sysctl_max_map_count) | |
2706 | return -ENOMEM; | |
2707 | ||
191c5424 | 2708 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2709 | return -ENOMEM; |
2710 | ||
1da177e4 | 2711 | /* Can we just expand an old private anonymous mapping? */ |
ba470de4 | 2712 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
19a809af | 2713 | NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); |
ba470de4 | 2714 | if (vma) |
1da177e4 LT |
2715 | goto out; |
2716 | ||
2717 | /* | |
2718 | * create a vma struct for an anonymous mapping | |
2719 | */ | |
c5e3b83e | 2720 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2721 | if (!vma) { |
2722 | vm_unacct_memory(len >> PAGE_SHIFT); | |
2723 | return -ENOMEM; | |
2724 | } | |
1da177e4 | 2725 | |
5beb4930 | 2726 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
2727 | vma->vm_mm = mm; |
2728 | vma->vm_start = addr; | |
2729 | vma->vm_end = addr + len; | |
2730 | vma->vm_pgoff = pgoff; | |
2731 | vma->vm_flags = flags; | |
3ed75eb8 | 2732 | vma->vm_page_prot = vm_get_page_prot(flags); |
1da177e4 LT |
2733 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2734 | out: | |
3af9e859 | 2735 | perf_event_mmap(vma); |
1da177e4 | 2736 | mm->total_vm += len >> PAGE_SHIFT; |
84638335 | 2737 | mm->data_vm += len >> PAGE_SHIFT; |
128557ff ML |
2738 | if (flags & VM_LOCKED) |
2739 | mm->locked_vm += (len >> PAGE_SHIFT); | |
d9104d1c | 2740 | vma->vm_flags |= VM_SOFTDIRTY; |
5d22fc25 | 2741 | return 0; |
1da177e4 LT |
2742 | } |
2743 | ||
5d22fc25 | 2744 | int vm_brk(unsigned long addr, unsigned long len) |
e4eb1ff6 LT |
2745 | { |
2746 | struct mm_struct *mm = current->mm; | |
5d22fc25 | 2747 | int ret; |
128557ff | 2748 | bool populate; |
e4eb1ff6 | 2749 | |
2d6c9282 MH |
2750 | if (down_write_killable(&mm->mmap_sem)) |
2751 | return -EINTR; | |
2752 | ||
e4eb1ff6 | 2753 | ret = do_brk(addr, len); |
128557ff | 2754 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
e4eb1ff6 | 2755 | up_write(&mm->mmap_sem); |
5d22fc25 | 2756 | if (populate && !ret) |
128557ff | 2757 | mm_populate(addr, len); |
e4eb1ff6 LT |
2758 | return ret; |
2759 | } | |
2760 | EXPORT_SYMBOL(vm_brk); | |
1da177e4 LT |
2761 | |
2762 | /* Release all mmaps. */ | |
2763 | void exit_mmap(struct mm_struct *mm) | |
2764 | { | |
d16dfc55 | 2765 | struct mmu_gather tlb; |
ba470de4 | 2766 | struct vm_area_struct *vma; |
1da177e4 LT |
2767 | unsigned long nr_accounted = 0; |
2768 | ||
d6dd61c8 | 2769 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 2770 | mmu_notifier_release(mm); |
d6dd61c8 | 2771 | |
ba470de4 RR |
2772 | if (mm->locked_vm) { |
2773 | vma = mm->mmap; | |
2774 | while (vma) { | |
2775 | if (vma->vm_flags & VM_LOCKED) | |
2776 | munlock_vma_pages_all(vma); | |
2777 | vma = vma->vm_next; | |
2778 | } | |
2779 | } | |
9480c53e JF |
2780 | |
2781 | arch_exit_mmap(mm); | |
2782 | ||
ba470de4 | 2783 | vma = mm->mmap; |
9480c53e JF |
2784 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
2785 | return; | |
2786 | ||
1da177e4 | 2787 | lru_add_drain(); |
1da177e4 | 2788 | flush_cache_mm(mm); |
2b047252 | 2789 | tlb_gather_mmu(&tlb, mm, 0, -1); |
901608d9 | 2790 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
e0da382c | 2791 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
4f74d2c8 | 2792 | unmap_vmas(&tlb, vma, 0, -1); |
9ba69294 | 2793 | |
6ee8630e | 2794 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); |
853f5e26 | 2795 | tlb_finish_mmu(&tlb, 0, -1); |
1da177e4 | 2796 | |
1da177e4 | 2797 | /* |
8f4f8c16 HD |
2798 | * Walk the list again, actually closing and freeing it, |
2799 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 2800 | */ |
4f74d2c8 LT |
2801 | while (vma) { |
2802 | if (vma->vm_flags & VM_ACCOUNT) | |
2803 | nr_accounted += vma_pages(vma); | |
a8fb5618 | 2804 | vma = remove_vma(vma); |
4f74d2c8 LT |
2805 | } |
2806 | vm_unacct_memory(nr_accounted); | |
1da177e4 LT |
2807 | } |
2808 | ||
2809 | /* Insert vm structure into process list sorted by address | |
2810 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
c8c06efa | 2811 | * then i_mmap_rwsem is taken here. |
1da177e4 | 2812 | */ |
6597d783 | 2813 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2814 | { |
6597d783 HD |
2815 | struct vm_area_struct *prev; |
2816 | struct rb_node **rb_link, *rb_parent; | |
1da177e4 | 2817 | |
c9d13f5f CG |
2818 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
2819 | &prev, &rb_link, &rb_parent)) | |
2820 | return -ENOMEM; | |
2821 | if ((vma->vm_flags & VM_ACCOUNT) && | |
2822 | security_vm_enough_memory_mm(mm, vma_pages(vma))) | |
2823 | return -ENOMEM; | |
2824 | ||
1da177e4 LT |
2825 | /* |
2826 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
2827 | * until its first write fault, when page's anon_vma and index | |
2828 | * are set. But now set the vm_pgoff it will almost certainly | |
2829 | * end up with (unless mremap moves it elsewhere before that | |
2830 | * first wfault), so /proc/pid/maps tells a consistent story. | |
2831 | * | |
2832 | * By setting it to reflect the virtual start address of the | |
2833 | * vma, merges and splits can happen in a seamless way, just | |
2834 | * using the existing file pgoff checks and manipulations. | |
2835 | * Similarly in do_mmap_pgoff and in do_brk. | |
2836 | */ | |
8a9cc3b5 | 2837 | if (vma_is_anonymous(vma)) { |
1da177e4 LT |
2838 | BUG_ON(vma->anon_vma); |
2839 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
2840 | } | |
2b144498 | 2841 | |
1da177e4 LT |
2842 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2843 | return 0; | |
2844 | } | |
2845 | ||
2846 | /* | |
2847 | * Copy the vma structure to a new location in the same mm, | |
2848 | * prior to moving page table entries, to effect an mremap move. | |
2849 | */ | |
2850 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
2851 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
2852 | bool *need_rmap_locks) | |
1da177e4 LT |
2853 | { |
2854 | struct vm_area_struct *vma = *vmap; | |
2855 | unsigned long vma_start = vma->vm_start; | |
2856 | struct mm_struct *mm = vma->vm_mm; | |
2857 | struct vm_area_struct *new_vma, *prev; | |
2858 | struct rb_node **rb_link, *rb_parent; | |
948f017b | 2859 | bool faulted_in_anon_vma = true; |
1da177e4 LT |
2860 | |
2861 | /* | |
2862 | * If anonymous vma has not yet been faulted, update new pgoff | |
2863 | * to match new location, to increase its chance of merging. | |
2864 | */ | |
ce75799b | 2865 | if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { |
1da177e4 | 2866 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
2867 | faulted_in_anon_vma = false; |
2868 | } | |
1da177e4 | 2869 | |
6597d783 HD |
2870 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
2871 | return NULL; /* should never get here */ | |
1da177e4 | 2872 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
19a809af AA |
2873 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
2874 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
2875 | if (new_vma) { |
2876 | /* | |
2877 | * Source vma may have been merged into new_vma | |
2878 | */ | |
948f017b AA |
2879 | if (unlikely(vma_start >= new_vma->vm_start && |
2880 | vma_start < new_vma->vm_end)) { | |
2881 | /* | |
2882 | * The only way we can get a vma_merge with | |
2883 | * self during an mremap is if the vma hasn't | |
2884 | * been faulted in yet and we were allowed to | |
2885 | * reset the dst vma->vm_pgoff to the | |
2886 | * destination address of the mremap to allow | |
2887 | * the merge to happen. mremap must change the | |
2888 | * vm_pgoff linearity between src and dst vmas | |
2889 | * (in turn preventing a vma_merge) to be | |
2890 | * safe. It is only safe to keep the vm_pgoff | |
2891 | * linear if there are no pages mapped yet. | |
2892 | */ | |
81d1b09c | 2893 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
38a76013 | 2894 | *vmap = vma = new_vma; |
108d6642 | 2895 | } |
38a76013 | 2896 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 2897 | } else { |
e94b1766 | 2898 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
e3975891 CG |
2899 | if (!new_vma) |
2900 | goto out; | |
2901 | *new_vma = *vma; | |
2902 | new_vma->vm_start = addr; | |
2903 | new_vma->vm_end = addr + len; | |
2904 | new_vma->vm_pgoff = pgoff; | |
2905 | if (vma_dup_policy(vma, new_vma)) | |
2906 | goto out_free_vma; | |
2907 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); | |
2908 | if (anon_vma_clone(new_vma, vma)) | |
2909 | goto out_free_mempol; | |
2910 | if (new_vma->vm_file) | |
2911 | get_file(new_vma->vm_file); | |
2912 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
2913 | new_vma->vm_ops->open(new_vma); | |
2914 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
2915 | *need_rmap_locks = false; | |
1da177e4 LT |
2916 | } |
2917 | return new_vma; | |
5beb4930 | 2918 | |
e3975891 | 2919 | out_free_mempol: |
ef0855d3 | 2920 | mpol_put(vma_policy(new_vma)); |
e3975891 | 2921 | out_free_vma: |
5beb4930 | 2922 | kmem_cache_free(vm_area_cachep, new_vma); |
e3975891 | 2923 | out: |
5beb4930 | 2924 | return NULL; |
1da177e4 | 2925 | } |
119f657c | 2926 | |
2927 | /* | |
2928 | * Return true if the calling process may expand its vm space by the passed | |
2929 | * number of pages | |
2930 | */ | |
84638335 | 2931 | bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) |
119f657c | 2932 | { |
84638335 KK |
2933 | if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) |
2934 | return false; | |
119f657c | 2935 | |
d977d56c KK |
2936 | if (is_data_mapping(flags) && |
2937 | mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { | |
f4fcd558 KK |
2938 | /* Workaround for Valgrind */ |
2939 | if (rlimit(RLIMIT_DATA) == 0 && | |
2940 | mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) | |
2941 | return true; | |
2942 | if (!ignore_rlimit_data) { | |
2943 | pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n", | |
d977d56c KK |
2944 | current->comm, current->pid, |
2945 | (mm->data_vm + npages) << PAGE_SHIFT, | |
2946 | rlimit(RLIMIT_DATA)); | |
d977d56c | 2947 | return false; |
f4fcd558 | 2948 | } |
d977d56c | 2949 | } |
119f657c | 2950 | |
84638335 KK |
2951 | return true; |
2952 | } | |
2953 | ||
2954 | void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) | |
2955 | { | |
2956 | mm->total_vm += npages; | |
2957 | ||
d977d56c | 2958 | if (is_exec_mapping(flags)) |
84638335 | 2959 | mm->exec_vm += npages; |
d977d56c | 2960 | else if (is_stack_mapping(flags)) |
84638335 | 2961 | mm->stack_vm += npages; |
d977d56c | 2962 | else if (is_data_mapping(flags)) |
84638335 | 2963 | mm->data_vm += npages; |
119f657c | 2964 | } |
fa5dc22f | 2965 | |
a62c34bd AL |
2966 | static int special_mapping_fault(struct vm_area_struct *vma, |
2967 | struct vm_fault *vmf); | |
2968 | ||
2969 | /* | |
2970 | * Having a close hook prevents vma merging regardless of flags. | |
2971 | */ | |
2972 | static void special_mapping_close(struct vm_area_struct *vma) | |
2973 | { | |
2974 | } | |
2975 | ||
2976 | static const char *special_mapping_name(struct vm_area_struct *vma) | |
2977 | { | |
2978 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; | |
2979 | } | |
2980 | ||
b059a453 DS |
2981 | static int special_mapping_mremap(struct vm_area_struct *new_vma) |
2982 | { | |
2983 | struct vm_special_mapping *sm = new_vma->vm_private_data; | |
2984 | ||
2985 | if (sm->mremap) | |
2986 | return sm->mremap(sm, new_vma); | |
2987 | return 0; | |
2988 | } | |
2989 | ||
a62c34bd AL |
2990 | static const struct vm_operations_struct special_mapping_vmops = { |
2991 | .close = special_mapping_close, | |
2992 | .fault = special_mapping_fault, | |
b059a453 | 2993 | .mremap = special_mapping_mremap, |
a62c34bd AL |
2994 | .name = special_mapping_name, |
2995 | }; | |
2996 | ||
2997 | static const struct vm_operations_struct legacy_special_mapping_vmops = { | |
2998 | .close = special_mapping_close, | |
2999 | .fault = special_mapping_fault, | |
3000 | }; | |
fa5dc22f | 3001 | |
b1d0e4f5 NP |
3002 | static int special_mapping_fault(struct vm_area_struct *vma, |
3003 | struct vm_fault *vmf) | |
fa5dc22f | 3004 | { |
b1d0e4f5 | 3005 | pgoff_t pgoff; |
fa5dc22f RM |
3006 | struct page **pages; |
3007 | ||
f872f540 | 3008 | if (vma->vm_ops == &legacy_special_mapping_vmops) { |
a62c34bd | 3009 | pages = vma->vm_private_data; |
f872f540 AL |
3010 | } else { |
3011 | struct vm_special_mapping *sm = vma->vm_private_data; | |
3012 | ||
3013 | if (sm->fault) | |
3014 | return sm->fault(sm, vma, vmf); | |
3015 | ||
3016 | pages = sm->pages; | |
3017 | } | |
a62c34bd | 3018 | |
8a9cc3b5 | 3019 | for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) |
b1d0e4f5 | 3020 | pgoff--; |
fa5dc22f RM |
3021 | |
3022 | if (*pages) { | |
3023 | struct page *page = *pages; | |
3024 | get_page(page); | |
b1d0e4f5 NP |
3025 | vmf->page = page; |
3026 | return 0; | |
fa5dc22f RM |
3027 | } |
3028 | ||
b1d0e4f5 | 3029 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
3030 | } |
3031 | ||
a62c34bd AL |
3032 | static struct vm_area_struct *__install_special_mapping( |
3033 | struct mm_struct *mm, | |
3034 | unsigned long addr, unsigned long len, | |
27f28b97 CG |
3035 | unsigned long vm_flags, void *priv, |
3036 | const struct vm_operations_struct *ops) | |
fa5dc22f | 3037 | { |
462e635e | 3038 | int ret; |
fa5dc22f RM |
3039 | struct vm_area_struct *vma; |
3040 | ||
3041 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
3042 | if (unlikely(vma == NULL)) | |
3935ed6a | 3043 | return ERR_PTR(-ENOMEM); |
fa5dc22f | 3044 | |
5beb4930 | 3045 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
fa5dc22f RM |
3046 | vma->vm_mm = mm; |
3047 | vma->vm_start = addr; | |
3048 | vma->vm_end = addr + len; | |
3049 | ||
d9104d1c | 3050 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; |
3ed75eb8 | 3051 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f | 3052 | |
a62c34bd AL |
3053 | vma->vm_ops = ops; |
3054 | vma->vm_private_data = priv; | |
fa5dc22f | 3055 | |
462e635e TO |
3056 | ret = insert_vm_struct(mm, vma); |
3057 | if (ret) | |
3058 | goto out; | |
fa5dc22f | 3059 | |
84638335 | 3060 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); |
fa5dc22f | 3061 | |
cdd6c482 | 3062 | perf_event_mmap(vma); |
089dd79d | 3063 | |
3935ed6a | 3064 | return vma; |
462e635e TO |
3065 | |
3066 | out: | |
3067 | kmem_cache_free(vm_area_cachep, vma); | |
3935ed6a SS |
3068 | return ERR_PTR(ret); |
3069 | } | |
3070 | ||
a62c34bd AL |
3071 | /* |
3072 | * Called with mm->mmap_sem held for writing. | |
3073 | * Insert a new vma covering the given region, with the given flags. | |
3074 | * Its pages are supplied by the given array of struct page *. | |
3075 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
3076 | * The region past the last page supplied will always produce SIGBUS. | |
3077 | * The array pointer and the pages it points to are assumed to stay alive | |
3078 | * for as long as this mapping might exist. | |
3079 | */ | |
3080 | struct vm_area_struct *_install_special_mapping( | |
3081 | struct mm_struct *mm, | |
3082 | unsigned long addr, unsigned long len, | |
3083 | unsigned long vm_flags, const struct vm_special_mapping *spec) | |
3084 | { | |
27f28b97 CG |
3085 | return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, |
3086 | &special_mapping_vmops); | |
a62c34bd AL |
3087 | } |
3088 | ||
3935ed6a SS |
3089 | int install_special_mapping(struct mm_struct *mm, |
3090 | unsigned long addr, unsigned long len, | |
3091 | unsigned long vm_flags, struct page **pages) | |
3092 | { | |
a62c34bd | 3093 | struct vm_area_struct *vma = __install_special_mapping( |
27f28b97 CG |
3094 | mm, addr, len, vm_flags, (void *)pages, |
3095 | &legacy_special_mapping_vmops); | |
3935ed6a | 3096 | |
14bd5b45 | 3097 | return PTR_ERR_OR_ZERO(vma); |
fa5dc22f | 3098 | } |
7906d00c AA |
3099 | |
3100 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3101 | ||
454ed842 | 3102 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3103 | { |
bf181b9f | 3104 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3105 | /* |
3106 | * The LSB of head.next can't change from under us | |
3107 | * because we hold the mm_all_locks_mutex. | |
3108 | */ | |
572043c9 | 3109 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); |
7906d00c AA |
3110 | /* |
3111 | * We can safely modify head.next after taking the | |
5a505085 | 3112 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3113 | * the same anon_vma we won't take it again. |
3114 | * | |
3115 | * No need of atomic instructions here, head.next | |
3116 | * can't change from under us thanks to the | |
5a505085 | 3117 | * anon_vma->root->rwsem. |
7906d00c AA |
3118 | */ |
3119 | if (__test_and_set_bit(0, (unsigned long *) | |
bf181b9f | 3120 | &anon_vma->root->rb_root.rb_node)) |
7906d00c AA |
3121 | BUG(); |
3122 | } | |
3123 | } | |
3124 | ||
454ed842 | 3125 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3126 | { |
3127 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3128 | /* | |
3129 | * AS_MM_ALL_LOCKS can't change from under us because | |
3130 | * we hold the mm_all_locks_mutex. | |
3131 | * | |
3132 | * Operations on ->flags have to be atomic because | |
3133 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3134 | * mm_all_locks_mutex, there may be other cpus | |
3135 | * changing other bitflags in parallel to us. | |
3136 | */ | |
3137 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3138 | BUG(); | |
c8c06efa | 3139 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); |
7906d00c AA |
3140 | } |
3141 | } | |
3142 | ||
3143 | /* | |
3144 | * This operation locks against the VM for all pte/vma/mm related | |
3145 | * operations that could ever happen on a certain mm. This includes | |
3146 | * vmtruncate, try_to_unmap, and all page faults. | |
3147 | * | |
3148 | * The caller must take the mmap_sem in write mode before calling | |
3149 | * mm_take_all_locks(). The caller isn't allowed to release the | |
3150 | * mmap_sem until mm_drop_all_locks() returns. | |
3151 | * | |
3152 | * mmap_sem in write mode is required in order to block all operations | |
3153 | * that could modify pagetables and free pages without need of | |
27ba0644 | 3154 | * altering the vma layout. It's also needed in write mode to avoid new |
7906d00c AA |
3155 | * anon_vmas to be associated with existing vmas. |
3156 | * | |
3157 | * A single task can't take more than one mm_take_all_locks() in a row | |
3158 | * or it would deadlock. | |
3159 | * | |
bf181b9f | 3160 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3161 | * mapping->flags avoid to take the same lock twice, if more than one |
3162 | * vma in this mm is backed by the same anon_vma or address_space. | |
3163 | * | |
88f306b6 KS |
3164 | * We take locks in following order, accordingly to comment at beginning |
3165 | * of mm/rmap.c: | |
3166 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for | |
3167 | * hugetlb mapping); | |
3168 | * - all i_mmap_rwsem locks; | |
3169 | * - all anon_vma->rwseml | |
3170 | * | |
3171 | * We can take all locks within these types randomly because the VM code | |
3172 | * doesn't nest them and we protected from parallel mm_take_all_locks() by | |
3173 | * mm_all_locks_mutex. | |
7906d00c AA |
3174 | * |
3175 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3176 | * that may have to take thousand of locks. | |
3177 | * | |
3178 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3179 | */ | |
3180 | int mm_take_all_locks(struct mm_struct *mm) | |
3181 | { | |
3182 | struct vm_area_struct *vma; | |
5beb4930 | 3183 | struct anon_vma_chain *avc; |
7906d00c AA |
3184 | |
3185 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3186 | ||
3187 | mutex_lock(&mm_all_locks_mutex); | |
3188 | ||
3189 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3190 | if (signal_pending(current)) | |
3191 | goto out_unlock; | |
88f306b6 KS |
3192 | if (vma->vm_file && vma->vm_file->f_mapping && |
3193 | is_vm_hugetlb_page(vma)) | |
3194 | vm_lock_mapping(mm, vma->vm_file->f_mapping); | |
3195 | } | |
3196 | ||
3197 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3198 | if (signal_pending(current)) | |
3199 | goto out_unlock; | |
3200 | if (vma->vm_file && vma->vm_file->f_mapping && | |
3201 | !is_vm_hugetlb_page(vma)) | |
454ed842 | 3202 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3203 | } |
7cd5a02f PZ |
3204 | |
3205 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3206 | if (signal_pending(current)) | |
3207 | goto out_unlock; | |
3208 | if (vma->anon_vma) | |
5beb4930 RR |
3209 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3210 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3211 | } |
7cd5a02f | 3212 | |
584cff54 | 3213 | return 0; |
7906d00c AA |
3214 | |
3215 | out_unlock: | |
584cff54 KC |
3216 | mm_drop_all_locks(mm); |
3217 | return -EINTR; | |
7906d00c AA |
3218 | } |
3219 | ||
3220 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3221 | { | |
bf181b9f | 3222 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3223 | /* |
3224 | * The LSB of head.next can't change to 0 from under | |
3225 | * us because we hold the mm_all_locks_mutex. | |
3226 | * | |
3227 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3228 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3229 | * never see our bitflag. |
3230 | * | |
3231 | * No need of atomic instructions here, head.next | |
3232 | * can't change from under us until we release the | |
5a505085 | 3233 | * anon_vma->root->rwsem. |
7906d00c AA |
3234 | */ |
3235 | if (!__test_and_clear_bit(0, (unsigned long *) | |
bf181b9f | 3236 | &anon_vma->root->rb_root.rb_node)) |
7906d00c | 3237 | BUG(); |
08b52706 | 3238 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3239 | } |
3240 | } | |
3241 | ||
3242 | static void vm_unlock_mapping(struct address_space *mapping) | |
3243 | { | |
3244 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3245 | /* | |
3246 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3247 | * because we hold the mm_all_locks_mutex. | |
3248 | */ | |
83cde9e8 | 3249 | i_mmap_unlock_write(mapping); |
7906d00c AA |
3250 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3251 | &mapping->flags)) | |
3252 | BUG(); | |
3253 | } | |
3254 | } | |
3255 | ||
3256 | /* | |
3257 | * The mmap_sem cannot be released by the caller until | |
3258 | * mm_drop_all_locks() returns. | |
3259 | */ | |
3260 | void mm_drop_all_locks(struct mm_struct *mm) | |
3261 | { | |
3262 | struct vm_area_struct *vma; | |
5beb4930 | 3263 | struct anon_vma_chain *avc; |
7906d00c AA |
3264 | |
3265 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3266 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); | |
3267 | ||
3268 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3269 | if (vma->anon_vma) | |
5beb4930 RR |
3270 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3271 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3272 | if (vma->vm_file && vma->vm_file->f_mapping) |
3273 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3274 | } | |
3275 | ||
3276 | mutex_unlock(&mm_all_locks_mutex); | |
3277 | } | |
8feae131 DH |
3278 | |
3279 | /* | |
3280 | * initialise the VMA slab | |
3281 | */ | |
3282 | void __init mmap_init(void) | |
3283 | { | |
00a62ce9 KM |
3284 | int ret; |
3285 | ||
908c7f19 | 3286 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 3287 | VM_BUG_ON(ret); |
8feae131 | 3288 | } |
c9b1d098 AS |
3289 | |
3290 | /* | |
3291 | * Initialise sysctl_user_reserve_kbytes. | |
3292 | * | |
3293 | * This is intended to prevent a user from starting a single memory hogging | |
3294 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3295 | * mode. | |
3296 | * | |
3297 | * The default value is min(3% of free memory, 128MB) | |
3298 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3299 | */ | |
1640879a | 3300 | static int init_user_reserve(void) |
c9b1d098 AS |
3301 | { |
3302 | unsigned long free_kbytes; | |
3303 | ||
3304 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3305 | ||
3306 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3307 | return 0; | |
3308 | } | |
a64fb3cd | 3309 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
3310 | |
3311 | /* | |
3312 | * Initialise sysctl_admin_reserve_kbytes. | |
3313 | * | |
3314 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3315 | * to log in and kill a memory hogging process. | |
3316 | * | |
3317 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3318 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3319 | * only reserve 3% of free pages by default. | |
3320 | */ | |
1640879a | 3321 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3322 | { |
3323 | unsigned long free_kbytes; | |
3324 | ||
3325 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3326 | ||
3327 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3328 | return 0; | |
3329 | } | |
a64fb3cd | 3330 | subsys_initcall(init_admin_reserve); |
1640879a AS |
3331 | |
3332 | /* | |
3333 | * Reinititalise user and admin reserves if memory is added or removed. | |
3334 | * | |
3335 | * The default user reserve max is 128MB, and the default max for the | |
3336 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3337 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3338 | * and tools like top. It may make sense to increase or even disable the | |
3339 | * reserve depending on the existence of swap or variations in the recovery | |
3340 | * tools. So, the admin may have changed them. | |
3341 | * | |
3342 | * If memory is added and the reserves have been eliminated or increased above | |
3343 | * the default max, then we'll trust the admin. | |
3344 | * | |
3345 | * If memory is removed and there isn't enough free memory, then we | |
3346 | * need to reset the reserves. | |
3347 | * | |
3348 | * Otherwise keep the reserve set by the admin. | |
3349 | */ | |
3350 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3351 | unsigned long action, void *data) | |
3352 | { | |
3353 | unsigned long tmp, free_kbytes; | |
3354 | ||
3355 | switch (action) { | |
3356 | case MEM_ONLINE: | |
3357 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3358 | tmp = sysctl_user_reserve_kbytes; | |
3359 | if (0 < tmp && tmp < (1UL << 17)) | |
3360 | init_user_reserve(); | |
3361 | ||
3362 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3363 | tmp = sysctl_admin_reserve_kbytes; | |
3364 | if (0 < tmp && tmp < (1UL << 13)) | |
3365 | init_admin_reserve(); | |
3366 | ||
3367 | break; | |
3368 | case MEM_OFFLINE: | |
3369 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3370 | ||
3371 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3372 | init_user_reserve(); | |
3373 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3374 | sysctl_user_reserve_kbytes); | |
3375 | } | |
3376 | ||
3377 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3378 | init_admin_reserve(); | |
3379 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3380 | sysctl_admin_reserve_kbytes); | |
3381 | } | |
3382 | break; | |
3383 | default: | |
3384 | break; | |
3385 | } | |
3386 | return NOTIFY_OK; | |
3387 | } | |
3388 | ||
3389 | static struct notifier_block reserve_mem_nb = { | |
3390 | .notifier_call = reserve_mem_notifier, | |
3391 | }; | |
3392 | ||
3393 | static int __meminit init_reserve_notifier(void) | |
3394 | { | |
3395 | if (register_hotmemory_notifier(&reserve_mem_nb)) | |
b1de0d13 | 3396 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
1640879a AS |
3397 | |
3398 | return 0; | |
3399 | } | |
a64fb3cd | 3400 | subsys_initcall(init_reserve_notifier); |