]>
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; | |
6d2329f8 | 119 | pgprot_t vm_page_prot; |
64e45507 | 120 | |
6d2329f8 AA |
121 | vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); |
122 | if (vma_wants_writenotify(vma, vm_page_prot)) { | |
64e45507 | 123 | vm_flags &= ~VM_SHARED; |
6d2329f8 | 124 | vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); |
64e45507 | 125 | } |
6d2329f8 AA |
126 | /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */ |
127 | WRITE_ONCE(vma->vm_page_prot, vm_page_prot); | |
64e45507 PF |
128 | } |
129 | ||
1da177e4 | 130 | /* |
c8c06efa | 131 | * Requires inode->i_mapping->i_mmap_rwsem |
1da177e4 LT |
132 | */ |
133 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
134 | struct file *file, struct address_space *mapping) | |
135 | { | |
136 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 137 | atomic_inc(&file_inode(file)->i_writecount); |
1da177e4 | 138 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 139 | mapping_unmap_writable(mapping); |
1da177e4 LT |
140 | |
141 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 142 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
143 | flush_dcache_mmap_unlock(mapping); |
144 | } | |
145 | ||
146 | /* | |
6b2dbba8 | 147 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 148 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 149 | */ |
a8fb5618 | 150 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
151 | { |
152 | struct file *file = vma->vm_file; | |
153 | ||
1da177e4 LT |
154 | if (file) { |
155 | struct address_space *mapping = file->f_mapping; | |
83cde9e8 | 156 | i_mmap_lock_write(mapping); |
1da177e4 | 157 | __remove_shared_vm_struct(vma, file, mapping); |
83cde9e8 | 158 | i_mmap_unlock_write(mapping); |
1da177e4 | 159 | } |
a8fb5618 HD |
160 | } |
161 | ||
162 | /* | |
163 | * Close a vm structure and free it, returning the next. | |
164 | */ | |
165 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
166 | { | |
167 | struct vm_area_struct *next = vma->vm_next; | |
168 | ||
a8fb5618 | 169 | might_sleep(); |
1da177e4 LT |
170 | if (vma->vm_ops && vma->vm_ops->close) |
171 | vma->vm_ops->close(vma); | |
e9714acf | 172 | if (vma->vm_file) |
a8fb5618 | 173 | fput(vma->vm_file); |
f0be3d32 | 174 | mpol_put(vma_policy(vma)); |
1da177e4 | 175 | kmem_cache_free(vm_area_cachep, vma); |
a8fb5618 | 176 | return next; |
1da177e4 LT |
177 | } |
178 | ||
5d22fc25 | 179 | static int do_brk(unsigned long addr, unsigned long len); |
e4eb1ff6 | 180 | |
6a6160a7 | 181 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 | 182 | { |
8764b338 | 183 | unsigned long retval; |
1da177e4 LT |
184 | unsigned long newbrk, oldbrk; |
185 | struct mm_struct *mm = current->mm; | |
a5b4592c | 186 | unsigned long min_brk; |
128557ff | 187 | bool populate; |
1da177e4 | 188 | |
dc0ef0df MH |
189 | if (down_write_killable(&mm->mmap_sem)) |
190 | return -EINTR; | |
1da177e4 | 191 | |
a5b4592c | 192 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
193 | /* |
194 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
195 | * randomize_va_space to 2, which will still cause mm->start_brk | |
196 | * to be arbitrarily shifted | |
197 | */ | |
4471a675 | 198 | if (current->brk_randomized) |
5520e894 JK |
199 | min_brk = mm->start_brk; |
200 | else | |
201 | min_brk = mm->end_data; | |
a5b4592c JK |
202 | #else |
203 | min_brk = mm->start_brk; | |
204 | #endif | |
205 | if (brk < min_brk) | |
1da177e4 | 206 | goto out; |
1e624196 RG |
207 | |
208 | /* | |
209 | * Check against rlimit here. If this check is done later after the test | |
210 | * of oldbrk with newbrk then it can escape the test and let the data | |
211 | * segment grow beyond its set limit the in case where the limit is | |
212 | * not page aligned -Ram Gupta | |
213 | */ | |
8764b338 CG |
214 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
215 | mm->end_data, mm->start_data)) | |
1e624196 RG |
216 | goto out; |
217 | ||
1da177e4 LT |
218 | newbrk = PAGE_ALIGN(brk); |
219 | oldbrk = PAGE_ALIGN(mm->brk); | |
220 | if (oldbrk == newbrk) | |
221 | goto set_brk; | |
222 | ||
223 | /* Always allow shrinking brk. */ | |
224 | if (brk <= mm->brk) { | |
225 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
226 | goto set_brk; | |
227 | goto out; | |
228 | } | |
229 | ||
1da177e4 LT |
230 | /* Check against existing mmap mappings. */ |
231 | if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) | |
232 | goto out; | |
233 | ||
234 | /* Ok, looks good - let it rip. */ | |
5d22fc25 | 235 | if (do_brk(oldbrk, newbrk-oldbrk) < 0) |
1da177e4 | 236 | goto out; |
128557ff | 237 | |
1da177e4 LT |
238 | set_brk: |
239 | mm->brk = brk; | |
128557ff ML |
240 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
241 | up_write(&mm->mmap_sem); | |
242 | if (populate) | |
243 | mm_populate(oldbrk, newbrk - oldbrk); | |
244 | return brk; | |
245 | ||
1da177e4 LT |
246 | out: |
247 | retval = mm->brk; | |
248 | up_write(&mm->mmap_sem); | |
249 | return retval; | |
250 | } | |
251 | ||
d3737187 ML |
252 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
253 | { | |
254 | unsigned long max, subtree_gap; | |
255 | max = vma->vm_start; | |
256 | if (vma->vm_prev) | |
257 | max -= vma->vm_prev->vm_end; | |
258 | if (vma->vm_rb.rb_left) { | |
259 | subtree_gap = rb_entry(vma->vm_rb.rb_left, | |
260 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
261 | if (subtree_gap > max) | |
262 | max = subtree_gap; | |
263 | } | |
264 | if (vma->vm_rb.rb_right) { | |
265 | subtree_gap = rb_entry(vma->vm_rb.rb_right, | |
266 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
267 | if (subtree_gap > max) | |
268 | max = subtree_gap; | |
269 | } | |
270 | return max; | |
271 | } | |
272 | ||
ed8ea815 | 273 | #ifdef CONFIG_DEBUG_VM_RB |
acf128d0 | 274 | static int browse_rb(struct mm_struct *mm) |
1da177e4 | 275 | { |
acf128d0 | 276 | struct rb_root *root = &mm->mm_rb; |
5a0768f6 | 277 | int i = 0, j, bug = 0; |
1da177e4 LT |
278 | struct rb_node *nd, *pn = NULL; |
279 | unsigned long prev = 0, pend = 0; | |
280 | ||
281 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
282 | struct vm_area_struct *vma; | |
283 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
5a0768f6 | 284 | if (vma->vm_start < prev) { |
ff26f70f AM |
285 | pr_emerg("vm_start %lx < prev %lx\n", |
286 | vma->vm_start, prev); | |
5a0768f6 ML |
287 | bug = 1; |
288 | } | |
289 | if (vma->vm_start < pend) { | |
ff26f70f AM |
290 | pr_emerg("vm_start %lx < pend %lx\n", |
291 | vma->vm_start, pend); | |
5a0768f6 ML |
292 | bug = 1; |
293 | } | |
294 | if (vma->vm_start > vma->vm_end) { | |
ff26f70f AM |
295 | pr_emerg("vm_start %lx > vm_end %lx\n", |
296 | vma->vm_start, vma->vm_end); | |
5a0768f6 ML |
297 | bug = 1; |
298 | } | |
acf128d0 | 299 | spin_lock(&mm->page_table_lock); |
5a0768f6 | 300 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { |
8542bdfc | 301 | pr_emerg("free gap %lx, correct %lx\n", |
5a0768f6 ML |
302 | vma->rb_subtree_gap, |
303 | vma_compute_subtree_gap(vma)); | |
304 | bug = 1; | |
305 | } | |
acf128d0 | 306 | spin_unlock(&mm->page_table_lock); |
1da177e4 LT |
307 | i++; |
308 | pn = nd; | |
d1af65d1 DM |
309 | prev = vma->vm_start; |
310 | pend = vma->vm_end; | |
1da177e4 LT |
311 | } |
312 | j = 0; | |
5a0768f6 | 313 | for (nd = pn; nd; nd = rb_prev(nd)) |
1da177e4 | 314 | j++; |
5a0768f6 | 315 | if (i != j) { |
8542bdfc | 316 | pr_emerg("backwards %d, forwards %d\n", j, i); |
5a0768f6 | 317 | bug = 1; |
1da177e4 | 318 | } |
5a0768f6 | 319 | return bug ? -1 : i; |
1da177e4 LT |
320 | } |
321 | ||
d3737187 ML |
322 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
323 | { | |
324 | struct rb_node *nd; | |
325 | ||
326 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
327 | struct vm_area_struct *vma; | |
328 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
96dad67f SL |
329 | VM_BUG_ON_VMA(vma != ignore && |
330 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma), | |
331 | vma); | |
1da177e4 | 332 | } |
1da177e4 LT |
333 | } |
334 | ||
eafd4dc4 | 335 | static void validate_mm(struct mm_struct *mm) |
1da177e4 LT |
336 | { |
337 | int bug = 0; | |
338 | int i = 0; | |
5a0768f6 | 339 | unsigned long highest_address = 0; |
ed8ea815 | 340 | struct vm_area_struct *vma = mm->mmap; |
ff26f70f | 341 | |
ed8ea815 | 342 | while (vma) { |
12352d3c | 343 | struct anon_vma *anon_vma = vma->anon_vma; |
ed8ea815 | 344 | struct anon_vma_chain *avc; |
ff26f70f | 345 | |
12352d3c KK |
346 | if (anon_vma) { |
347 | anon_vma_lock_read(anon_vma); | |
348 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
349 | anon_vma_interval_tree_verify(avc); | |
350 | anon_vma_unlock_read(anon_vma); | |
351 | } | |
352 | ||
5a0768f6 | 353 | highest_address = vma->vm_end; |
ed8ea815 | 354 | vma = vma->vm_next; |
1da177e4 LT |
355 | i++; |
356 | } | |
5a0768f6 | 357 | if (i != mm->map_count) { |
8542bdfc | 358 | pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); |
5a0768f6 ML |
359 | bug = 1; |
360 | } | |
361 | if (highest_address != mm->highest_vm_end) { | |
8542bdfc | 362 | pr_emerg("mm->highest_vm_end %lx, found %lx\n", |
ff26f70f | 363 | mm->highest_vm_end, highest_address); |
5a0768f6 ML |
364 | bug = 1; |
365 | } | |
acf128d0 | 366 | i = browse_rb(mm); |
5a0768f6 | 367 | if (i != mm->map_count) { |
ff26f70f AM |
368 | if (i != -1) |
369 | pr_emerg("map_count %d rb %d\n", mm->map_count, i); | |
5a0768f6 ML |
370 | bug = 1; |
371 | } | |
96dad67f | 372 | VM_BUG_ON_MM(bug, mm); |
1da177e4 LT |
373 | } |
374 | #else | |
d3737187 | 375 | #define validate_mm_rb(root, ignore) do { } while (0) |
1da177e4 LT |
376 | #define validate_mm(mm) do { } while (0) |
377 | #endif | |
378 | ||
d3737187 ML |
379 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
380 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) | |
381 | ||
382 | /* | |
383 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or | |
384 | * vma->vm_prev->vm_end values changed, without modifying the vma's position | |
385 | * in the rbtree. | |
386 | */ | |
387 | static void vma_gap_update(struct vm_area_struct *vma) | |
388 | { | |
389 | /* | |
390 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback | |
391 | * function that does exacltly what we want. | |
392 | */ | |
393 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); | |
394 | } | |
395 | ||
396 | static inline void vma_rb_insert(struct vm_area_struct *vma, | |
397 | struct rb_root *root) | |
398 | { | |
399 | /* All rb_subtree_gap values must be consistent prior to insertion */ | |
400 | validate_mm_rb(root, NULL); | |
401 | ||
402 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
403 | } | |
404 | ||
405 | static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) | |
406 | { | |
407 | /* | |
408 | * All rb_subtree_gap values must be consistent prior to erase, | |
409 | * with the possible exception of the vma being erased. | |
410 | */ | |
411 | validate_mm_rb(root, vma); | |
412 | ||
413 | /* | |
414 | * Note rb_erase_augmented is a fairly large inline function, | |
415 | * so make sure we instantiate it only once with our desired | |
416 | * augmented rbtree callbacks. | |
417 | */ | |
418 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
419 | } | |
420 | ||
bf181b9f ML |
421 | /* |
422 | * vma has some anon_vma assigned, and is already inserted on that | |
423 | * anon_vma's interval trees. | |
424 | * | |
425 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
426 | * vma must be removed from the anon_vma's interval trees using | |
427 | * anon_vma_interval_tree_pre_update_vma(). | |
428 | * | |
429 | * After the update, the vma will be reinserted using | |
430 | * anon_vma_interval_tree_post_update_vma(). | |
431 | * | |
432 | * The entire update must be protected by exclusive mmap_sem and by | |
433 | * the root anon_vma's mutex. | |
434 | */ | |
435 | static inline void | |
436 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
437 | { | |
438 | struct anon_vma_chain *avc; | |
439 | ||
440 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
441 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
442 | } | |
443 | ||
444 | static inline void | |
445 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
446 | { | |
447 | struct anon_vma_chain *avc; | |
448 | ||
449 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
450 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
451 | } | |
452 | ||
6597d783 HD |
453 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
454 | unsigned long end, struct vm_area_struct **pprev, | |
455 | struct rb_node ***rb_link, struct rb_node **rb_parent) | |
1da177e4 | 456 | { |
6597d783 | 457 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
1da177e4 LT |
458 | |
459 | __rb_link = &mm->mm_rb.rb_node; | |
460 | rb_prev = __rb_parent = NULL; | |
1da177e4 LT |
461 | |
462 | while (*__rb_link) { | |
463 | struct vm_area_struct *vma_tmp; | |
464 | ||
465 | __rb_parent = *__rb_link; | |
466 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
467 | ||
468 | if (vma_tmp->vm_end > addr) { | |
6597d783 HD |
469 | /* Fail if an existing vma overlaps the area */ |
470 | if (vma_tmp->vm_start < end) | |
471 | return -ENOMEM; | |
1da177e4 LT |
472 | __rb_link = &__rb_parent->rb_left; |
473 | } else { | |
474 | rb_prev = __rb_parent; | |
475 | __rb_link = &__rb_parent->rb_right; | |
476 | } | |
477 | } | |
478 | ||
479 | *pprev = NULL; | |
480 | if (rb_prev) | |
481 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
482 | *rb_link = __rb_link; | |
483 | *rb_parent = __rb_parent; | |
6597d783 | 484 | return 0; |
1da177e4 LT |
485 | } |
486 | ||
e8420a8e CH |
487 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
488 | unsigned long addr, unsigned long end) | |
489 | { | |
490 | unsigned long nr_pages = 0; | |
491 | struct vm_area_struct *vma; | |
492 | ||
493 | /* Find first overlaping mapping */ | |
494 | vma = find_vma_intersection(mm, addr, end); | |
495 | if (!vma) | |
496 | return 0; | |
497 | ||
498 | nr_pages = (min(end, vma->vm_end) - | |
499 | max(addr, vma->vm_start)) >> PAGE_SHIFT; | |
500 | ||
501 | /* Iterate over the rest of the overlaps */ | |
502 | for (vma = vma->vm_next; vma; vma = vma->vm_next) { | |
503 | unsigned long overlap_len; | |
504 | ||
505 | if (vma->vm_start > end) | |
506 | break; | |
507 | ||
508 | overlap_len = min(end, vma->vm_end) - vma->vm_start; | |
509 | nr_pages += overlap_len >> PAGE_SHIFT; | |
510 | } | |
511 | ||
512 | return nr_pages; | |
513 | } | |
514 | ||
1da177e4 LT |
515 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
516 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
517 | { | |
d3737187 ML |
518 | /* Update tracking information for the gap following the new vma. */ |
519 | if (vma->vm_next) | |
520 | vma_gap_update(vma->vm_next); | |
521 | else | |
522 | mm->highest_vm_end = vma->vm_end; | |
523 | ||
524 | /* | |
525 | * vma->vm_prev wasn't known when we followed the rbtree to find the | |
526 | * correct insertion point for that vma. As a result, we could not | |
527 | * update the vma vm_rb parents rb_subtree_gap values on the way down. | |
528 | * So, we first insert the vma with a zero rb_subtree_gap value | |
529 | * (to be consistent with what we did on the way down), and then | |
530 | * immediately update the gap to the correct value. Finally we | |
531 | * rebalance the rbtree after all augmented values have been set. | |
532 | */ | |
1da177e4 | 533 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
d3737187 ML |
534 | vma->rb_subtree_gap = 0; |
535 | vma_gap_update(vma); | |
536 | vma_rb_insert(vma, &mm->mm_rb); | |
1da177e4 LT |
537 | } |
538 | ||
cb8f488c | 539 | static void __vma_link_file(struct vm_area_struct *vma) |
1da177e4 | 540 | { |
48aae425 | 541 | struct file *file; |
1da177e4 LT |
542 | |
543 | file = vma->vm_file; | |
544 | if (file) { | |
545 | struct address_space *mapping = file->f_mapping; | |
546 | ||
547 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 548 | atomic_dec(&file_inode(file)->i_writecount); |
1da177e4 | 549 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 550 | atomic_inc(&mapping->i_mmap_writable); |
1da177e4 LT |
551 | |
552 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 553 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 LT |
554 | flush_dcache_mmap_unlock(mapping); |
555 | } | |
556 | } | |
557 | ||
558 | static void | |
559 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
560 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
561 | struct rb_node *rb_parent) | |
562 | { | |
563 | __vma_link_list(mm, vma, prev, rb_parent); | |
564 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
1da177e4 LT |
565 | } |
566 | ||
567 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
568 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
569 | struct rb_node *rb_parent) | |
570 | { | |
571 | struct address_space *mapping = NULL; | |
572 | ||
64ac4940 | 573 | if (vma->vm_file) { |
1da177e4 | 574 | mapping = vma->vm_file->f_mapping; |
83cde9e8 | 575 | i_mmap_lock_write(mapping); |
64ac4940 | 576 | } |
1da177e4 LT |
577 | |
578 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
579 | __vma_link_file(vma); | |
580 | ||
1da177e4 | 581 | if (mapping) |
83cde9e8 | 582 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
583 | |
584 | mm->map_count++; | |
585 | validate_mm(mm); | |
586 | } | |
587 | ||
588 | /* | |
88f6b4c3 | 589 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
6b2dbba8 | 590 | * mm's list and rbtree. It has already been inserted into the interval tree. |
1da177e4 | 591 | */ |
48aae425 | 592 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 593 | { |
6597d783 | 594 | struct vm_area_struct *prev; |
48aae425 | 595 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 596 | |
6597d783 HD |
597 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
598 | &prev, &rb_link, &rb_parent)) | |
599 | BUG(); | |
1da177e4 LT |
600 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
601 | mm->map_count++; | |
602 | } | |
603 | ||
e86f15ee AA |
604 | static __always_inline void __vma_unlink_common(struct mm_struct *mm, |
605 | struct vm_area_struct *vma, | |
606 | struct vm_area_struct *prev, | |
607 | bool has_prev) | |
1da177e4 | 608 | { |
d3737187 | 609 | struct vm_area_struct *next; |
297c5eee | 610 | |
d3737187 | 611 | vma_rb_erase(vma, &mm->mm_rb); |
e86f15ee AA |
612 | next = vma->vm_next; |
613 | if (has_prev) | |
614 | prev->vm_next = next; | |
615 | else { | |
616 | prev = vma->vm_prev; | |
617 | if (prev) | |
618 | prev->vm_next = next; | |
619 | else | |
620 | mm->mmap = next; | |
621 | } | |
297c5eee LT |
622 | if (next) |
623 | next->vm_prev = prev; | |
615d6e87 DB |
624 | |
625 | /* Kill the cache */ | |
626 | vmacache_invalidate(mm); | |
1da177e4 LT |
627 | } |
628 | ||
e86f15ee AA |
629 | static inline void __vma_unlink_prev(struct mm_struct *mm, |
630 | struct vm_area_struct *vma, | |
631 | struct vm_area_struct *prev) | |
632 | { | |
633 | __vma_unlink_common(mm, vma, prev, true); | |
634 | } | |
635 | ||
636 | static inline void __vma_unlink(struct mm_struct *mm, | |
637 | struct vm_area_struct *vma) | |
638 | { | |
639 | __vma_unlink_common(mm, vma, NULL, false); | |
640 | } | |
641 | ||
1da177e4 LT |
642 | /* |
643 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
644 | * is already present in an i_mmap tree without adjusting the tree. | |
645 | * The following helper function should be used when such adjustments | |
646 | * are necessary. The "insert" vma (if any) is to be inserted | |
647 | * before we drop the necessary locks. | |
648 | */ | |
e86f15ee AA |
649 | int __vma_adjust(struct vm_area_struct *vma, unsigned long start, |
650 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, | |
651 | struct vm_area_struct *expand) | |
1da177e4 LT |
652 | { |
653 | struct mm_struct *mm = vma->vm_mm; | |
e86f15ee | 654 | struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; |
1da177e4 | 655 | struct address_space *mapping = NULL; |
6b2dbba8 | 656 | struct rb_root *root = NULL; |
012f1800 | 657 | struct anon_vma *anon_vma = NULL; |
1da177e4 | 658 | struct file *file = vma->vm_file; |
d3737187 | 659 | bool start_changed = false, end_changed = false; |
1da177e4 LT |
660 | long adjust_next = 0; |
661 | int remove_next = 0; | |
662 | ||
663 | if (next && !insert) { | |
734537c9 | 664 | struct vm_area_struct *exporter = NULL, *importer = NULL; |
287d97ac | 665 | |
1da177e4 LT |
666 | if (end >= next->vm_end) { |
667 | /* | |
668 | * vma expands, overlapping all the next, and | |
669 | * perhaps the one after too (mprotect case 6). | |
e86f15ee AA |
670 | * The only two other cases that gets here are |
671 | * case 1, case 7 and case 8. | |
1da177e4 | 672 | */ |
e86f15ee AA |
673 | if (next == expand) { |
674 | /* | |
675 | * The only case where we don't expand "vma" | |
676 | * and we expand "next" instead is case 8. | |
677 | */ | |
678 | VM_WARN_ON(end != next->vm_end); | |
679 | /* | |
680 | * remove_next == 3 means we're | |
681 | * removing "vma" and that to do so we | |
682 | * swapped "vma" and "next". | |
683 | */ | |
684 | remove_next = 3; | |
685 | VM_WARN_ON(file != next->vm_file); | |
686 | swap(vma, next); | |
687 | } else { | |
688 | VM_WARN_ON(expand != vma); | |
689 | /* | |
690 | * case 1, 6, 7, remove_next == 2 is case 6, | |
691 | * remove_next == 1 is case 1 or 7. | |
692 | */ | |
693 | remove_next = 1 + (end > next->vm_end); | |
694 | VM_WARN_ON(remove_next == 2 && | |
695 | end != next->vm_next->vm_end); | |
696 | VM_WARN_ON(remove_next == 1 && | |
697 | end != next->vm_end); | |
698 | /* trim end to next, for case 6 first pass */ | |
699 | end = next->vm_end; | |
700 | } | |
701 | ||
287d97ac | 702 | exporter = next; |
1da177e4 | 703 | importer = vma; |
734537c9 KS |
704 | |
705 | /* | |
706 | * If next doesn't have anon_vma, import from vma after | |
707 | * next, if the vma overlaps with it. | |
708 | */ | |
97a42cd4 | 709 | if (remove_next == 2 && !next->anon_vma) |
734537c9 KS |
710 | exporter = next->vm_next; |
711 | ||
1da177e4 LT |
712 | } else if (end > next->vm_start) { |
713 | /* | |
714 | * vma expands, overlapping part of the next: | |
715 | * mprotect case 5 shifting the boundary up. | |
716 | */ | |
717 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
287d97ac | 718 | exporter = next; |
1da177e4 | 719 | importer = vma; |
e86f15ee | 720 | VM_WARN_ON(expand != importer); |
1da177e4 LT |
721 | } else if (end < vma->vm_end) { |
722 | /* | |
723 | * vma shrinks, and !insert tells it's not | |
724 | * split_vma inserting another: so it must be | |
725 | * mprotect case 4 shifting the boundary down. | |
726 | */ | |
cc71aba3 | 727 | adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); |
287d97ac | 728 | exporter = vma; |
1da177e4 | 729 | importer = next; |
e86f15ee | 730 | VM_WARN_ON(expand != importer); |
1da177e4 | 731 | } |
1da177e4 | 732 | |
5beb4930 RR |
733 | /* |
734 | * Easily overlooked: when mprotect shifts the boundary, | |
735 | * make sure the expanding vma has anon_vma set if the | |
736 | * shrinking vma had, to cover any anon pages imported. | |
737 | */ | |
287d97ac | 738 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
c4ea95d7 DF |
739 | int error; |
740 | ||
b800c91a | 741 | importer->anon_vma = exporter->anon_vma; |
c4ea95d7 | 742 | error = anon_vma_clone(importer, exporter); |
3fe89b3e | 743 | if (error) |
c4ea95d7 | 744 | return error; |
5beb4930 RR |
745 | } |
746 | } | |
734537c9 | 747 | again: |
e86f15ee | 748 | vma_adjust_trans_huge(orig_vma, start, end, adjust_next); |
37f9f559 | 749 | |
1da177e4 LT |
750 | if (file) { |
751 | mapping = file->f_mapping; | |
27ba0644 KS |
752 | root = &mapping->i_mmap; |
753 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); | |
682968e0 | 754 | |
27ba0644 KS |
755 | if (adjust_next) |
756 | uprobe_munmap(next, next->vm_start, next->vm_end); | |
682968e0 | 757 | |
83cde9e8 | 758 | i_mmap_lock_write(mapping); |
1da177e4 | 759 | if (insert) { |
1da177e4 | 760 | /* |
6b2dbba8 | 761 | * Put into interval tree now, so instantiated pages |
1da177e4 LT |
762 | * are visible to arm/parisc __flush_dcache_page |
763 | * throughout; but we cannot insert into address | |
764 | * space until vma start or end is updated. | |
765 | */ | |
766 | __vma_link_file(insert); | |
767 | } | |
768 | } | |
769 | ||
bf181b9f ML |
770 | anon_vma = vma->anon_vma; |
771 | if (!anon_vma && adjust_next) | |
772 | anon_vma = next->anon_vma; | |
773 | if (anon_vma) { | |
e86f15ee AA |
774 | VM_WARN_ON(adjust_next && next->anon_vma && |
775 | anon_vma != next->anon_vma); | |
4fc3f1d6 | 776 | anon_vma_lock_write(anon_vma); |
bf181b9f ML |
777 | anon_vma_interval_tree_pre_update_vma(vma); |
778 | if (adjust_next) | |
779 | anon_vma_interval_tree_pre_update_vma(next); | |
780 | } | |
012f1800 | 781 | |
1da177e4 LT |
782 | if (root) { |
783 | flush_dcache_mmap_lock(mapping); | |
6b2dbba8 | 784 | vma_interval_tree_remove(vma, root); |
1da177e4 | 785 | if (adjust_next) |
6b2dbba8 | 786 | vma_interval_tree_remove(next, root); |
1da177e4 LT |
787 | } |
788 | ||
d3737187 ML |
789 | if (start != vma->vm_start) { |
790 | vma->vm_start = start; | |
791 | start_changed = true; | |
792 | } | |
793 | if (end != vma->vm_end) { | |
794 | vma->vm_end = end; | |
795 | end_changed = true; | |
796 | } | |
1da177e4 LT |
797 | vma->vm_pgoff = pgoff; |
798 | if (adjust_next) { | |
799 | next->vm_start += adjust_next << PAGE_SHIFT; | |
800 | next->vm_pgoff += adjust_next; | |
801 | } | |
802 | ||
803 | if (root) { | |
804 | if (adjust_next) | |
6b2dbba8 ML |
805 | vma_interval_tree_insert(next, root); |
806 | vma_interval_tree_insert(vma, root); | |
1da177e4 LT |
807 | flush_dcache_mmap_unlock(mapping); |
808 | } | |
809 | ||
810 | if (remove_next) { | |
811 | /* | |
812 | * vma_merge has merged next into vma, and needs | |
813 | * us to remove next before dropping the locks. | |
814 | */ | |
e86f15ee AA |
815 | if (remove_next != 3) |
816 | __vma_unlink_prev(mm, next, vma); | |
817 | else | |
818 | /* vma is not before next if they've been swapped */ | |
819 | __vma_unlink(mm, next); | |
1da177e4 LT |
820 | if (file) |
821 | __remove_shared_vm_struct(next, file, mapping); | |
1da177e4 LT |
822 | } else if (insert) { |
823 | /* | |
824 | * split_vma has split insert from vma, and needs | |
825 | * us to insert it before dropping the locks | |
826 | * (it may either follow vma or precede it). | |
827 | */ | |
828 | __insert_vm_struct(mm, insert); | |
d3737187 ML |
829 | } else { |
830 | if (start_changed) | |
831 | vma_gap_update(vma); | |
832 | if (end_changed) { | |
833 | if (!next) | |
834 | mm->highest_vm_end = end; | |
835 | else if (!adjust_next) | |
836 | vma_gap_update(next); | |
837 | } | |
1da177e4 LT |
838 | } |
839 | ||
bf181b9f ML |
840 | if (anon_vma) { |
841 | anon_vma_interval_tree_post_update_vma(vma); | |
842 | if (adjust_next) | |
843 | anon_vma_interval_tree_post_update_vma(next); | |
08b52706 | 844 | anon_vma_unlock_write(anon_vma); |
bf181b9f | 845 | } |
1da177e4 | 846 | if (mapping) |
83cde9e8 | 847 | i_mmap_unlock_write(mapping); |
1da177e4 | 848 | |
2b144498 | 849 | if (root) { |
7b2d81d4 | 850 | uprobe_mmap(vma); |
2b144498 SD |
851 | |
852 | if (adjust_next) | |
7b2d81d4 | 853 | uprobe_mmap(next); |
2b144498 SD |
854 | } |
855 | ||
1da177e4 | 856 | if (remove_next) { |
925d1c40 | 857 | if (file) { |
cbc91f71 | 858 | uprobe_munmap(next, next->vm_start, next->vm_end); |
1da177e4 | 859 | fput(file); |
925d1c40 | 860 | } |
5beb4930 RR |
861 | if (next->anon_vma) |
862 | anon_vma_merge(vma, next); | |
1da177e4 | 863 | mm->map_count--; |
3964acd0 | 864 | mpol_put(vma_policy(next)); |
1da177e4 LT |
865 | kmem_cache_free(vm_area_cachep, next); |
866 | /* | |
867 | * In mprotect's case 6 (see comments on vma_merge), | |
868 | * we must remove another next too. It would clutter | |
869 | * up the code too much to do both in one go. | |
870 | */ | |
e86f15ee AA |
871 | if (remove_next != 3) { |
872 | /* | |
873 | * If "next" was removed and vma->vm_end was | |
874 | * expanded (up) over it, in turn | |
875 | * "next->vm_prev->vm_end" changed and the | |
876 | * "vma->vm_next" gap must be updated. | |
877 | */ | |
878 | next = vma->vm_next; | |
879 | } else { | |
880 | /* | |
881 | * For the scope of the comment "next" and | |
882 | * "vma" considered pre-swap(): if "vma" was | |
883 | * removed, next->vm_start was expanded (down) | |
884 | * over it and the "next" gap must be updated. | |
885 | * Because of the swap() the post-swap() "vma" | |
886 | * actually points to pre-swap() "next" | |
887 | * (post-swap() "next" as opposed is now a | |
888 | * dangling pointer). | |
889 | */ | |
890 | next = vma; | |
891 | } | |
734537c9 KS |
892 | if (remove_next == 2) { |
893 | remove_next = 1; | |
894 | end = next->vm_end; | |
1da177e4 | 895 | goto again; |
734537c9 | 896 | } |
d3737187 ML |
897 | else if (next) |
898 | vma_gap_update(next); | |
fb8c41e9 AA |
899 | else { |
900 | /* | |
901 | * If remove_next == 2 we obviously can't | |
902 | * reach this path. | |
903 | * | |
904 | * If remove_next == 3 we can't reach this | |
905 | * path because pre-swap() next is always not | |
906 | * NULL. pre-swap() "next" is not being | |
907 | * removed and its next->vm_end is not altered | |
908 | * (and furthermore "end" already matches | |
909 | * next->vm_end in remove_next == 3). | |
910 | * | |
911 | * We reach this only in the remove_next == 1 | |
912 | * case if the "next" vma that was removed was | |
913 | * the highest vma of the mm. However in such | |
914 | * case next->vm_end == "end" and the extended | |
915 | * "vma" has vma->vm_end == next->vm_end so | |
916 | * mm->highest_vm_end doesn't need any update | |
917 | * in remove_next == 1 case. | |
918 | */ | |
919 | VM_WARN_ON(mm->highest_vm_end != end); | |
920 | } | |
1da177e4 | 921 | } |
2b144498 | 922 | if (insert && file) |
7b2d81d4 | 923 | uprobe_mmap(insert); |
1da177e4 LT |
924 | |
925 | validate_mm(mm); | |
5beb4930 RR |
926 | |
927 | return 0; | |
1da177e4 LT |
928 | } |
929 | ||
930 | /* | |
931 | * If the vma has a ->close operation then the driver probably needs to release | |
932 | * per-vma resources, so we don't attempt to merge those. | |
933 | */ | |
1da177e4 | 934 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
19a809af AA |
935 | struct file *file, unsigned long vm_flags, |
936 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 937 | { |
34228d47 CG |
938 | /* |
939 | * VM_SOFTDIRTY should not prevent from VMA merging, if we | |
940 | * match the flags but dirty bit -- the caller should mark | |
941 | * merged VMA as dirty. If dirty bit won't be excluded from | |
942 | * comparison, we increase pressue on the memory system forcing | |
943 | * the kernel to generate new VMAs when old one could be | |
944 | * extended instead. | |
945 | */ | |
946 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) | |
1da177e4 LT |
947 | return 0; |
948 | if (vma->vm_file != file) | |
949 | return 0; | |
950 | if (vma->vm_ops && vma->vm_ops->close) | |
951 | return 0; | |
19a809af AA |
952 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) |
953 | return 0; | |
1da177e4 LT |
954 | return 1; |
955 | } | |
956 | ||
957 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
965f55de SL |
958 | struct anon_vma *anon_vma2, |
959 | struct vm_area_struct *vma) | |
1da177e4 | 960 | { |
965f55de SL |
961 | /* |
962 | * The list_is_singular() test is to avoid merging VMA cloned from | |
963 | * parents. This can improve scalability caused by anon_vma lock. | |
964 | */ | |
965 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
966 | list_is_singular(&vma->anon_vma_chain))) | |
967 | return 1; | |
968 | return anon_vma1 == anon_vma2; | |
1da177e4 LT |
969 | } |
970 | ||
971 | /* | |
972 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
973 | * in front of (at a lower virtual address and file offset than) the vma. | |
974 | * | |
975 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
976 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
977 | * | |
978 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
979 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
980 | * wrap, nor mmaps which cover the final page at index -1UL. | |
981 | */ | |
982 | static int | |
983 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
984 | struct anon_vma *anon_vma, struct file *file, |
985 | pgoff_t vm_pgoff, | |
986 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 987 | { |
19a809af | 988 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 989 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
990 | if (vma->vm_pgoff == vm_pgoff) |
991 | return 1; | |
992 | } | |
993 | return 0; | |
994 | } | |
995 | ||
996 | /* | |
997 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
998 | * beyond (at a higher virtual address and file offset than) the vma. | |
999 | * | |
1000 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
1001 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
1002 | */ | |
1003 | static int | |
1004 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
1005 | struct anon_vma *anon_vma, struct file *file, |
1006 | pgoff_t vm_pgoff, | |
1007 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 1008 | { |
19a809af | 1009 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 1010 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 1011 | pgoff_t vm_pglen; |
d6e93217 | 1012 | vm_pglen = vma_pages(vma); |
1da177e4 LT |
1013 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
1014 | return 1; | |
1015 | } | |
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
1021 | * whether that can be merged with its predecessor or its successor. | |
1022 | * Or both (it neatly fills a hole). | |
1023 | * | |
1024 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
1025 | * certain not to be mapped by the time vma_merge is called; but when | |
1026 | * called for mprotect, it is certain to be already mapped (either at | |
1027 | * an offset within prev, or at the start of next), and the flags of | |
1028 | * this area are about to be changed to vm_flags - and the no-change | |
1029 | * case has already been eliminated. | |
1030 | * | |
1031 | * The following mprotect cases have to be considered, where AAAA is | |
1032 | * the area passed down from mprotect_fixup, never extending beyond one | |
1033 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
1034 | * | |
1035 | * AAAA AAAA AAAA AAAA | |
1036 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
1037 | * cannot merge might become might become might become | |
1038 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
1039 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
e86f15ee | 1040 | * mremap move: PPPPXXXXXXXX 8 |
1da177e4 LT |
1041 | * AAAA |
1042 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
1043 | * might become case 1 below case 2 below case 3 below | |
1044 | * | |
e86f15ee AA |
1045 | * It is important for case 8 that the the vma NNNN overlapping the |
1046 | * region AAAA is never going to extended over XXXX. Instead XXXX must | |
1047 | * be extended in region AAAA and NNNN must be removed. This way in | |
1048 | * all cases where vma_merge succeeds, the moment vma_adjust drops the | |
1049 | * rmap_locks, the properties of the merged vma will be already | |
1050 | * correct for the whole merged range. Some of those properties like | |
1051 | * vm_page_prot/vm_flags may be accessed by rmap_walks and they must | |
1052 | * be correct for the whole merged range immediately after the | |
1053 | * rmap_locks are released. Otherwise if XXXX would be removed and | |
1054 | * NNNN would be extended over the XXXX range, remove_migration_ptes | |
1055 | * or other rmap walkers (if working on addresses beyond the "end" | |
1056 | * parameter) may establish ptes with the wrong permissions of NNNN | |
1057 | * instead of the right permissions of XXXX. | |
1da177e4 LT |
1058 | */ |
1059 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
1060 | struct vm_area_struct *prev, unsigned long addr, | |
1061 | unsigned long end, unsigned long vm_flags, | |
cc71aba3 | 1062 | struct anon_vma *anon_vma, struct file *file, |
19a809af AA |
1063 | pgoff_t pgoff, struct mempolicy *policy, |
1064 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 LT |
1065 | { |
1066 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
1067 | struct vm_area_struct *area, *next; | |
5beb4930 | 1068 | int err; |
1da177e4 LT |
1069 | |
1070 | /* | |
1071 | * We later require that vma->vm_flags == vm_flags, | |
1072 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
1073 | */ | |
1074 | if (vm_flags & VM_SPECIAL) | |
1075 | return NULL; | |
1076 | ||
1077 | if (prev) | |
1078 | next = prev->vm_next; | |
1079 | else | |
1080 | next = mm->mmap; | |
1081 | area = next; | |
e86f15ee | 1082 | if (area && area->vm_end == end) /* cases 6, 7, 8 */ |
1da177e4 LT |
1083 | next = next->vm_next; |
1084 | ||
e86f15ee AA |
1085 | /* verify some invariant that must be enforced by the caller */ |
1086 | VM_WARN_ON(prev && addr <= prev->vm_start); | |
1087 | VM_WARN_ON(area && end > area->vm_end); | |
1088 | VM_WARN_ON(addr >= end); | |
1089 | ||
1da177e4 LT |
1090 | /* |
1091 | * Can it merge with the predecessor? | |
1092 | */ | |
1093 | if (prev && prev->vm_end == addr && | |
cc71aba3 | 1094 | mpol_equal(vma_policy(prev), policy) && |
1da177e4 | 1095 | can_vma_merge_after(prev, vm_flags, |
19a809af AA |
1096 | anon_vma, file, pgoff, |
1097 | vm_userfaultfd_ctx)) { | |
1da177e4 LT |
1098 | /* |
1099 | * OK, it can. Can we now merge in the successor as well? | |
1100 | */ | |
1101 | if (next && end == next->vm_start && | |
1102 | mpol_equal(policy, vma_policy(next)) && | |
1103 | can_vma_merge_before(next, vm_flags, | |
19a809af AA |
1104 | anon_vma, file, |
1105 | pgoff+pglen, | |
1106 | vm_userfaultfd_ctx) && | |
1da177e4 | 1107 | is_mergeable_anon_vma(prev->anon_vma, |
965f55de | 1108 | next->anon_vma, NULL)) { |
1da177e4 | 1109 | /* cases 1, 6 */ |
e86f15ee AA |
1110 | err = __vma_adjust(prev, prev->vm_start, |
1111 | next->vm_end, prev->vm_pgoff, NULL, | |
1112 | prev); | |
1da177e4 | 1113 | } else /* cases 2, 5, 7 */ |
e86f15ee AA |
1114 | err = __vma_adjust(prev, prev->vm_start, |
1115 | end, prev->vm_pgoff, NULL, prev); | |
5beb4930 RR |
1116 | if (err) |
1117 | return NULL; | |
6d50e60c | 1118 | khugepaged_enter_vma_merge(prev, vm_flags); |
1da177e4 LT |
1119 | return prev; |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * Can this new request be merged in front of next? | |
1124 | */ | |
1125 | if (next && end == next->vm_start && | |
cc71aba3 | 1126 | mpol_equal(policy, vma_policy(next)) && |
1da177e4 | 1127 | can_vma_merge_before(next, vm_flags, |
19a809af AA |
1128 | anon_vma, file, pgoff+pglen, |
1129 | vm_userfaultfd_ctx)) { | |
1da177e4 | 1130 | if (prev && addr < prev->vm_end) /* case 4 */ |
e86f15ee AA |
1131 | err = __vma_adjust(prev, prev->vm_start, |
1132 | addr, prev->vm_pgoff, NULL, next); | |
1133 | else { /* cases 3, 8 */ | |
1134 | err = __vma_adjust(area, addr, next->vm_end, | |
1135 | next->vm_pgoff - pglen, NULL, next); | |
1136 | /* | |
1137 | * In case 3 area is already equal to next and | |
1138 | * this is a noop, but in case 8 "area" has | |
1139 | * been removed and next was expanded over it. | |
1140 | */ | |
1141 | area = next; | |
1142 | } | |
5beb4930 RR |
1143 | if (err) |
1144 | return NULL; | |
6d50e60c | 1145 | khugepaged_enter_vma_merge(area, vm_flags); |
1da177e4 LT |
1146 | return area; |
1147 | } | |
1148 | ||
1149 | return NULL; | |
1150 | } | |
1151 | ||
d0e9fe17 LT |
1152 | /* |
1153 | * Rough compatbility check to quickly see if it's even worth looking | |
1154 | * at sharing an anon_vma. | |
1155 | * | |
1156 | * They need to have the same vm_file, and the flags can only differ | |
1157 | * in things that mprotect may change. | |
1158 | * | |
1159 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1160 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1161 | * there is a vm_ops->close() function, because that indicates that the | |
1162 | * driver is doing some kind of reference counting. But that doesn't | |
1163 | * really matter for the anon_vma sharing case. | |
1164 | */ | |
1165 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1166 | { | |
1167 | return a->vm_end == b->vm_start && | |
1168 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1169 | a->vm_file == b->vm_file && | |
34228d47 | 1170 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && |
d0e9fe17 LT |
1171 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
1172 | } | |
1173 | ||
1174 | /* | |
1175 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1176 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1177 | * the same as 'old', the other will be the new one that is trying | |
1178 | * to share the anon_vma. | |
1179 | * | |
1180 | * NOTE! This runs with mm_sem held for reading, so it is possible that | |
1181 | * the anon_vma of 'old' is concurrently in the process of being set up | |
1182 | * by another page fault trying to merge _that_. But that's ok: if it | |
1183 | * is being set up, that automatically means that it will be a singleton | |
1184 | * acceptable for merging, so we can do all of this optimistically. But | |
4db0c3c2 | 1185 | * we do that READ_ONCE() to make sure that we never re-load the pointer. |
d0e9fe17 LT |
1186 | * |
1187 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1188 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1189 | * is to return an anon_vma that is "complex" due to having gone through | |
1190 | * a fork). | |
1191 | * | |
1192 | * We also make sure that the two vma's are compatible (adjacent, | |
1193 | * and with the same memory policies). That's all stable, even with just | |
1194 | * a read lock on the mm_sem. | |
1195 | */ | |
1196 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1197 | { | |
1198 | if (anon_vma_compatible(a, b)) { | |
4db0c3c2 | 1199 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); |
d0e9fe17 LT |
1200 | |
1201 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1202 | return anon_vma; | |
1203 | } | |
1204 | return NULL; | |
1205 | } | |
1206 | ||
1da177e4 LT |
1207 | /* |
1208 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1209 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1210 | * to allocate a new anon_vma. It checks because a repetitive | |
1211 | * sequence of mprotects and faults may otherwise lead to distinct | |
1212 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1213 | * mprotect. | |
1214 | */ | |
1215 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1216 | { | |
d0e9fe17 | 1217 | struct anon_vma *anon_vma; |
1da177e4 | 1218 | struct vm_area_struct *near; |
1da177e4 LT |
1219 | |
1220 | near = vma->vm_next; | |
1221 | if (!near) | |
1222 | goto try_prev; | |
1223 | ||
d0e9fe17 LT |
1224 | anon_vma = reusable_anon_vma(near, vma, near); |
1225 | if (anon_vma) | |
1226 | return anon_vma; | |
1da177e4 | 1227 | try_prev: |
9be34c9d | 1228 | near = vma->vm_prev; |
1da177e4 LT |
1229 | if (!near) |
1230 | goto none; | |
1231 | ||
d0e9fe17 LT |
1232 | anon_vma = reusable_anon_vma(near, near, vma); |
1233 | if (anon_vma) | |
1234 | return anon_vma; | |
1da177e4 LT |
1235 | none: |
1236 | /* | |
1237 | * There's no absolute need to look only at touching neighbours: | |
1238 | * we could search further afield for "compatible" anon_vmas. | |
1239 | * But it would probably just be a waste of time searching, | |
1240 | * or lead to too many vmas hanging off the same anon_vma. | |
1241 | * We're trying to allow mprotect remerging later on, | |
1242 | * not trying to minimize memory used for anon_vmas. | |
1243 | */ | |
1244 | return NULL; | |
1245 | } | |
1246 | ||
40401530 AV |
1247 | /* |
1248 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1249 | * low as possible but still greater than mmap_min_addr | |
1250 | */ | |
1251 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1252 | { | |
1253 | hint &= PAGE_MASK; | |
1254 | if (((void *)hint != NULL) && | |
1255 | (hint < mmap_min_addr)) | |
1256 | return PAGE_ALIGN(mmap_min_addr); | |
1257 | return hint; | |
1258 | } | |
1259 | ||
363ee17f DB |
1260 | static inline int mlock_future_check(struct mm_struct *mm, |
1261 | unsigned long flags, | |
1262 | unsigned long len) | |
1263 | { | |
1264 | unsigned long locked, lock_limit; | |
1265 | ||
1266 | /* mlock MCL_FUTURE? */ | |
1267 | if (flags & VM_LOCKED) { | |
1268 | locked = len >> PAGE_SHIFT; | |
1269 | locked += mm->locked_vm; | |
1270 | lock_limit = rlimit(RLIMIT_MEMLOCK); | |
1271 | lock_limit >>= PAGE_SHIFT; | |
1272 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
1273 | return -EAGAIN; | |
1274 | } | |
1275 | return 0; | |
1276 | } | |
1277 | ||
1da177e4 | 1278 | /* |
27f5de79 | 1279 | * The caller must hold down_write(¤t->mm->mmap_sem). |
1da177e4 | 1280 | */ |
1fcfd8db | 1281 | unsigned long do_mmap(struct file *file, unsigned long addr, |
1da177e4 | 1282 | unsigned long len, unsigned long prot, |
1fcfd8db ON |
1283 | unsigned long flags, vm_flags_t vm_flags, |
1284 | unsigned long pgoff, unsigned long *populate) | |
1da177e4 | 1285 | { |
cc71aba3 | 1286 | struct mm_struct *mm = current->mm; |
62b5f7d0 | 1287 | int pkey = 0; |
1da177e4 | 1288 | |
41badc15 | 1289 | *populate = 0; |
bebeb3d6 | 1290 | |
e37609bb PK |
1291 | if (!len) |
1292 | return -EINVAL; | |
1293 | ||
1da177e4 LT |
1294 | /* |
1295 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1296 | * | |
1297 | * (the exception is when the underlying filesystem is noexec | |
1298 | * mounted, in which case we dont add PROT_EXEC.) | |
1299 | */ | |
1300 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
90f8572b | 1301 | if (!(file && path_noexec(&file->f_path))) |
1da177e4 LT |
1302 | prot |= PROT_EXEC; |
1303 | ||
7cd94146 EP |
1304 | if (!(flags & MAP_FIXED)) |
1305 | addr = round_hint_to_min(addr); | |
1306 | ||
1da177e4 LT |
1307 | /* Careful about overflows.. */ |
1308 | len = PAGE_ALIGN(len); | |
9206de95 | 1309 | if (!len) |
1da177e4 LT |
1310 | return -ENOMEM; |
1311 | ||
1312 | /* offset overflow? */ | |
1313 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
cc71aba3 | 1314 | return -EOVERFLOW; |
1da177e4 LT |
1315 | |
1316 | /* Too many mappings? */ | |
1317 | if (mm->map_count > sysctl_max_map_count) | |
1318 | return -ENOMEM; | |
1319 | ||
1320 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1321 | * that it represents a valid section of the address space. | |
1322 | */ | |
1323 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
de1741a1 | 1324 | if (offset_in_page(addr)) |
1da177e4 LT |
1325 | return addr; |
1326 | ||
62b5f7d0 DH |
1327 | if (prot == PROT_EXEC) { |
1328 | pkey = execute_only_pkey(mm); | |
1329 | if (pkey < 0) | |
1330 | pkey = 0; | |
1331 | } | |
1332 | ||
1da177e4 LT |
1333 | /* Do simple checking here so the lower-level routines won't have |
1334 | * to. we assume access permissions have been handled by the open | |
1335 | * of the memory object, so we don't do any here. | |
1336 | */ | |
62b5f7d0 | 1337 | vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | |
1da177e4 LT |
1338 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1339 | ||
cdf7b341 | 1340 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1341 | if (!can_do_mlock()) |
1342 | return -EPERM; | |
ba470de4 | 1343 | |
363ee17f DB |
1344 | if (mlock_future_check(mm, vm_flags, len)) |
1345 | return -EAGAIN; | |
1da177e4 | 1346 | |
1da177e4 | 1347 | if (file) { |
077bf22b ON |
1348 | struct inode *inode = file_inode(file); |
1349 | ||
1da177e4 LT |
1350 | switch (flags & MAP_TYPE) { |
1351 | case MAP_SHARED: | |
1352 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
1353 | return -EACCES; | |
1354 | ||
1355 | /* | |
1356 | * Make sure we don't allow writing to an append-only | |
1357 | * file.. | |
1358 | */ | |
1359 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1360 | return -EACCES; | |
1361 | ||
1362 | /* | |
1363 | * Make sure there are no mandatory locks on the file. | |
1364 | */ | |
d7a06983 | 1365 | if (locks_verify_locked(file)) |
1da177e4 LT |
1366 | return -EAGAIN; |
1367 | ||
1368 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
1369 | if (!(file->f_mode & FMODE_WRITE)) | |
1370 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
1371 | ||
1372 | /* fall through */ | |
1373 | case MAP_PRIVATE: | |
1374 | if (!(file->f_mode & FMODE_READ)) | |
1375 | return -EACCES; | |
90f8572b | 1376 | if (path_noexec(&file->f_path)) { |
80c5606c LT |
1377 | if (vm_flags & VM_EXEC) |
1378 | return -EPERM; | |
1379 | vm_flags &= ~VM_MAYEXEC; | |
1380 | } | |
80c5606c | 1381 | |
72c2d531 | 1382 | if (!file->f_op->mmap) |
80c5606c | 1383 | return -ENODEV; |
b2c56e4f ON |
1384 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1385 | return -EINVAL; | |
1da177e4 LT |
1386 | break; |
1387 | ||
1388 | default: | |
1389 | return -EINVAL; | |
1390 | } | |
1391 | } else { | |
1392 | switch (flags & MAP_TYPE) { | |
1393 | case MAP_SHARED: | |
b2c56e4f ON |
1394 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1395 | return -EINVAL; | |
ce363942 TH |
1396 | /* |
1397 | * Ignore pgoff. | |
1398 | */ | |
1399 | pgoff = 0; | |
1da177e4 LT |
1400 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1401 | break; | |
1402 | case MAP_PRIVATE: | |
1403 | /* | |
1404 | * Set pgoff according to addr for anon_vma. | |
1405 | */ | |
1406 | pgoff = addr >> PAGE_SHIFT; | |
1407 | break; | |
1408 | default: | |
1409 | return -EINVAL; | |
1410 | } | |
1411 | } | |
1412 | ||
c22c0d63 ML |
1413 | /* |
1414 | * Set 'VM_NORESERVE' if we should not account for the | |
1415 | * memory use of this mapping. | |
1416 | */ | |
1417 | if (flags & MAP_NORESERVE) { | |
1418 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1419 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1420 | vm_flags |= VM_NORESERVE; | |
1421 | ||
1422 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1423 | if (file && is_file_hugepages(file)) | |
1424 | vm_flags |= VM_NORESERVE; | |
1425 | } | |
1426 | ||
1427 | addr = mmap_region(file, addr, len, vm_flags, pgoff); | |
09a9f1d2 ML |
1428 | if (!IS_ERR_VALUE(addr) && |
1429 | ((vm_flags & VM_LOCKED) || | |
1430 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1431 | *populate = len; |
bebeb3d6 | 1432 | return addr; |
0165ab44 | 1433 | } |
6be5ceb0 | 1434 | |
66f0dc48 HD |
1435 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1436 | unsigned long, prot, unsigned long, flags, | |
1437 | unsigned long, fd, unsigned long, pgoff) | |
1438 | { | |
1439 | struct file *file = NULL; | |
1e3ee14b | 1440 | unsigned long retval; |
66f0dc48 HD |
1441 | |
1442 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1443 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1444 | file = fget(fd); |
1445 | if (!file) | |
1e3ee14b | 1446 | return -EBADF; |
af73e4d9 NH |
1447 | if (is_file_hugepages(file)) |
1448 | len = ALIGN(len, huge_page_size(hstate_file(file))); | |
493af578 JE |
1449 | retval = -EINVAL; |
1450 | if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) | |
1451 | goto out_fput; | |
66f0dc48 HD |
1452 | } else if (flags & MAP_HUGETLB) { |
1453 | struct user_struct *user = NULL; | |
c103a4dc | 1454 | struct hstate *hs; |
af73e4d9 | 1455 | |
c103a4dc | 1456 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); |
091d0d55 LZ |
1457 | if (!hs) |
1458 | return -EINVAL; | |
1459 | ||
1460 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1461 | /* |
1462 | * VM_NORESERVE is used because the reservations will be | |
1463 | * taken when vm_ops->mmap() is called | |
1464 | * A dummy user value is used because we are not locking | |
1465 | * memory so no accounting is necessary | |
1466 | */ | |
af73e4d9 | 1467 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f AK |
1468 | VM_NORESERVE, |
1469 | &user, HUGETLB_ANONHUGE_INODE, | |
1470 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); | |
66f0dc48 HD |
1471 | if (IS_ERR(file)) |
1472 | return PTR_ERR(file); | |
1473 | } | |
1474 | ||
1475 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1476 | ||
9fbeb5ab | 1477 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
493af578 | 1478 | out_fput: |
66f0dc48 HD |
1479 | if (file) |
1480 | fput(file); | |
66f0dc48 HD |
1481 | return retval; |
1482 | } | |
1483 | ||
a4679373 CH |
1484 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1485 | struct mmap_arg_struct { | |
1486 | unsigned long addr; | |
1487 | unsigned long len; | |
1488 | unsigned long prot; | |
1489 | unsigned long flags; | |
1490 | unsigned long fd; | |
1491 | unsigned long offset; | |
1492 | }; | |
1493 | ||
1494 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1495 | { | |
1496 | struct mmap_arg_struct a; | |
1497 | ||
1498 | if (copy_from_user(&a, arg, sizeof(a))) | |
1499 | return -EFAULT; | |
de1741a1 | 1500 | if (offset_in_page(a.offset)) |
a4679373 CH |
1501 | return -EINVAL; |
1502 | ||
1503 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1504 | a.offset >> PAGE_SHIFT); | |
1505 | } | |
1506 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1507 | ||
4e950f6f AD |
1508 | /* |
1509 | * Some shared mappigns will want the pages marked read-only | |
1510 | * to track write events. If so, we'll downgrade vm_page_prot | |
1511 | * to the private version (using protection_map[] without the | |
1512 | * VM_SHARED bit). | |
1513 | */ | |
6d2329f8 | 1514 | int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) |
4e950f6f | 1515 | { |
ca16d140 | 1516 | vm_flags_t vm_flags = vma->vm_flags; |
8a04446a | 1517 | const struct vm_operations_struct *vm_ops = vma->vm_ops; |
4e950f6f AD |
1518 | |
1519 | /* If it was private or non-writable, the write bit is already clear */ | |
1520 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1521 | return 0; | |
1522 | ||
1523 | /* The backer wishes to know when pages are first written to? */ | |
8a04446a | 1524 | if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) |
4e950f6f AD |
1525 | return 1; |
1526 | ||
64e45507 PF |
1527 | /* The open routine did something to the protections that pgprot_modify |
1528 | * won't preserve? */ | |
6d2329f8 AA |
1529 | if (pgprot_val(vm_page_prot) != |
1530 | pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) | |
4e950f6f AD |
1531 | return 0; |
1532 | ||
64e45507 PF |
1533 | /* Do we need to track softdirty? */ |
1534 | if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) | |
1535 | return 1; | |
1536 | ||
4e950f6f | 1537 | /* Specialty mapping? */ |
4b6e1e37 | 1538 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1539 | return 0; |
1540 | ||
1541 | /* Can the mapping track the dirty pages? */ | |
1542 | return vma->vm_file && vma->vm_file->f_mapping && | |
1543 | mapping_cap_account_dirty(vma->vm_file->f_mapping); | |
1544 | } | |
1545 | ||
fc8744ad LT |
1546 | /* |
1547 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1548 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1549 | */ |
ca16d140 | 1550 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1551 | { |
5a6fe125 MG |
1552 | /* |
1553 | * hugetlb has its own accounting separate from the core VM | |
1554 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1555 | */ | |
1556 | if (file && is_file_hugepages(file)) | |
1557 | return 0; | |
1558 | ||
fc8744ad LT |
1559 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1560 | } | |
1561 | ||
0165ab44 | 1562 | unsigned long mmap_region(struct file *file, unsigned long addr, |
c22c0d63 | 1563 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) |
0165ab44 MS |
1564 | { |
1565 | struct mm_struct *mm = current->mm; | |
1566 | struct vm_area_struct *vma, *prev; | |
0165ab44 MS |
1567 | int error; |
1568 | struct rb_node **rb_link, *rb_parent; | |
1569 | unsigned long charged = 0; | |
0165ab44 | 1570 | |
e8420a8e | 1571 | /* Check against address space limit. */ |
84638335 | 1572 | if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { |
e8420a8e CH |
1573 | unsigned long nr_pages; |
1574 | ||
1575 | /* | |
1576 | * MAP_FIXED may remove pages of mappings that intersects with | |
1577 | * requested mapping. Account for the pages it would unmap. | |
1578 | */ | |
e8420a8e CH |
1579 | nr_pages = count_vma_pages_range(mm, addr, addr + len); |
1580 | ||
84638335 KK |
1581 | if (!may_expand_vm(mm, vm_flags, |
1582 | (len >> PAGE_SHIFT) - nr_pages)) | |
e8420a8e CH |
1583 | return -ENOMEM; |
1584 | } | |
1585 | ||
1da177e4 | 1586 | /* Clear old maps */ |
9fcd1457 RV |
1587 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
1588 | &rb_parent)) { | |
1da177e4 LT |
1589 | if (do_munmap(mm, addr, len)) |
1590 | return -ENOMEM; | |
1da177e4 LT |
1591 | } |
1592 | ||
fc8744ad LT |
1593 | /* |
1594 | * Private writable mapping: check memory availability | |
1595 | */ | |
5a6fe125 | 1596 | if (accountable_mapping(file, vm_flags)) { |
fc8744ad | 1597 | charged = len >> PAGE_SHIFT; |
191c5424 | 1598 | if (security_vm_enough_memory_mm(mm, charged)) |
fc8744ad LT |
1599 | return -ENOMEM; |
1600 | vm_flags |= VM_ACCOUNT; | |
1da177e4 LT |
1601 | } |
1602 | ||
1603 | /* | |
de33c8db | 1604 | * Can we just expand an old mapping? |
1da177e4 | 1605 | */ |
19a809af AA |
1606 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, |
1607 | NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); | |
de33c8db LT |
1608 | if (vma) |
1609 | goto out; | |
1da177e4 LT |
1610 | |
1611 | /* | |
1612 | * Determine the object being mapped and call the appropriate | |
1613 | * specific mapper. the address has already been validated, but | |
1614 | * not unmapped, but the maps are removed from the list. | |
1615 | */ | |
c5e3b83e | 1616 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1617 | if (!vma) { |
1618 | error = -ENOMEM; | |
1619 | goto unacct_error; | |
1620 | } | |
1da177e4 LT |
1621 | |
1622 | vma->vm_mm = mm; | |
1623 | vma->vm_start = addr; | |
1624 | vma->vm_end = addr + len; | |
1625 | vma->vm_flags = vm_flags; | |
3ed75eb8 | 1626 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
1da177e4 | 1627 | vma->vm_pgoff = pgoff; |
5beb4930 | 1628 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
1629 | |
1630 | if (file) { | |
1da177e4 LT |
1631 | if (vm_flags & VM_DENYWRITE) { |
1632 | error = deny_write_access(file); | |
1633 | if (error) | |
1634 | goto free_vma; | |
1da177e4 | 1635 | } |
4bb5f5d9 DH |
1636 | if (vm_flags & VM_SHARED) { |
1637 | error = mapping_map_writable(file->f_mapping); | |
1638 | if (error) | |
1639 | goto allow_write_and_free_vma; | |
1640 | } | |
1641 | ||
1642 | /* ->mmap() can change vma->vm_file, but must guarantee that | |
1643 | * vma_link() below can deny write-access if VM_DENYWRITE is set | |
1644 | * and map writably if VM_SHARED is set. This usually means the | |
1645 | * new file must not have been exposed to user-space, yet. | |
1646 | */ | |
cb0942b8 | 1647 | vma->vm_file = get_file(file); |
1da177e4 LT |
1648 | error = file->f_op->mmap(file, vma); |
1649 | if (error) | |
1650 | goto unmap_and_free_vma; | |
f8dbf0a7 HS |
1651 | |
1652 | /* Can addr have changed?? | |
1653 | * | |
1654 | * Answer: Yes, several device drivers can do it in their | |
1655 | * f_op->mmap method. -DaveM | |
2897b4d2 JK |
1656 | * Bug: If addr is changed, prev, rb_link, rb_parent should |
1657 | * be updated for vma_link() | |
f8dbf0a7 | 1658 | */ |
2897b4d2 JK |
1659 | WARN_ON_ONCE(addr != vma->vm_start); |
1660 | ||
f8dbf0a7 | 1661 | addr = vma->vm_start; |
f8dbf0a7 | 1662 | vm_flags = vma->vm_flags; |
1da177e4 LT |
1663 | } else if (vm_flags & VM_SHARED) { |
1664 | error = shmem_zero_setup(vma); | |
1665 | if (error) | |
1666 | goto free_vma; | |
1667 | } | |
1668 | ||
de33c8db | 1669 | vma_link(mm, vma, prev, rb_link, rb_parent); |
4d3d5b41 | 1670 | /* Once vma denies write, undo our temporary denial count */ |
4bb5f5d9 DH |
1671 | if (file) { |
1672 | if (vm_flags & VM_SHARED) | |
1673 | mapping_unmap_writable(file->f_mapping); | |
1674 | if (vm_flags & VM_DENYWRITE) | |
1675 | allow_write_access(file); | |
1676 | } | |
e8686772 | 1677 | file = vma->vm_file; |
4d3d5b41 | 1678 | out: |
cdd6c482 | 1679 | perf_event_mmap(vma); |
0a4a9391 | 1680 | |
84638335 | 1681 | vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); |
1da177e4 | 1682 | if (vm_flags & VM_LOCKED) { |
bebeb3d6 ML |
1683 | if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || |
1684 | vma == get_gate_vma(current->mm))) | |
06f9d8c2 | 1685 | mm->locked_vm += (len >> PAGE_SHIFT); |
bebeb3d6 | 1686 | else |
de60f5f1 | 1687 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
bebeb3d6 | 1688 | } |
2b144498 | 1689 | |
c7a3a88c ON |
1690 | if (file) |
1691 | uprobe_mmap(vma); | |
2b144498 | 1692 | |
d9104d1c CG |
1693 | /* |
1694 | * New (or expanded) vma always get soft dirty status. | |
1695 | * Otherwise user-space soft-dirty page tracker won't | |
1696 | * be able to distinguish situation when vma area unmapped, | |
1697 | * then new mapped in-place (which must be aimed as | |
1698 | * a completely new data area). | |
1699 | */ | |
1700 | vma->vm_flags |= VM_SOFTDIRTY; | |
1701 | ||
64e45507 PF |
1702 | vma_set_page_prot(vma); |
1703 | ||
1da177e4 LT |
1704 | return addr; |
1705 | ||
1706 | unmap_and_free_vma: | |
1da177e4 LT |
1707 | vma->vm_file = NULL; |
1708 | fput(file); | |
1709 | ||
1710 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1711 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1712 | charged = 0; | |
4bb5f5d9 DH |
1713 | if (vm_flags & VM_SHARED) |
1714 | mapping_unmap_writable(file->f_mapping); | |
1715 | allow_write_and_free_vma: | |
1716 | if (vm_flags & VM_DENYWRITE) | |
1717 | allow_write_access(file); | |
1da177e4 LT |
1718 | free_vma: |
1719 | kmem_cache_free(vm_area_cachep, vma); | |
1720 | unacct_error: | |
1721 | if (charged) | |
1722 | vm_unacct_memory(charged); | |
1723 | return error; | |
1724 | } | |
1725 | ||
db4fbfb9 ML |
1726 | unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
1727 | { | |
1728 | /* | |
1729 | * We implement the search by looking for an rbtree node that | |
1730 | * immediately follows a suitable gap. That is, | |
1731 | * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; | |
1732 | * - gap_end = vma->vm_start >= info->low_limit + length; | |
1733 | * - gap_end - gap_start >= length | |
1734 | */ | |
1735 | ||
1736 | struct mm_struct *mm = current->mm; | |
1737 | struct vm_area_struct *vma; | |
1738 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1739 | ||
1740 | /* Adjust search length to account for worst case alignment overhead */ | |
1741 | length = info->length + info->align_mask; | |
1742 | if (length < info->length) | |
1743 | return -ENOMEM; | |
1744 | ||
1745 | /* Adjust search limits by the desired length */ | |
1746 | if (info->high_limit < length) | |
1747 | return -ENOMEM; | |
1748 | high_limit = info->high_limit - length; | |
1749 | ||
1750 | if (info->low_limit > high_limit) | |
1751 | return -ENOMEM; | |
1752 | low_limit = info->low_limit + length; | |
1753 | ||
1754 | /* Check if rbtree root looks promising */ | |
1755 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1756 | goto check_highest; | |
1757 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1758 | if (vma->rb_subtree_gap < length) | |
1759 | goto check_highest; | |
1760 | ||
1761 | while (true) { | |
1762 | /* Visit left subtree if it looks promising */ | |
1763 | gap_end = vma->vm_start; | |
1764 | if (gap_end >= low_limit && vma->vm_rb.rb_left) { | |
1765 | struct vm_area_struct *left = | |
1766 | rb_entry(vma->vm_rb.rb_left, | |
1767 | struct vm_area_struct, vm_rb); | |
1768 | if (left->rb_subtree_gap >= length) { | |
1769 | vma = left; | |
1770 | continue; | |
1771 | } | |
1772 | } | |
1773 | ||
1774 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1775 | check_current: | |
1776 | /* Check if current node has a suitable gap */ | |
1777 | if (gap_start > high_limit) | |
1778 | return -ENOMEM; | |
1779 | if (gap_end >= low_limit && gap_end - gap_start >= length) | |
1780 | goto found; | |
1781 | ||
1782 | /* Visit right subtree if it looks promising */ | |
1783 | if (vma->vm_rb.rb_right) { | |
1784 | struct vm_area_struct *right = | |
1785 | rb_entry(vma->vm_rb.rb_right, | |
1786 | struct vm_area_struct, vm_rb); | |
1787 | if (right->rb_subtree_gap >= length) { | |
1788 | vma = right; | |
1789 | continue; | |
1790 | } | |
1791 | } | |
1792 | ||
1793 | /* Go back up the rbtree to find next candidate node */ | |
1794 | while (true) { | |
1795 | struct rb_node *prev = &vma->vm_rb; | |
1796 | if (!rb_parent(prev)) | |
1797 | goto check_highest; | |
1798 | vma = rb_entry(rb_parent(prev), | |
1799 | struct vm_area_struct, vm_rb); | |
1800 | if (prev == vma->vm_rb.rb_left) { | |
1801 | gap_start = vma->vm_prev->vm_end; | |
1802 | gap_end = vma->vm_start; | |
1803 | goto check_current; | |
1804 | } | |
1805 | } | |
1806 | } | |
1807 | ||
1808 | check_highest: | |
1809 | /* Check highest gap, which does not precede any rbtree node */ | |
1810 | gap_start = mm->highest_vm_end; | |
1811 | gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ | |
1812 | if (gap_start > high_limit) | |
1813 | return -ENOMEM; | |
1814 | ||
1815 | found: | |
1816 | /* We found a suitable gap. Clip it with the original low_limit. */ | |
1817 | if (gap_start < info->low_limit) | |
1818 | gap_start = info->low_limit; | |
1819 | ||
1820 | /* Adjust gap address to the desired alignment */ | |
1821 | gap_start += (info->align_offset - gap_start) & info->align_mask; | |
1822 | ||
1823 | VM_BUG_ON(gap_start + info->length > info->high_limit); | |
1824 | VM_BUG_ON(gap_start + info->length > gap_end); | |
1825 | return gap_start; | |
1826 | } | |
1827 | ||
1828 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) | |
1829 | { | |
1830 | struct mm_struct *mm = current->mm; | |
1831 | struct vm_area_struct *vma; | |
1832 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1833 | ||
1834 | /* Adjust search length to account for worst case alignment overhead */ | |
1835 | length = info->length + info->align_mask; | |
1836 | if (length < info->length) | |
1837 | return -ENOMEM; | |
1838 | ||
1839 | /* | |
1840 | * Adjust search limits by the desired length. | |
1841 | * See implementation comment at top of unmapped_area(). | |
1842 | */ | |
1843 | gap_end = info->high_limit; | |
1844 | if (gap_end < length) | |
1845 | return -ENOMEM; | |
1846 | high_limit = gap_end - length; | |
1847 | ||
1848 | if (info->low_limit > high_limit) | |
1849 | return -ENOMEM; | |
1850 | low_limit = info->low_limit + length; | |
1851 | ||
1852 | /* Check highest gap, which does not precede any rbtree node */ | |
1853 | gap_start = mm->highest_vm_end; | |
1854 | if (gap_start <= high_limit) | |
1855 | goto found_highest; | |
1856 | ||
1857 | /* Check if rbtree root looks promising */ | |
1858 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1859 | return -ENOMEM; | |
1860 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1861 | if (vma->rb_subtree_gap < length) | |
1862 | return -ENOMEM; | |
1863 | ||
1864 | while (true) { | |
1865 | /* Visit right subtree if it looks promising */ | |
1866 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1867 | if (gap_start <= high_limit && vma->vm_rb.rb_right) { | |
1868 | struct vm_area_struct *right = | |
1869 | rb_entry(vma->vm_rb.rb_right, | |
1870 | struct vm_area_struct, vm_rb); | |
1871 | if (right->rb_subtree_gap >= length) { | |
1872 | vma = right; | |
1873 | continue; | |
1874 | } | |
1875 | } | |
1876 | ||
1877 | check_current: | |
1878 | /* Check if current node has a suitable gap */ | |
1879 | gap_end = vma->vm_start; | |
1880 | if (gap_end < low_limit) | |
1881 | return -ENOMEM; | |
1882 | if (gap_start <= high_limit && gap_end - gap_start >= length) | |
1883 | goto found; | |
1884 | ||
1885 | /* Visit left subtree if it looks promising */ | |
1886 | if (vma->vm_rb.rb_left) { | |
1887 | struct vm_area_struct *left = | |
1888 | rb_entry(vma->vm_rb.rb_left, | |
1889 | struct vm_area_struct, vm_rb); | |
1890 | if (left->rb_subtree_gap >= length) { | |
1891 | vma = left; | |
1892 | continue; | |
1893 | } | |
1894 | } | |
1895 | ||
1896 | /* Go back up the rbtree to find next candidate node */ | |
1897 | while (true) { | |
1898 | struct rb_node *prev = &vma->vm_rb; | |
1899 | if (!rb_parent(prev)) | |
1900 | return -ENOMEM; | |
1901 | vma = rb_entry(rb_parent(prev), | |
1902 | struct vm_area_struct, vm_rb); | |
1903 | if (prev == vma->vm_rb.rb_right) { | |
1904 | gap_start = vma->vm_prev ? | |
1905 | vma->vm_prev->vm_end : 0; | |
1906 | goto check_current; | |
1907 | } | |
1908 | } | |
1909 | } | |
1910 | ||
1911 | found: | |
1912 | /* We found a suitable gap. Clip it with the original high_limit. */ | |
1913 | if (gap_end > info->high_limit) | |
1914 | gap_end = info->high_limit; | |
1915 | ||
1916 | found_highest: | |
1917 | /* Compute highest gap address at the desired alignment */ | |
1918 | gap_end -= info->length; | |
1919 | gap_end -= (gap_end - info->align_offset) & info->align_mask; | |
1920 | ||
1921 | VM_BUG_ON(gap_end < info->low_limit); | |
1922 | VM_BUG_ON(gap_end < gap_start); | |
1923 | return gap_end; | |
1924 | } | |
1925 | ||
1da177e4 LT |
1926 | /* Get an address range which is currently unmapped. |
1927 | * For shmat() with addr=0. | |
1928 | * | |
1929 | * Ugly calling convention alert: | |
1930 | * Return value with the low bits set means error value, | |
1931 | * ie | |
1932 | * if (ret & ~PAGE_MASK) | |
1933 | * error = ret; | |
1934 | * | |
1935 | * This function "knows" that -ENOMEM has the bits set. | |
1936 | */ | |
1937 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1938 | unsigned long | |
1939 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1940 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1941 | { | |
1942 | struct mm_struct *mm = current->mm; | |
1943 | struct vm_area_struct *vma; | |
db4fbfb9 | 1944 | struct vm_unmapped_area_info info; |
1da177e4 | 1945 | |
2afc745f | 1946 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1947 | return -ENOMEM; |
1948 | ||
06abdfb4 BH |
1949 | if (flags & MAP_FIXED) |
1950 | return addr; | |
1951 | ||
1da177e4 LT |
1952 | if (addr) { |
1953 | addr = PAGE_ALIGN(addr); | |
1954 | vma = find_vma(mm, addr); | |
2afc745f | 1955 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1956 | (!vma || addr + len <= vma->vm_start)) |
1957 | return addr; | |
1958 | } | |
1da177e4 | 1959 | |
db4fbfb9 ML |
1960 | info.flags = 0; |
1961 | info.length = len; | |
4e99b021 | 1962 | info.low_limit = mm->mmap_base; |
db4fbfb9 ML |
1963 | info.high_limit = TASK_SIZE; |
1964 | info.align_mask = 0; | |
1965 | return vm_unmapped_area(&info); | |
1da177e4 | 1966 | } |
cc71aba3 | 1967 | #endif |
1da177e4 | 1968 | |
1da177e4 LT |
1969 | /* |
1970 | * This mmap-allocator allocates new areas top-down from below the | |
1971 | * stack's low limit (the base): | |
1972 | */ | |
1973 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1974 | unsigned long | |
1975 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1976 | const unsigned long len, const unsigned long pgoff, | |
1977 | const unsigned long flags) | |
1978 | { | |
1979 | struct vm_area_struct *vma; | |
1980 | struct mm_struct *mm = current->mm; | |
db4fbfb9 ML |
1981 | unsigned long addr = addr0; |
1982 | struct vm_unmapped_area_info info; | |
1da177e4 LT |
1983 | |
1984 | /* requested length too big for entire address space */ | |
2afc745f | 1985 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1986 | return -ENOMEM; |
1987 | ||
06abdfb4 BH |
1988 | if (flags & MAP_FIXED) |
1989 | return addr; | |
1990 | ||
1da177e4 LT |
1991 | /* requesting a specific address */ |
1992 | if (addr) { | |
1993 | addr = PAGE_ALIGN(addr); | |
1994 | vma = find_vma(mm, addr); | |
2afc745f | 1995 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1996 | (!vma || addr + len <= vma->vm_start)) |
1997 | return addr; | |
1998 | } | |
1999 | ||
db4fbfb9 ML |
2000 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
2001 | info.length = len; | |
2afc745f | 2002 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); |
db4fbfb9 ML |
2003 | info.high_limit = mm->mmap_base; |
2004 | info.align_mask = 0; | |
2005 | addr = vm_unmapped_area(&info); | |
b716ad95 | 2006 | |
1da177e4 LT |
2007 | /* |
2008 | * A failed mmap() very likely causes application failure, | |
2009 | * so fall back to the bottom-up function here. This scenario | |
2010 | * can happen with large stack limits and large mmap() | |
2011 | * allocations. | |
2012 | */ | |
de1741a1 | 2013 | if (offset_in_page(addr)) { |
db4fbfb9 ML |
2014 | VM_BUG_ON(addr != -ENOMEM); |
2015 | info.flags = 0; | |
2016 | info.low_limit = TASK_UNMAPPED_BASE; | |
2017 | info.high_limit = TASK_SIZE; | |
2018 | addr = vm_unmapped_area(&info); | |
2019 | } | |
1da177e4 LT |
2020 | |
2021 | return addr; | |
2022 | } | |
2023 | #endif | |
2024 | ||
1da177e4 LT |
2025 | unsigned long |
2026 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
2027 | unsigned long pgoff, unsigned long flags) | |
2028 | { | |
06abdfb4 BH |
2029 | unsigned long (*get_area)(struct file *, unsigned long, |
2030 | unsigned long, unsigned long, unsigned long); | |
2031 | ||
9206de95 AV |
2032 | unsigned long error = arch_mmap_check(addr, len, flags); |
2033 | if (error) | |
2034 | return error; | |
2035 | ||
2036 | /* Careful about overflows.. */ | |
2037 | if (len > TASK_SIZE) | |
2038 | return -ENOMEM; | |
2039 | ||
06abdfb4 | 2040 | get_area = current->mm->get_unmapped_area; |
c01d5b30 HD |
2041 | if (file) { |
2042 | if (file->f_op->get_unmapped_area) | |
2043 | get_area = file->f_op->get_unmapped_area; | |
2044 | } else if (flags & MAP_SHARED) { | |
2045 | /* | |
2046 | * mmap_region() will call shmem_zero_setup() to create a file, | |
2047 | * so use shmem's get_unmapped_area in case it can be huge. | |
2048 | * do_mmap_pgoff() will clear pgoff, so match alignment. | |
2049 | */ | |
2050 | pgoff = 0; | |
2051 | get_area = shmem_get_unmapped_area; | |
2052 | } | |
2053 | ||
06abdfb4 BH |
2054 | addr = get_area(file, addr, len, pgoff, flags); |
2055 | if (IS_ERR_VALUE(addr)) | |
2056 | return addr; | |
1da177e4 | 2057 | |
07ab67c8 LT |
2058 | if (addr > TASK_SIZE - len) |
2059 | return -ENOMEM; | |
de1741a1 | 2060 | if (offset_in_page(addr)) |
07ab67c8 | 2061 | return -EINVAL; |
06abdfb4 | 2062 | |
9ac4ed4b AV |
2063 | error = security_mmap_addr(addr); |
2064 | return error ? error : addr; | |
1da177e4 LT |
2065 | } |
2066 | ||
2067 | EXPORT_SYMBOL(get_unmapped_area); | |
2068 | ||
2069 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
48aae425 | 2070 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2071 | { |
615d6e87 DB |
2072 | struct rb_node *rb_node; |
2073 | struct vm_area_struct *vma; | |
1da177e4 | 2074 | |
841e31e5 | 2075 | /* Check the cache first. */ |
615d6e87 DB |
2076 | vma = vmacache_find(mm, addr); |
2077 | if (likely(vma)) | |
2078 | return vma; | |
841e31e5 | 2079 | |
615d6e87 | 2080 | rb_node = mm->mm_rb.rb_node; |
841e31e5 | 2081 | |
615d6e87 DB |
2082 | while (rb_node) { |
2083 | struct vm_area_struct *tmp; | |
2084 | ||
2085 | tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
2086 | ||
2087 | if (tmp->vm_end > addr) { | |
2088 | vma = tmp; | |
2089 | if (tmp->vm_start <= addr) | |
2090 | break; | |
2091 | rb_node = rb_node->rb_left; | |
2092 | } else | |
2093 | rb_node = rb_node->rb_right; | |
1da177e4 | 2094 | } |
615d6e87 DB |
2095 | |
2096 | if (vma) | |
2097 | vmacache_update(addr, vma); | |
1da177e4 LT |
2098 | return vma; |
2099 | } | |
2100 | ||
2101 | EXPORT_SYMBOL(find_vma); | |
2102 | ||
6bd4837d KM |
2103 | /* |
2104 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. | |
6bd4837d | 2105 | */ |
1da177e4 LT |
2106 | struct vm_area_struct * |
2107 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
2108 | struct vm_area_struct **pprev) | |
2109 | { | |
6bd4837d | 2110 | struct vm_area_struct *vma; |
1da177e4 | 2111 | |
6bd4837d | 2112 | vma = find_vma(mm, addr); |
83cd904d MP |
2113 | if (vma) { |
2114 | *pprev = vma->vm_prev; | |
2115 | } else { | |
2116 | struct rb_node *rb_node = mm->mm_rb.rb_node; | |
2117 | *pprev = NULL; | |
2118 | while (rb_node) { | |
2119 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
2120 | rb_node = rb_node->rb_right; | |
2121 | } | |
2122 | } | |
6bd4837d | 2123 | return vma; |
1da177e4 LT |
2124 | } |
2125 | ||
2126 | /* | |
2127 | * Verify that the stack growth is acceptable and | |
2128 | * update accounting. This is shared with both the | |
2129 | * grow-up and grow-down cases. | |
2130 | */ | |
48aae425 | 2131 | static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) |
1da177e4 LT |
2132 | { |
2133 | struct mm_struct *mm = vma->vm_mm; | |
2134 | struct rlimit *rlim = current->signal->rlim; | |
690eac53 | 2135 | unsigned long new_start, actual_size; |
1da177e4 LT |
2136 | |
2137 | /* address space limit tests */ | |
84638335 | 2138 | if (!may_expand_vm(mm, vma->vm_flags, grow)) |
1da177e4 LT |
2139 | return -ENOMEM; |
2140 | ||
2141 | /* Stack limit test */ | |
690eac53 LT |
2142 | actual_size = size; |
2143 | if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) | |
2144 | actual_size -= PAGE_SIZE; | |
4db0c3c2 | 2145 | if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur)) |
1da177e4 LT |
2146 | return -ENOMEM; |
2147 | ||
2148 | /* mlock limit tests */ | |
2149 | if (vma->vm_flags & VM_LOCKED) { | |
2150 | unsigned long locked; | |
2151 | unsigned long limit; | |
2152 | locked = mm->locked_vm + grow; | |
4db0c3c2 | 2153 | limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); |
59e99e5b | 2154 | limit >>= PAGE_SHIFT; |
1da177e4 LT |
2155 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
2156 | return -ENOMEM; | |
2157 | } | |
2158 | ||
0d59a01b AL |
2159 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
2160 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
2161 | vma->vm_end - size; | |
2162 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
2163 | return -EFAULT; | |
2164 | ||
1da177e4 LT |
2165 | /* |
2166 | * Overcommit.. This must be the final test, as it will | |
2167 | * update security statistics. | |
2168 | */ | |
05fa199d | 2169 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
2170 | return -ENOMEM; |
2171 | ||
1da177e4 LT |
2172 | return 0; |
2173 | } | |
2174 | ||
46dea3d0 | 2175 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 2176 | /* |
46dea3d0 HD |
2177 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
2178 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 2179 | */ |
46dea3d0 | 2180 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 2181 | { |
09357814 | 2182 | struct mm_struct *mm = vma->vm_mm; |
12352d3c | 2183 | int error = 0; |
1da177e4 LT |
2184 | |
2185 | if (!(vma->vm_flags & VM_GROWSUP)) | |
2186 | return -EFAULT; | |
2187 | ||
12352d3c KK |
2188 | /* Guard against wrapping around to address 0. */ |
2189 | if (address < PAGE_ALIGN(address+4)) | |
2190 | address = PAGE_ALIGN(address+4); | |
2191 | else | |
2192 | return -ENOMEM; | |
2193 | ||
2194 | /* We must make sure the anon_vma is allocated. */ | |
1da177e4 LT |
2195 | if (unlikely(anon_vma_prepare(vma))) |
2196 | return -ENOMEM; | |
1da177e4 LT |
2197 | |
2198 | /* | |
2199 | * vma->vm_start/vm_end cannot change under us because the caller | |
2200 | * is required to hold the mmap_sem in read mode. We need the | |
2201 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2202 | */ | |
12352d3c | 2203 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2204 | |
2205 | /* Somebody else might have raced and expanded it already */ | |
2206 | if (address > vma->vm_end) { | |
2207 | unsigned long size, grow; | |
2208 | ||
2209 | size = address - vma->vm_start; | |
2210 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
2211 | ||
42c36f63 HD |
2212 | error = -ENOMEM; |
2213 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
2214 | error = acct_stack_growth(vma, size, grow); | |
2215 | if (!error) { | |
4128997b ML |
2216 | /* |
2217 | * vma_gap_update() doesn't support concurrent | |
2218 | * updates, but we only hold a shared mmap_sem | |
2219 | * lock here, so we need to protect against | |
2220 | * concurrent vma expansions. | |
12352d3c | 2221 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2222 | * we don't guarantee that all growable vmas |
2223 | * in a mm share the same root anon vma. | |
2224 | * So, we reuse mm->page_table_lock to guard | |
2225 | * against concurrent vma expansions. | |
2226 | */ | |
09357814 | 2227 | spin_lock(&mm->page_table_lock); |
87e8827b | 2228 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2229 | mm->locked_vm += grow; |
84638335 | 2230 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2231 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 2232 | vma->vm_end = address; |
bf181b9f | 2233 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 ML |
2234 | if (vma->vm_next) |
2235 | vma_gap_update(vma->vm_next); | |
2236 | else | |
09357814 ON |
2237 | mm->highest_vm_end = address; |
2238 | spin_unlock(&mm->page_table_lock); | |
4128997b | 2239 | |
42c36f63 HD |
2240 | perf_event_mmap(vma); |
2241 | } | |
3af9e859 | 2242 | } |
1da177e4 | 2243 | } |
12352d3c | 2244 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2245 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2246 | validate_mm(mm); |
1da177e4 LT |
2247 | return error; |
2248 | } | |
46dea3d0 HD |
2249 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2250 | ||
1da177e4 LT |
2251 | /* |
2252 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
2253 | */ | |
d05f3169 | 2254 | int expand_downwards(struct vm_area_struct *vma, |
b6a2fea3 | 2255 | unsigned long address) |
1da177e4 | 2256 | { |
09357814 | 2257 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 LT |
2258 | int error; |
2259 | ||
8869477a | 2260 | address &= PAGE_MASK; |
e5467859 | 2261 | error = security_mmap_addr(address); |
8869477a EP |
2262 | if (error) |
2263 | return error; | |
2264 | ||
12352d3c KK |
2265 | /* We must make sure the anon_vma is allocated. */ |
2266 | if (unlikely(anon_vma_prepare(vma))) | |
2267 | return -ENOMEM; | |
1da177e4 LT |
2268 | |
2269 | /* | |
2270 | * vma->vm_start/vm_end cannot change under us because the caller | |
2271 | * is required to hold the mmap_sem in read mode. We need the | |
2272 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2273 | */ | |
12352d3c | 2274 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2275 | |
2276 | /* Somebody else might have raced and expanded it already */ | |
2277 | if (address < vma->vm_start) { | |
2278 | unsigned long size, grow; | |
2279 | ||
2280 | size = vma->vm_end - address; | |
2281 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2282 | ||
a626ca6a LT |
2283 | error = -ENOMEM; |
2284 | if (grow <= vma->vm_pgoff) { | |
2285 | error = acct_stack_growth(vma, size, grow); | |
2286 | if (!error) { | |
4128997b ML |
2287 | /* |
2288 | * vma_gap_update() doesn't support concurrent | |
2289 | * updates, but we only hold a shared mmap_sem | |
2290 | * lock here, so we need to protect against | |
2291 | * concurrent vma expansions. | |
12352d3c | 2292 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2293 | * we don't guarantee that all growable vmas |
2294 | * in a mm share the same root anon vma. | |
2295 | * So, we reuse mm->page_table_lock to guard | |
2296 | * against concurrent vma expansions. | |
2297 | */ | |
09357814 | 2298 | spin_lock(&mm->page_table_lock); |
87e8827b | 2299 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2300 | mm->locked_vm += grow; |
84638335 | 2301 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2302 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2303 | vma->vm_start = address; |
2304 | vma->vm_pgoff -= grow; | |
bf181b9f | 2305 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 | 2306 | vma_gap_update(vma); |
09357814 | 2307 | spin_unlock(&mm->page_table_lock); |
4128997b | 2308 | |
a626ca6a LT |
2309 | perf_event_mmap(vma); |
2310 | } | |
1da177e4 LT |
2311 | } |
2312 | } | |
12352d3c | 2313 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2314 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2315 | validate_mm(mm); |
1da177e4 LT |
2316 | return error; |
2317 | } | |
2318 | ||
09884964 LT |
2319 | /* |
2320 | * Note how expand_stack() refuses to expand the stack all the way to | |
2321 | * abut the next virtual mapping, *unless* that mapping itself is also | |
2322 | * a stack mapping. We want to leave room for a guard page, after all | |
2323 | * (the guard page itself is not added here, that is done by the | |
2324 | * actual page faulting logic) | |
2325 | * | |
2326 | * This matches the behavior of the guard page logic (see mm/memory.c: | |
2327 | * check_stack_guard_page()), which only allows the guard page to be | |
2328 | * removed under these circumstances. | |
2329 | */ | |
b6a2fea3 OW |
2330 | #ifdef CONFIG_STACK_GROWSUP |
2331 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2332 | { | |
09884964 LT |
2333 | struct vm_area_struct *next; |
2334 | ||
2335 | address &= PAGE_MASK; | |
2336 | next = vma->vm_next; | |
2337 | if (next && next->vm_start == address + PAGE_SIZE) { | |
2338 | if (!(next->vm_flags & VM_GROWSUP)) | |
2339 | return -ENOMEM; | |
2340 | } | |
b6a2fea3 OW |
2341 | return expand_upwards(vma, address); |
2342 | } | |
2343 | ||
2344 | struct vm_area_struct * | |
2345 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2346 | { | |
2347 | struct vm_area_struct *vma, *prev; | |
2348 | ||
2349 | addr &= PAGE_MASK; | |
2350 | vma = find_vma_prev(mm, addr, &prev); | |
2351 | if (vma && (vma->vm_start <= addr)) | |
2352 | return vma; | |
1c127185 | 2353 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2354 | return NULL; |
cea10a19 | 2355 | if (prev->vm_flags & VM_LOCKED) |
fc05f566 | 2356 | populate_vma_page_range(prev, addr, prev->vm_end, NULL); |
b6a2fea3 OW |
2357 | return prev; |
2358 | } | |
2359 | #else | |
2360 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2361 | { | |
09884964 LT |
2362 | struct vm_area_struct *prev; |
2363 | ||
2364 | address &= PAGE_MASK; | |
2365 | prev = vma->vm_prev; | |
2366 | if (prev && prev->vm_end == address) { | |
2367 | if (!(prev->vm_flags & VM_GROWSDOWN)) | |
2368 | return -ENOMEM; | |
2369 | } | |
b6a2fea3 OW |
2370 | return expand_downwards(vma, address); |
2371 | } | |
2372 | ||
1da177e4 | 2373 | struct vm_area_struct * |
cc71aba3 | 2374 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2375 | { |
cc71aba3 | 2376 | struct vm_area_struct *vma; |
1da177e4 LT |
2377 | unsigned long start; |
2378 | ||
2379 | addr &= PAGE_MASK; | |
cc71aba3 | 2380 | vma = find_vma(mm, addr); |
1da177e4 LT |
2381 | if (!vma) |
2382 | return NULL; | |
2383 | if (vma->vm_start <= addr) | |
2384 | return vma; | |
2385 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2386 | return NULL; | |
2387 | start = vma->vm_start; | |
2388 | if (expand_stack(vma, addr)) | |
2389 | return NULL; | |
cea10a19 | 2390 | if (vma->vm_flags & VM_LOCKED) |
fc05f566 | 2391 | populate_vma_page_range(vma, addr, start, NULL); |
1da177e4 LT |
2392 | return vma; |
2393 | } | |
2394 | #endif | |
2395 | ||
e1d6d01a JB |
2396 | EXPORT_SYMBOL_GPL(find_extend_vma); |
2397 | ||
1da177e4 | 2398 | /* |
2c0b3814 | 2399 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 2400 | * so release them, and do the vma updates. |
2c0b3814 HD |
2401 | * |
2402 | * Called with the mm semaphore held. | |
1da177e4 | 2403 | */ |
2c0b3814 | 2404 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2405 | { |
4f74d2c8 LT |
2406 | unsigned long nr_accounted = 0; |
2407 | ||
365e9c87 HD |
2408 | /* Update high watermark before we lower total_vm */ |
2409 | update_hiwater_vm(mm); | |
1da177e4 | 2410 | do { |
2c0b3814 HD |
2411 | long nrpages = vma_pages(vma); |
2412 | ||
4f74d2c8 LT |
2413 | if (vma->vm_flags & VM_ACCOUNT) |
2414 | nr_accounted += nrpages; | |
84638335 | 2415 | vm_stat_account(mm, vma->vm_flags, -nrpages); |
a8fb5618 | 2416 | vma = remove_vma(vma); |
146425a3 | 2417 | } while (vma); |
4f74d2c8 | 2418 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2419 | validate_mm(mm); |
2420 | } | |
2421 | ||
2422 | /* | |
2423 | * Get rid of page table information in the indicated region. | |
2424 | * | |
f10df686 | 2425 | * Called with the mm semaphore held. |
1da177e4 LT |
2426 | */ |
2427 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
2428 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
2429 | unsigned long start, unsigned long end) | |
1da177e4 | 2430 | { |
cc71aba3 | 2431 | struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; |
d16dfc55 | 2432 | struct mmu_gather tlb; |
1da177e4 LT |
2433 | |
2434 | lru_add_drain(); | |
2b047252 | 2435 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 2436 | update_hiwater_rss(mm); |
4f74d2c8 | 2437 | unmap_vmas(&tlb, vma, start, end); |
d16dfc55 | 2438 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
6ee8630e | 2439 | next ? next->vm_start : USER_PGTABLES_CEILING); |
d16dfc55 | 2440 | tlb_finish_mmu(&tlb, start, end); |
1da177e4 LT |
2441 | } |
2442 | ||
2443 | /* | |
2444 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
2445 | * vma list as we go.. | |
2446 | */ | |
2447 | static void | |
2448 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
2449 | struct vm_area_struct *prev, unsigned long end) | |
2450 | { | |
2451 | struct vm_area_struct **insertion_point; | |
2452 | struct vm_area_struct *tail_vma = NULL; | |
2453 | ||
2454 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
297c5eee | 2455 | vma->vm_prev = NULL; |
1da177e4 | 2456 | do { |
d3737187 | 2457 | vma_rb_erase(vma, &mm->mm_rb); |
1da177e4 LT |
2458 | mm->map_count--; |
2459 | tail_vma = vma; | |
2460 | vma = vma->vm_next; | |
2461 | } while (vma && vma->vm_start < end); | |
2462 | *insertion_point = vma; | |
d3737187 | 2463 | if (vma) { |
297c5eee | 2464 | vma->vm_prev = prev; |
d3737187 ML |
2465 | vma_gap_update(vma); |
2466 | } else | |
2467 | mm->highest_vm_end = prev ? prev->vm_end : 0; | |
1da177e4 | 2468 | tail_vma->vm_next = NULL; |
615d6e87 DB |
2469 | |
2470 | /* Kill the cache */ | |
2471 | vmacache_invalidate(mm); | |
1da177e4 LT |
2472 | } |
2473 | ||
2474 | /* | |
659ace58 KM |
2475 | * __split_vma() bypasses sysctl_max_map_count checking. We use this on the |
2476 | * munmap path where it doesn't make sense to fail. | |
1da177e4 | 2477 | */ |
cc71aba3 | 2478 | static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
2479 | unsigned long addr, int new_below) |
2480 | { | |
1da177e4 | 2481 | struct vm_area_struct *new; |
e3975891 | 2482 | int err; |
1da177e4 | 2483 | |
a5516438 AK |
2484 | if (is_vm_hugetlb_page(vma) && (addr & |
2485 | ~(huge_page_mask(hstate_vma(vma))))) | |
1da177e4 LT |
2486 | return -EINVAL; |
2487 | ||
e94b1766 | 2488 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 | 2489 | if (!new) |
e3975891 | 2490 | return -ENOMEM; |
1da177e4 LT |
2491 | |
2492 | /* most fields are the same, copy all, and then fixup */ | |
2493 | *new = *vma; | |
2494 | ||
5beb4930 RR |
2495 | INIT_LIST_HEAD(&new->anon_vma_chain); |
2496 | ||
1da177e4 LT |
2497 | if (new_below) |
2498 | new->vm_end = addr; | |
2499 | else { | |
2500 | new->vm_start = addr; | |
2501 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2502 | } | |
2503 | ||
ef0855d3 ON |
2504 | err = vma_dup_policy(vma, new); |
2505 | if (err) | |
5beb4930 | 2506 | goto out_free_vma; |
1da177e4 | 2507 | |
c4ea95d7 DF |
2508 | err = anon_vma_clone(new, vma); |
2509 | if (err) | |
5beb4930 RR |
2510 | goto out_free_mpol; |
2511 | ||
e9714acf | 2512 | if (new->vm_file) |
1da177e4 LT |
2513 | get_file(new->vm_file); |
2514 | ||
2515 | if (new->vm_ops && new->vm_ops->open) | |
2516 | new->vm_ops->open(new); | |
2517 | ||
2518 | if (new_below) | |
5beb4930 | 2519 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
1da177e4 LT |
2520 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
2521 | else | |
5beb4930 | 2522 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
1da177e4 | 2523 | |
5beb4930 RR |
2524 | /* Success. */ |
2525 | if (!err) | |
2526 | return 0; | |
2527 | ||
2528 | /* Clean everything up if vma_adjust failed. */ | |
58927533 RR |
2529 | if (new->vm_ops && new->vm_ops->close) |
2530 | new->vm_ops->close(new); | |
e9714acf | 2531 | if (new->vm_file) |
5beb4930 | 2532 | fput(new->vm_file); |
2aeadc30 | 2533 | unlink_anon_vmas(new); |
5beb4930 | 2534 | out_free_mpol: |
ef0855d3 | 2535 | mpol_put(vma_policy(new)); |
5beb4930 RR |
2536 | out_free_vma: |
2537 | kmem_cache_free(vm_area_cachep, new); | |
5beb4930 | 2538 | return err; |
1da177e4 LT |
2539 | } |
2540 | ||
659ace58 KM |
2541 | /* |
2542 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2543 | * either for the first part or the tail. | |
2544 | */ | |
2545 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, | |
2546 | unsigned long addr, int new_below) | |
2547 | { | |
2548 | if (mm->map_count >= sysctl_max_map_count) | |
2549 | return -ENOMEM; | |
2550 | ||
2551 | return __split_vma(mm, vma, addr, new_below); | |
2552 | } | |
2553 | ||
1da177e4 LT |
2554 | /* Munmap is split into 2 main parts -- this part which finds |
2555 | * what needs doing, and the areas themselves, which do the | |
2556 | * work. This now handles partial unmappings. | |
2557 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
2558 | */ | |
2559 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
2560 | { | |
2561 | unsigned long end; | |
146425a3 | 2562 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 | 2563 | |
de1741a1 | 2564 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) |
1da177e4 LT |
2565 | return -EINVAL; |
2566 | ||
cc71aba3 | 2567 | len = PAGE_ALIGN(len); |
2568 | if (len == 0) | |
1da177e4 LT |
2569 | return -EINVAL; |
2570 | ||
2571 | /* Find the first overlapping VMA */ | |
9be34c9d | 2572 | vma = find_vma(mm, start); |
146425a3 | 2573 | if (!vma) |
1da177e4 | 2574 | return 0; |
9be34c9d | 2575 | prev = vma->vm_prev; |
146425a3 | 2576 | /* we have start < vma->vm_end */ |
1da177e4 LT |
2577 | |
2578 | /* if it doesn't overlap, we have nothing.. */ | |
2579 | end = start + len; | |
146425a3 | 2580 | if (vma->vm_start >= end) |
1da177e4 LT |
2581 | return 0; |
2582 | ||
2583 | /* | |
2584 | * If we need to split any vma, do it now to save pain later. | |
2585 | * | |
2586 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2587 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2588 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2589 | */ | |
146425a3 | 2590 | if (start > vma->vm_start) { |
659ace58 KM |
2591 | int error; |
2592 | ||
2593 | /* | |
2594 | * Make sure that map_count on return from munmap() will | |
2595 | * not exceed its limit; but let map_count go just above | |
2596 | * its limit temporarily, to help free resources as expected. | |
2597 | */ | |
2598 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
2599 | return -ENOMEM; | |
2600 | ||
2601 | error = __split_vma(mm, vma, start, 0); | |
1da177e4 LT |
2602 | if (error) |
2603 | return error; | |
146425a3 | 2604 | prev = vma; |
1da177e4 LT |
2605 | } |
2606 | ||
2607 | /* Does it split the last one? */ | |
2608 | last = find_vma(mm, end); | |
2609 | if (last && end > last->vm_start) { | |
659ace58 | 2610 | int error = __split_vma(mm, last, end, 1); |
1da177e4 LT |
2611 | if (error) |
2612 | return error; | |
2613 | } | |
cc71aba3 | 2614 | vma = prev ? prev->vm_next : mm->mmap; |
1da177e4 | 2615 | |
ba470de4 RR |
2616 | /* |
2617 | * unlock any mlock()ed ranges before detaching vmas | |
2618 | */ | |
2619 | if (mm->locked_vm) { | |
2620 | struct vm_area_struct *tmp = vma; | |
2621 | while (tmp && tmp->vm_start < end) { | |
2622 | if (tmp->vm_flags & VM_LOCKED) { | |
2623 | mm->locked_vm -= vma_pages(tmp); | |
2624 | munlock_vma_pages_all(tmp); | |
2625 | } | |
2626 | tmp = tmp->vm_next; | |
2627 | } | |
2628 | } | |
2629 | ||
1da177e4 LT |
2630 | /* |
2631 | * Remove the vma's, and unmap the actual pages | |
2632 | */ | |
146425a3 HD |
2633 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
2634 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 | 2635 | |
1de4fa14 DH |
2636 | arch_unmap(mm, vma, start, end); |
2637 | ||
1da177e4 | 2638 | /* Fix up all other VM information */ |
2c0b3814 | 2639 | remove_vma_list(mm, vma); |
1da177e4 LT |
2640 | |
2641 | return 0; | |
2642 | } | |
1da177e4 | 2643 | |
bfce281c | 2644 | int vm_munmap(unsigned long start, size_t len) |
1da177e4 LT |
2645 | { |
2646 | int ret; | |
bfce281c | 2647 | struct mm_struct *mm = current->mm; |
1da177e4 | 2648 | |
ae798783 MH |
2649 | if (down_write_killable(&mm->mmap_sem)) |
2650 | return -EINTR; | |
2651 | ||
a46ef99d | 2652 | ret = do_munmap(mm, start, len); |
1da177e4 LT |
2653 | up_write(&mm->mmap_sem); |
2654 | return ret; | |
2655 | } | |
a46ef99d LT |
2656 | EXPORT_SYMBOL(vm_munmap); |
2657 | ||
2658 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2659 | { | |
dc0ef0df MH |
2660 | int ret; |
2661 | struct mm_struct *mm = current->mm; | |
2662 | ||
a46ef99d | 2663 | profile_munmap(addr); |
dc0ef0df MH |
2664 | if (down_write_killable(&mm->mmap_sem)) |
2665 | return -EINTR; | |
2666 | ret = do_munmap(mm, addr, len); | |
2667 | up_write(&mm->mmap_sem); | |
2668 | return ret; | |
a46ef99d | 2669 | } |
1da177e4 | 2670 | |
c8d78c18 KS |
2671 | |
2672 | /* | |
2673 | * Emulation of deprecated remap_file_pages() syscall. | |
2674 | */ | |
2675 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, | |
2676 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) | |
2677 | { | |
2678 | ||
2679 | struct mm_struct *mm = current->mm; | |
2680 | struct vm_area_struct *vma; | |
2681 | unsigned long populate = 0; | |
2682 | unsigned long ret = -EINVAL; | |
2683 | struct file *file; | |
2684 | ||
756a025f JP |
2685 | pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n", |
2686 | current->comm, current->pid); | |
c8d78c18 KS |
2687 | |
2688 | if (prot) | |
2689 | return ret; | |
2690 | start = start & PAGE_MASK; | |
2691 | size = size & PAGE_MASK; | |
2692 | ||
2693 | if (start + size <= start) | |
2694 | return ret; | |
2695 | ||
2696 | /* Does pgoff wrap? */ | |
2697 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) | |
2698 | return ret; | |
2699 | ||
dc0ef0df MH |
2700 | if (down_write_killable(&mm->mmap_sem)) |
2701 | return -EINTR; | |
2702 | ||
c8d78c18 KS |
2703 | vma = find_vma(mm, start); |
2704 | ||
2705 | if (!vma || !(vma->vm_flags & VM_SHARED)) | |
2706 | goto out; | |
2707 | ||
48f7df32 | 2708 | if (start < vma->vm_start) |
c8d78c18 KS |
2709 | goto out; |
2710 | ||
48f7df32 KS |
2711 | if (start + size > vma->vm_end) { |
2712 | struct vm_area_struct *next; | |
2713 | ||
2714 | for (next = vma->vm_next; next; next = next->vm_next) { | |
2715 | /* hole between vmas ? */ | |
2716 | if (next->vm_start != next->vm_prev->vm_end) | |
2717 | goto out; | |
2718 | ||
2719 | if (next->vm_file != vma->vm_file) | |
2720 | goto out; | |
2721 | ||
2722 | if (next->vm_flags != vma->vm_flags) | |
2723 | goto out; | |
2724 | ||
2725 | if (start + size <= next->vm_end) | |
2726 | break; | |
2727 | } | |
2728 | ||
2729 | if (!next) | |
2730 | goto out; | |
c8d78c18 KS |
2731 | } |
2732 | ||
2733 | prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; | |
2734 | prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; | |
2735 | prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; | |
2736 | ||
2737 | flags &= MAP_NONBLOCK; | |
2738 | flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; | |
2739 | if (vma->vm_flags & VM_LOCKED) { | |
48f7df32 | 2740 | struct vm_area_struct *tmp; |
c8d78c18 | 2741 | flags |= MAP_LOCKED; |
48f7df32 | 2742 | |
c8d78c18 | 2743 | /* drop PG_Mlocked flag for over-mapped range */ |
48f7df32 KS |
2744 | for (tmp = vma; tmp->vm_start >= start + size; |
2745 | tmp = tmp->vm_next) { | |
9a73f61b KS |
2746 | /* |
2747 | * Split pmd and munlock page on the border | |
2748 | * of the range. | |
2749 | */ | |
2750 | vma_adjust_trans_huge(tmp, start, start + size, 0); | |
2751 | ||
48f7df32 KS |
2752 | munlock_vma_pages_range(tmp, |
2753 | max(tmp->vm_start, start), | |
2754 | min(tmp->vm_end, start + size)); | |
2755 | } | |
c8d78c18 KS |
2756 | } |
2757 | ||
2758 | file = get_file(vma->vm_file); | |
2759 | ret = do_mmap_pgoff(vma->vm_file, start, size, | |
2760 | prot, flags, pgoff, &populate); | |
2761 | fput(file); | |
2762 | out: | |
2763 | up_write(&mm->mmap_sem); | |
2764 | if (populate) | |
2765 | mm_populate(ret, populate); | |
2766 | if (!IS_ERR_VALUE(ret)) | |
2767 | ret = 0; | |
2768 | return ret; | |
2769 | } | |
2770 | ||
1da177e4 LT |
2771 | static inline void verify_mm_writelocked(struct mm_struct *mm) |
2772 | { | |
a241ec65 | 2773 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
2774 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
2775 | WARN_ON(1); | |
2776 | up_read(&mm->mmap_sem); | |
2777 | } | |
2778 | #endif | |
2779 | } | |
2780 | ||
2781 | /* | |
2782 | * this is really a simplified "do_mmap". it only handles | |
2783 | * anonymous maps. eventually we may be able to do some | |
2784 | * brk-specific accounting here. | |
2785 | */ | |
ba093a6d | 2786 | static int do_brk(unsigned long addr, unsigned long request) |
1da177e4 | 2787 | { |
cc71aba3 | 2788 | struct mm_struct *mm = current->mm; |
2789 | struct vm_area_struct *vma, *prev; | |
ba093a6d | 2790 | unsigned long flags, len; |
cc71aba3 | 2791 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 2792 | pgoff_t pgoff = addr >> PAGE_SHIFT; |
3a459756 | 2793 | int error; |
1da177e4 | 2794 | |
ba093a6d KC |
2795 | len = PAGE_ALIGN(request); |
2796 | if (len < request) | |
2797 | return -ENOMEM; | |
1da177e4 | 2798 | if (!len) |
5d22fc25 | 2799 | return 0; |
1da177e4 | 2800 | |
3a459756 KK |
2801 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; |
2802 | ||
2c6a1016 | 2803 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
de1741a1 | 2804 | if (offset_in_page(error)) |
3a459756 KK |
2805 | return error; |
2806 | ||
363ee17f DB |
2807 | error = mlock_future_check(mm, mm->def_flags, len); |
2808 | if (error) | |
2809 | return error; | |
1da177e4 LT |
2810 | |
2811 | /* | |
2812 | * mm->mmap_sem is required to protect against another thread | |
2813 | * changing the mappings in case we sleep. | |
2814 | */ | |
2815 | verify_mm_writelocked(mm); | |
2816 | ||
2817 | /* | |
2818 | * Clear old maps. this also does some error checking for us | |
2819 | */ | |
9fcd1457 RV |
2820 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
2821 | &rb_parent)) { | |
1da177e4 LT |
2822 | if (do_munmap(mm, addr, len)) |
2823 | return -ENOMEM; | |
1da177e4 LT |
2824 | } |
2825 | ||
2826 | /* Check against address space limits *after* clearing old maps... */ | |
84638335 | 2827 | if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) |
1da177e4 LT |
2828 | return -ENOMEM; |
2829 | ||
2830 | if (mm->map_count > sysctl_max_map_count) | |
2831 | return -ENOMEM; | |
2832 | ||
191c5424 | 2833 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2834 | return -ENOMEM; |
2835 | ||
1da177e4 | 2836 | /* Can we just expand an old private anonymous mapping? */ |
ba470de4 | 2837 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
19a809af | 2838 | NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); |
ba470de4 | 2839 | if (vma) |
1da177e4 LT |
2840 | goto out; |
2841 | ||
2842 | /* | |
2843 | * create a vma struct for an anonymous mapping | |
2844 | */ | |
c5e3b83e | 2845 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2846 | if (!vma) { |
2847 | vm_unacct_memory(len >> PAGE_SHIFT); | |
2848 | return -ENOMEM; | |
2849 | } | |
1da177e4 | 2850 | |
5beb4930 | 2851 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
2852 | vma->vm_mm = mm; |
2853 | vma->vm_start = addr; | |
2854 | vma->vm_end = addr + len; | |
2855 | vma->vm_pgoff = pgoff; | |
2856 | vma->vm_flags = flags; | |
3ed75eb8 | 2857 | vma->vm_page_prot = vm_get_page_prot(flags); |
1da177e4 LT |
2858 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2859 | out: | |
3af9e859 | 2860 | perf_event_mmap(vma); |
1da177e4 | 2861 | mm->total_vm += len >> PAGE_SHIFT; |
84638335 | 2862 | mm->data_vm += len >> PAGE_SHIFT; |
128557ff ML |
2863 | if (flags & VM_LOCKED) |
2864 | mm->locked_vm += (len >> PAGE_SHIFT); | |
d9104d1c | 2865 | vma->vm_flags |= VM_SOFTDIRTY; |
5d22fc25 | 2866 | return 0; |
1da177e4 LT |
2867 | } |
2868 | ||
5d22fc25 | 2869 | int vm_brk(unsigned long addr, unsigned long len) |
e4eb1ff6 LT |
2870 | { |
2871 | struct mm_struct *mm = current->mm; | |
5d22fc25 | 2872 | int ret; |
128557ff | 2873 | bool populate; |
e4eb1ff6 | 2874 | |
2d6c9282 MH |
2875 | if (down_write_killable(&mm->mmap_sem)) |
2876 | return -EINTR; | |
2877 | ||
e4eb1ff6 | 2878 | ret = do_brk(addr, len); |
128557ff | 2879 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
e4eb1ff6 | 2880 | up_write(&mm->mmap_sem); |
5d22fc25 | 2881 | if (populate && !ret) |
128557ff | 2882 | mm_populate(addr, len); |
e4eb1ff6 LT |
2883 | return ret; |
2884 | } | |
2885 | EXPORT_SYMBOL(vm_brk); | |
1da177e4 LT |
2886 | |
2887 | /* Release all mmaps. */ | |
2888 | void exit_mmap(struct mm_struct *mm) | |
2889 | { | |
d16dfc55 | 2890 | struct mmu_gather tlb; |
ba470de4 | 2891 | struct vm_area_struct *vma; |
1da177e4 LT |
2892 | unsigned long nr_accounted = 0; |
2893 | ||
d6dd61c8 | 2894 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 2895 | mmu_notifier_release(mm); |
d6dd61c8 | 2896 | |
ba470de4 RR |
2897 | if (mm->locked_vm) { |
2898 | vma = mm->mmap; | |
2899 | while (vma) { | |
2900 | if (vma->vm_flags & VM_LOCKED) | |
2901 | munlock_vma_pages_all(vma); | |
2902 | vma = vma->vm_next; | |
2903 | } | |
2904 | } | |
9480c53e JF |
2905 | |
2906 | arch_exit_mmap(mm); | |
2907 | ||
ba470de4 | 2908 | vma = mm->mmap; |
9480c53e JF |
2909 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
2910 | return; | |
2911 | ||
1da177e4 | 2912 | lru_add_drain(); |
1da177e4 | 2913 | flush_cache_mm(mm); |
2b047252 | 2914 | tlb_gather_mmu(&tlb, mm, 0, -1); |
901608d9 | 2915 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
e0da382c | 2916 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
4f74d2c8 | 2917 | unmap_vmas(&tlb, vma, 0, -1); |
9ba69294 | 2918 | |
6ee8630e | 2919 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); |
853f5e26 | 2920 | tlb_finish_mmu(&tlb, 0, -1); |
1da177e4 | 2921 | |
1da177e4 | 2922 | /* |
8f4f8c16 HD |
2923 | * Walk the list again, actually closing and freeing it, |
2924 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 2925 | */ |
4f74d2c8 LT |
2926 | while (vma) { |
2927 | if (vma->vm_flags & VM_ACCOUNT) | |
2928 | nr_accounted += vma_pages(vma); | |
a8fb5618 | 2929 | vma = remove_vma(vma); |
4f74d2c8 LT |
2930 | } |
2931 | vm_unacct_memory(nr_accounted); | |
1da177e4 LT |
2932 | } |
2933 | ||
2934 | /* Insert vm structure into process list sorted by address | |
2935 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
c8c06efa | 2936 | * then i_mmap_rwsem is taken here. |
1da177e4 | 2937 | */ |
6597d783 | 2938 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2939 | { |
6597d783 HD |
2940 | struct vm_area_struct *prev; |
2941 | struct rb_node **rb_link, *rb_parent; | |
1da177e4 | 2942 | |
c9d13f5f CG |
2943 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
2944 | &prev, &rb_link, &rb_parent)) | |
2945 | return -ENOMEM; | |
2946 | if ((vma->vm_flags & VM_ACCOUNT) && | |
2947 | security_vm_enough_memory_mm(mm, vma_pages(vma))) | |
2948 | return -ENOMEM; | |
2949 | ||
1da177e4 LT |
2950 | /* |
2951 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
2952 | * until its first write fault, when page's anon_vma and index | |
2953 | * are set. But now set the vm_pgoff it will almost certainly | |
2954 | * end up with (unless mremap moves it elsewhere before that | |
2955 | * first wfault), so /proc/pid/maps tells a consistent story. | |
2956 | * | |
2957 | * By setting it to reflect the virtual start address of the | |
2958 | * vma, merges and splits can happen in a seamless way, just | |
2959 | * using the existing file pgoff checks and manipulations. | |
2960 | * Similarly in do_mmap_pgoff and in do_brk. | |
2961 | */ | |
8a9cc3b5 | 2962 | if (vma_is_anonymous(vma)) { |
1da177e4 LT |
2963 | BUG_ON(vma->anon_vma); |
2964 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
2965 | } | |
2b144498 | 2966 | |
1da177e4 LT |
2967 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2968 | return 0; | |
2969 | } | |
2970 | ||
2971 | /* | |
2972 | * Copy the vma structure to a new location in the same mm, | |
2973 | * prior to moving page table entries, to effect an mremap move. | |
2974 | */ | |
2975 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
2976 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
2977 | bool *need_rmap_locks) | |
1da177e4 LT |
2978 | { |
2979 | struct vm_area_struct *vma = *vmap; | |
2980 | unsigned long vma_start = vma->vm_start; | |
2981 | struct mm_struct *mm = vma->vm_mm; | |
2982 | struct vm_area_struct *new_vma, *prev; | |
2983 | struct rb_node **rb_link, *rb_parent; | |
948f017b | 2984 | bool faulted_in_anon_vma = true; |
1da177e4 LT |
2985 | |
2986 | /* | |
2987 | * If anonymous vma has not yet been faulted, update new pgoff | |
2988 | * to match new location, to increase its chance of merging. | |
2989 | */ | |
ce75799b | 2990 | if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { |
1da177e4 | 2991 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
2992 | faulted_in_anon_vma = false; |
2993 | } | |
1da177e4 | 2994 | |
6597d783 HD |
2995 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
2996 | return NULL; /* should never get here */ | |
1da177e4 | 2997 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
19a809af AA |
2998 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
2999 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
3000 | if (new_vma) { |
3001 | /* | |
3002 | * Source vma may have been merged into new_vma | |
3003 | */ | |
948f017b AA |
3004 | if (unlikely(vma_start >= new_vma->vm_start && |
3005 | vma_start < new_vma->vm_end)) { | |
3006 | /* | |
3007 | * The only way we can get a vma_merge with | |
3008 | * self during an mremap is if the vma hasn't | |
3009 | * been faulted in yet and we were allowed to | |
3010 | * reset the dst vma->vm_pgoff to the | |
3011 | * destination address of the mremap to allow | |
3012 | * the merge to happen. mremap must change the | |
3013 | * vm_pgoff linearity between src and dst vmas | |
3014 | * (in turn preventing a vma_merge) to be | |
3015 | * safe. It is only safe to keep the vm_pgoff | |
3016 | * linear if there are no pages mapped yet. | |
3017 | */ | |
81d1b09c | 3018 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
38a76013 | 3019 | *vmap = vma = new_vma; |
108d6642 | 3020 | } |
38a76013 | 3021 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 3022 | } else { |
e94b1766 | 3023 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
e3975891 CG |
3024 | if (!new_vma) |
3025 | goto out; | |
3026 | *new_vma = *vma; | |
3027 | new_vma->vm_start = addr; | |
3028 | new_vma->vm_end = addr + len; | |
3029 | new_vma->vm_pgoff = pgoff; | |
3030 | if (vma_dup_policy(vma, new_vma)) | |
3031 | goto out_free_vma; | |
3032 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); | |
3033 | if (anon_vma_clone(new_vma, vma)) | |
3034 | goto out_free_mempol; | |
3035 | if (new_vma->vm_file) | |
3036 | get_file(new_vma->vm_file); | |
3037 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
3038 | new_vma->vm_ops->open(new_vma); | |
3039 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
3040 | *need_rmap_locks = false; | |
1da177e4 LT |
3041 | } |
3042 | return new_vma; | |
5beb4930 | 3043 | |
e3975891 | 3044 | out_free_mempol: |
ef0855d3 | 3045 | mpol_put(vma_policy(new_vma)); |
e3975891 | 3046 | out_free_vma: |
5beb4930 | 3047 | kmem_cache_free(vm_area_cachep, new_vma); |
e3975891 | 3048 | out: |
5beb4930 | 3049 | return NULL; |
1da177e4 | 3050 | } |
119f657c | 3051 | |
3052 | /* | |
3053 | * Return true if the calling process may expand its vm space by the passed | |
3054 | * number of pages | |
3055 | */ | |
84638335 | 3056 | bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) |
119f657c | 3057 | { |
84638335 KK |
3058 | if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) |
3059 | return false; | |
119f657c | 3060 | |
d977d56c KK |
3061 | if (is_data_mapping(flags) && |
3062 | mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { | |
f4fcd558 KK |
3063 | /* Workaround for Valgrind */ |
3064 | if (rlimit(RLIMIT_DATA) == 0 && | |
3065 | mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) | |
3066 | return true; | |
3067 | if (!ignore_rlimit_data) { | |
3068 | pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n", | |
d977d56c KK |
3069 | current->comm, current->pid, |
3070 | (mm->data_vm + npages) << PAGE_SHIFT, | |
3071 | rlimit(RLIMIT_DATA)); | |
d977d56c | 3072 | return false; |
f4fcd558 | 3073 | } |
d977d56c | 3074 | } |
119f657c | 3075 | |
84638335 KK |
3076 | return true; |
3077 | } | |
3078 | ||
3079 | void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) | |
3080 | { | |
3081 | mm->total_vm += npages; | |
3082 | ||
d977d56c | 3083 | if (is_exec_mapping(flags)) |
84638335 | 3084 | mm->exec_vm += npages; |
d977d56c | 3085 | else if (is_stack_mapping(flags)) |
84638335 | 3086 | mm->stack_vm += npages; |
d977d56c | 3087 | else if (is_data_mapping(flags)) |
84638335 | 3088 | mm->data_vm += npages; |
119f657c | 3089 | } |
fa5dc22f | 3090 | |
a62c34bd AL |
3091 | static int special_mapping_fault(struct vm_area_struct *vma, |
3092 | struct vm_fault *vmf); | |
3093 | ||
3094 | /* | |
3095 | * Having a close hook prevents vma merging regardless of flags. | |
3096 | */ | |
3097 | static void special_mapping_close(struct vm_area_struct *vma) | |
3098 | { | |
3099 | } | |
3100 | ||
3101 | static const char *special_mapping_name(struct vm_area_struct *vma) | |
3102 | { | |
3103 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; | |
3104 | } | |
3105 | ||
b059a453 DS |
3106 | static int special_mapping_mremap(struct vm_area_struct *new_vma) |
3107 | { | |
3108 | struct vm_special_mapping *sm = new_vma->vm_private_data; | |
3109 | ||
3110 | if (sm->mremap) | |
3111 | return sm->mremap(sm, new_vma); | |
3112 | return 0; | |
3113 | } | |
3114 | ||
a62c34bd AL |
3115 | static const struct vm_operations_struct special_mapping_vmops = { |
3116 | .close = special_mapping_close, | |
3117 | .fault = special_mapping_fault, | |
b059a453 | 3118 | .mremap = special_mapping_mremap, |
a62c34bd AL |
3119 | .name = special_mapping_name, |
3120 | }; | |
3121 | ||
3122 | static const struct vm_operations_struct legacy_special_mapping_vmops = { | |
3123 | .close = special_mapping_close, | |
3124 | .fault = special_mapping_fault, | |
3125 | }; | |
fa5dc22f | 3126 | |
b1d0e4f5 NP |
3127 | static int special_mapping_fault(struct vm_area_struct *vma, |
3128 | struct vm_fault *vmf) | |
fa5dc22f | 3129 | { |
b1d0e4f5 | 3130 | pgoff_t pgoff; |
fa5dc22f RM |
3131 | struct page **pages; |
3132 | ||
f872f540 | 3133 | if (vma->vm_ops == &legacy_special_mapping_vmops) { |
a62c34bd | 3134 | pages = vma->vm_private_data; |
f872f540 AL |
3135 | } else { |
3136 | struct vm_special_mapping *sm = vma->vm_private_data; | |
3137 | ||
3138 | if (sm->fault) | |
3139 | return sm->fault(sm, vma, vmf); | |
3140 | ||
3141 | pages = sm->pages; | |
3142 | } | |
a62c34bd | 3143 | |
8a9cc3b5 | 3144 | for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) |
b1d0e4f5 | 3145 | pgoff--; |
fa5dc22f RM |
3146 | |
3147 | if (*pages) { | |
3148 | struct page *page = *pages; | |
3149 | get_page(page); | |
b1d0e4f5 NP |
3150 | vmf->page = page; |
3151 | return 0; | |
fa5dc22f RM |
3152 | } |
3153 | ||
b1d0e4f5 | 3154 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
3155 | } |
3156 | ||
a62c34bd AL |
3157 | static struct vm_area_struct *__install_special_mapping( |
3158 | struct mm_struct *mm, | |
3159 | unsigned long addr, unsigned long len, | |
27f28b97 CG |
3160 | unsigned long vm_flags, void *priv, |
3161 | const struct vm_operations_struct *ops) | |
fa5dc22f | 3162 | { |
462e635e | 3163 | int ret; |
fa5dc22f RM |
3164 | struct vm_area_struct *vma; |
3165 | ||
3166 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
3167 | if (unlikely(vma == NULL)) | |
3935ed6a | 3168 | return ERR_PTR(-ENOMEM); |
fa5dc22f | 3169 | |
5beb4930 | 3170 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
fa5dc22f RM |
3171 | vma->vm_mm = mm; |
3172 | vma->vm_start = addr; | |
3173 | vma->vm_end = addr + len; | |
3174 | ||
d9104d1c | 3175 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; |
3ed75eb8 | 3176 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f | 3177 | |
a62c34bd AL |
3178 | vma->vm_ops = ops; |
3179 | vma->vm_private_data = priv; | |
fa5dc22f | 3180 | |
462e635e TO |
3181 | ret = insert_vm_struct(mm, vma); |
3182 | if (ret) | |
3183 | goto out; | |
fa5dc22f | 3184 | |
84638335 | 3185 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); |
fa5dc22f | 3186 | |
cdd6c482 | 3187 | perf_event_mmap(vma); |
089dd79d | 3188 | |
3935ed6a | 3189 | return vma; |
462e635e TO |
3190 | |
3191 | out: | |
3192 | kmem_cache_free(vm_area_cachep, vma); | |
3935ed6a SS |
3193 | return ERR_PTR(ret); |
3194 | } | |
3195 | ||
2eefd878 DS |
3196 | bool vma_is_special_mapping(const struct vm_area_struct *vma, |
3197 | const struct vm_special_mapping *sm) | |
3198 | { | |
3199 | return vma->vm_private_data == sm && | |
3200 | (vma->vm_ops == &special_mapping_vmops || | |
3201 | vma->vm_ops == &legacy_special_mapping_vmops); | |
3202 | } | |
3203 | ||
a62c34bd AL |
3204 | /* |
3205 | * Called with mm->mmap_sem held for writing. | |
3206 | * Insert a new vma covering the given region, with the given flags. | |
3207 | * Its pages are supplied by the given array of struct page *. | |
3208 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
3209 | * The region past the last page supplied will always produce SIGBUS. | |
3210 | * The array pointer and the pages it points to are assumed to stay alive | |
3211 | * for as long as this mapping might exist. | |
3212 | */ | |
3213 | struct vm_area_struct *_install_special_mapping( | |
3214 | struct mm_struct *mm, | |
3215 | unsigned long addr, unsigned long len, | |
3216 | unsigned long vm_flags, const struct vm_special_mapping *spec) | |
3217 | { | |
27f28b97 CG |
3218 | return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, |
3219 | &special_mapping_vmops); | |
a62c34bd AL |
3220 | } |
3221 | ||
3935ed6a SS |
3222 | int install_special_mapping(struct mm_struct *mm, |
3223 | unsigned long addr, unsigned long len, | |
3224 | unsigned long vm_flags, struct page **pages) | |
3225 | { | |
a62c34bd | 3226 | struct vm_area_struct *vma = __install_special_mapping( |
27f28b97 CG |
3227 | mm, addr, len, vm_flags, (void *)pages, |
3228 | &legacy_special_mapping_vmops); | |
3935ed6a | 3229 | |
14bd5b45 | 3230 | return PTR_ERR_OR_ZERO(vma); |
fa5dc22f | 3231 | } |
7906d00c AA |
3232 | |
3233 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3234 | ||
454ed842 | 3235 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3236 | { |
bf181b9f | 3237 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3238 | /* |
3239 | * The LSB of head.next can't change from under us | |
3240 | * because we hold the mm_all_locks_mutex. | |
3241 | */ | |
572043c9 | 3242 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); |
7906d00c AA |
3243 | /* |
3244 | * We can safely modify head.next after taking the | |
5a505085 | 3245 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3246 | * the same anon_vma we won't take it again. |
3247 | * | |
3248 | * No need of atomic instructions here, head.next | |
3249 | * can't change from under us thanks to the | |
5a505085 | 3250 | * anon_vma->root->rwsem. |
7906d00c AA |
3251 | */ |
3252 | if (__test_and_set_bit(0, (unsigned long *) | |
bf181b9f | 3253 | &anon_vma->root->rb_root.rb_node)) |
7906d00c AA |
3254 | BUG(); |
3255 | } | |
3256 | } | |
3257 | ||
454ed842 | 3258 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3259 | { |
3260 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3261 | /* | |
3262 | * AS_MM_ALL_LOCKS can't change from under us because | |
3263 | * we hold the mm_all_locks_mutex. | |
3264 | * | |
3265 | * Operations on ->flags have to be atomic because | |
3266 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3267 | * mm_all_locks_mutex, there may be other cpus | |
3268 | * changing other bitflags in parallel to us. | |
3269 | */ | |
3270 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3271 | BUG(); | |
c8c06efa | 3272 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); |
7906d00c AA |
3273 | } |
3274 | } | |
3275 | ||
3276 | /* | |
3277 | * This operation locks against the VM for all pte/vma/mm related | |
3278 | * operations that could ever happen on a certain mm. This includes | |
3279 | * vmtruncate, try_to_unmap, and all page faults. | |
3280 | * | |
3281 | * The caller must take the mmap_sem in write mode before calling | |
3282 | * mm_take_all_locks(). The caller isn't allowed to release the | |
3283 | * mmap_sem until mm_drop_all_locks() returns. | |
3284 | * | |
3285 | * mmap_sem in write mode is required in order to block all operations | |
3286 | * that could modify pagetables and free pages without need of | |
27ba0644 | 3287 | * altering the vma layout. It's also needed in write mode to avoid new |
7906d00c AA |
3288 | * anon_vmas to be associated with existing vmas. |
3289 | * | |
3290 | * A single task can't take more than one mm_take_all_locks() in a row | |
3291 | * or it would deadlock. | |
3292 | * | |
bf181b9f | 3293 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3294 | * mapping->flags avoid to take the same lock twice, if more than one |
3295 | * vma in this mm is backed by the same anon_vma or address_space. | |
3296 | * | |
88f306b6 KS |
3297 | * We take locks in following order, accordingly to comment at beginning |
3298 | * of mm/rmap.c: | |
3299 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for | |
3300 | * hugetlb mapping); | |
3301 | * - all i_mmap_rwsem locks; | |
3302 | * - all anon_vma->rwseml | |
3303 | * | |
3304 | * We can take all locks within these types randomly because the VM code | |
3305 | * doesn't nest them and we protected from parallel mm_take_all_locks() by | |
3306 | * mm_all_locks_mutex. | |
7906d00c AA |
3307 | * |
3308 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3309 | * that may have to take thousand of locks. | |
3310 | * | |
3311 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3312 | */ | |
3313 | int mm_take_all_locks(struct mm_struct *mm) | |
3314 | { | |
3315 | struct vm_area_struct *vma; | |
5beb4930 | 3316 | struct anon_vma_chain *avc; |
7906d00c AA |
3317 | |
3318 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3319 | ||
3320 | mutex_lock(&mm_all_locks_mutex); | |
3321 | ||
3322 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3323 | if (signal_pending(current)) | |
3324 | goto out_unlock; | |
88f306b6 KS |
3325 | if (vma->vm_file && vma->vm_file->f_mapping && |
3326 | is_vm_hugetlb_page(vma)) | |
3327 | vm_lock_mapping(mm, vma->vm_file->f_mapping); | |
3328 | } | |
3329 | ||
3330 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3331 | if (signal_pending(current)) | |
3332 | goto out_unlock; | |
3333 | if (vma->vm_file && vma->vm_file->f_mapping && | |
3334 | !is_vm_hugetlb_page(vma)) | |
454ed842 | 3335 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3336 | } |
7cd5a02f PZ |
3337 | |
3338 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3339 | if (signal_pending(current)) | |
3340 | goto out_unlock; | |
3341 | if (vma->anon_vma) | |
5beb4930 RR |
3342 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3343 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3344 | } |
7cd5a02f | 3345 | |
584cff54 | 3346 | return 0; |
7906d00c AA |
3347 | |
3348 | out_unlock: | |
584cff54 KC |
3349 | mm_drop_all_locks(mm); |
3350 | return -EINTR; | |
7906d00c AA |
3351 | } |
3352 | ||
3353 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3354 | { | |
bf181b9f | 3355 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3356 | /* |
3357 | * The LSB of head.next can't change to 0 from under | |
3358 | * us because we hold the mm_all_locks_mutex. | |
3359 | * | |
3360 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3361 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3362 | * never see our bitflag. |
3363 | * | |
3364 | * No need of atomic instructions here, head.next | |
3365 | * can't change from under us until we release the | |
5a505085 | 3366 | * anon_vma->root->rwsem. |
7906d00c AA |
3367 | */ |
3368 | if (!__test_and_clear_bit(0, (unsigned long *) | |
bf181b9f | 3369 | &anon_vma->root->rb_root.rb_node)) |
7906d00c | 3370 | BUG(); |
08b52706 | 3371 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3372 | } |
3373 | } | |
3374 | ||
3375 | static void vm_unlock_mapping(struct address_space *mapping) | |
3376 | { | |
3377 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3378 | /* | |
3379 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3380 | * because we hold the mm_all_locks_mutex. | |
3381 | */ | |
83cde9e8 | 3382 | i_mmap_unlock_write(mapping); |
7906d00c AA |
3383 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3384 | &mapping->flags)) | |
3385 | BUG(); | |
3386 | } | |
3387 | } | |
3388 | ||
3389 | /* | |
3390 | * The mmap_sem cannot be released by the caller until | |
3391 | * mm_drop_all_locks() returns. | |
3392 | */ | |
3393 | void mm_drop_all_locks(struct mm_struct *mm) | |
3394 | { | |
3395 | struct vm_area_struct *vma; | |
5beb4930 | 3396 | struct anon_vma_chain *avc; |
7906d00c AA |
3397 | |
3398 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3399 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); | |
3400 | ||
3401 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3402 | if (vma->anon_vma) | |
5beb4930 RR |
3403 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3404 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3405 | if (vma->vm_file && vma->vm_file->f_mapping) |
3406 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3407 | } | |
3408 | ||
3409 | mutex_unlock(&mm_all_locks_mutex); | |
3410 | } | |
8feae131 DH |
3411 | |
3412 | /* | |
3413 | * initialise the VMA slab | |
3414 | */ | |
3415 | void __init mmap_init(void) | |
3416 | { | |
00a62ce9 KM |
3417 | int ret; |
3418 | ||
908c7f19 | 3419 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 3420 | VM_BUG_ON(ret); |
8feae131 | 3421 | } |
c9b1d098 AS |
3422 | |
3423 | /* | |
3424 | * Initialise sysctl_user_reserve_kbytes. | |
3425 | * | |
3426 | * This is intended to prevent a user from starting a single memory hogging | |
3427 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3428 | * mode. | |
3429 | * | |
3430 | * The default value is min(3% of free memory, 128MB) | |
3431 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3432 | */ | |
1640879a | 3433 | static int init_user_reserve(void) |
c9b1d098 AS |
3434 | { |
3435 | unsigned long free_kbytes; | |
3436 | ||
3437 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3438 | ||
3439 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3440 | return 0; | |
3441 | } | |
a64fb3cd | 3442 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
3443 | |
3444 | /* | |
3445 | * Initialise sysctl_admin_reserve_kbytes. | |
3446 | * | |
3447 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3448 | * to log in and kill a memory hogging process. | |
3449 | * | |
3450 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3451 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3452 | * only reserve 3% of free pages by default. | |
3453 | */ | |
1640879a | 3454 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3455 | { |
3456 | unsigned long free_kbytes; | |
3457 | ||
3458 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3459 | ||
3460 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3461 | return 0; | |
3462 | } | |
a64fb3cd | 3463 | subsys_initcall(init_admin_reserve); |
1640879a AS |
3464 | |
3465 | /* | |
3466 | * Reinititalise user and admin reserves if memory is added or removed. | |
3467 | * | |
3468 | * The default user reserve max is 128MB, and the default max for the | |
3469 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3470 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3471 | * and tools like top. It may make sense to increase or even disable the | |
3472 | * reserve depending on the existence of swap or variations in the recovery | |
3473 | * tools. So, the admin may have changed them. | |
3474 | * | |
3475 | * If memory is added and the reserves have been eliminated or increased above | |
3476 | * the default max, then we'll trust the admin. | |
3477 | * | |
3478 | * If memory is removed and there isn't enough free memory, then we | |
3479 | * need to reset the reserves. | |
3480 | * | |
3481 | * Otherwise keep the reserve set by the admin. | |
3482 | */ | |
3483 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3484 | unsigned long action, void *data) | |
3485 | { | |
3486 | unsigned long tmp, free_kbytes; | |
3487 | ||
3488 | switch (action) { | |
3489 | case MEM_ONLINE: | |
3490 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3491 | tmp = sysctl_user_reserve_kbytes; | |
3492 | if (0 < tmp && tmp < (1UL << 17)) | |
3493 | init_user_reserve(); | |
3494 | ||
3495 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3496 | tmp = sysctl_admin_reserve_kbytes; | |
3497 | if (0 < tmp && tmp < (1UL << 13)) | |
3498 | init_admin_reserve(); | |
3499 | ||
3500 | break; | |
3501 | case MEM_OFFLINE: | |
3502 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3503 | ||
3504 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3505 | init_user_reserve(); | |
3506 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3507 | sysctl_user_reserve_kbytes); | |
3508 | } | |
3509 | ||
3510 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3511 | init_admin_reserve(); | |
3512 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3513 | sysctl_admin_reserve_kbytes); | |
3514 | } | |
3515 | break; | |
3516 | default: | |
3517 | break; | |
3518 | } | |
3519 | return NOTIFY_OK; | |
3520 | } | |
3521 | ||
3522 | static struct notifier_block reserve_mem_nb = { | |
3523 | .notifier_call = reserve_mem_notifier, | |
3524 | }; | |
3525 | ||
3526 | static int __meminit init_reserve_notifier(void) | |
3527 | { | |
3528 | if (register_hotmemory_notifier(&reserve_mem_nb)) | |
b1de0d13 | 3529 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
1640879a AS |
3530 | |
3531 | return 0; | |
3532 | } | |
a64fb3cd | 3533 | subsys_initcall(init_reserve_notifier); |