]>
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> |
21292580 | 47 | #include <linux/oom.h> |
1da177e4 | 48 | |
7c0f6ba6 | 49 | #include <linux/uaccess.h> |
1da177e4 LT |
50 | #include <asm/cacheflush.h> |
51 | #include <asm/tlb.h> | |
d6dd61c8 | 52 | #include <asm/mmu_context.h> |
1da177e4 | 53 | |
42b77728 JB |
54 | #include "internal.h" |
55 | ||
3a459756 KK |
56 | #ifndef arch_mmap_check |
57 | #define arch_mmap_check(addr, len, flags) (0) | |
58 | #endif | |
59 | ||
d07e2259 DC |
60 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS |
61 | const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; | |
62 | const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; | |
63 | int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; | |
64 | #endif | |
65 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS | |
66 | const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; | |
67 | const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; | |
68 | int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; | |
69 | #endif | |
70 | ||
f4fcd558 | 71 | static bool ignore_rlimit_data; |
d977d56c | 72 | core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); |
d07e2259 | 73 | |
e0da382c HD |
74 | static void unmap_region(struct mm_struct *mm, |
75 | struct vm_area_struct *vma, struct vm_area_struct *prev, | |
76 | unsigned long start, unsigned long end); | |
77 | ||
1da177e4 LT |
78 | /* description of effects of mapping type and prot in current implementation. |
79 | * this is due to the limited x86 page protection hardware. The expected | |
80 | * behavior is in parens: | |
81 | * | |
82 | * map_type prot | |
83 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC | |
84 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
85 | * w: (no) no w: (no) no w: (yes) yes w: (no) no | |
86 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
cc71aba3 | 87 | * |
1da177e4 LT |
88 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes |
89 | * w: (no) no w: (no) no w: (copy) copy w: (no) no | |
90 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
91 | * | |
cab15ce6 CM |
92 | * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and |
93 | * MAP_PRIVATE: | |
94 | * r: (no) no | |
95 | * w: (no) no | |
96 | * x: (yes) yes | |
1da177e4 | 97 | */ |
ac34ceaf | 98 | pgprot_t protection_map[16] __ro_after_init = { |
1da177e4 LT |
99 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, |
100 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
101 | }; | |
102 | ||
804af2cf HD |
103 | pgprot_t vm_get_page_prot(unsigned long vm_flags) |
104 | { | |
b845f313 DK |
105 | return __pgprot(pgprot_val(protection_map[vm_flags & |
106 | (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | | |
107 | pgprot_val(arch_vm_get_page_prot(vm_flags))); | |
804af2cf HD |
108 | } |
109 | EXPORT_SYMBOL(vm_get_page_prot); | |
110 | ||
64e45507 PF |
111 | static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) |
112 | { | |
113 | return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); | |
114 | } | |
115 | ||
116 | /* Update vma->vm_page_prot to reflect vma->vm_flags. */ | |
117 | void vma_set_page_prot(struct vm_area_struct *vma) | |
118 | { | |
119 | unsigned long vm_flags = vma->vm_flags; | |
6d2329f8 | 120 | pgprot_t vm_page_prot; |
64e45507 | 121 | |
6d2329f8 AA |
122 | vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); |
123 | if (vma_wants_writenotify(vma, vm_page_prot)) { | |
64e45507 | 124 | vm_flags &= ~VM_SHARED; |
6d2329f8 | 125 | vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); |
64e45507 | 126 | } |
6d2329f8 AA |
127 | /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */ |
128 | WRITE_ONCE(vma->vm_page_prot, vm_page_prot); | |
64e45507 PF |
129 | } |
130 | ||
1da177e4 | 131 | /* |
c8c06efa | 132 | * Requires inode->i_mapping->i_mmap_rwsem |
1da177e4 LT |
133 | */ |
134 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
135 | struct file *file, struct address_space *mapping) | |
136 | { | |
137 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 138 | atomic_inc(&file_inode(file)->i_writecount); |
1da177e4 | 139 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 140 | mapping_unmap_writable(mapping); |
1da177e4 LT |
141 | |
142 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 143 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
144 | flush_dcache_mmap_unlock(mapping); |
145 | } | |
146 | ||
147 | /* | |
6b2dbba8 | 148 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 149 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 150 | */ |
a8fb5618 | 151 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
152 | { |
153 | struct file *file = vma->vm_file; | |
154 | ||
1da177e4 LT |
155 | if (file) { |
156 | struct address_space *mapping = file->f_mapping; | |
83cde9e8 | 157 | i_mmap_lock_write(mapping); |
1da177e4 | 158 | __remove_shared_vm_struct(vma, file, mapping); |
83cde9e8 | 159 | i_mmap_unlock_write(mapping); |
1da177e4 | 160 | } |
a8fb5618 HD |
161 | } |
162 | ||
163 | /* | |
164 | * Close a vm structure and free it, returning the next. | |
165 | */ | |
166 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
167 | { | |
168 | struct vm_area_struct *next = vma->vm_next; | |
169 | ||
a8fb5618 | 170 | might_sleep(); |
1da177e4 LT |
171 | if (vma->vm_ops && vma->vm_ops->close) |
172 | vma->vm_ops->close(vma); | |
e9714acf | 173 | if (vma->vm_file) |
a8fb5618 | 174 | fput(vma->vm_file); |
f0be3d32 | 175 | mpol_put(vma_policy(vma)); |
1da177e4 | 176 | kmem_cache_free(vm_area_cachep, vma); |
a8fb5618 | 177 | return next; |
1da177e4 LT |
178 | } |
179 | ||
897ab3e0 | 180 | static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf); |
e4eb1ff6 | 181 | |
6a6160a7 | 182 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 | 183 | { |
8764b338 | 184 | unsigned long retval; |
1da177e4 LT |
185 | unsigned long newbrk, oldbrk; |
186 | struct mm_struct *mm = current->mm; | |
1be7107f | 187 | struct vm_area_struct *next; |
a5b4592c | 188 | unsigned long min_brk; |
128557ff | 189 | bool populate; |
897ab3e0 | 190 | LIST_HEAD(uf); |
1da177e4 | 191 | |
dc0ef0df MH |
192 | if (down_write_killable(&mm->mmap_sem)) |
193 | return -EINTR; | |
1da177e4 | 194 | |
a5b4592c | 195 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
196 | /* |
197 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
198 | * randomize_va_space to 2, which will still cause mm->start_brk | |
199 | * to be arbitrarily shifted | |
200 | */ | |
4471a675 | 201 | if (current->brk_randomized) |
5520e894 JK |
202 | min_brk = mm->start_brk; |
203 | else | |
204 | min_brk = mm->end_data; | |
a5b4592c JK |
205 | #else |
206 | min_brk = mm->start_brk; | |
207 | #endif | |
208 | if (brk < min_brk) | |
1da177e4 | 209 | goto out; |
1e624196 RG |
210 | |
211 | /* | |
212 | * Check against rlimit here. If this check is done later after the test | |
213 | * of oldbrk with newbrk then it can escape the test and let the data | |
214 | * segment grow beyond its set limit the in case where the limit is | |
215 | * not page aligned -Ram Gupta | |
216 | */ | |
8764b338 CG |
217 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
218 | mm->end_data, mm->start_data)) | |
1e624196 RG |
219 | goto out; |
220 | ||
1da177e4 LT |
221 | newbrk = PAGE_ALIGN(brk); |
222 | oldbrk = PAGE_ALIGN(mm->brk); | |
223 | if (oldbrk == newbrk) | |
224 | goto set_brk; | |
225 | ||
226 | /* Always allow shrinking brk. */ | |
227 | if (brk <= mm->brk) { | |
897ab3e0 | 228 | if (!do_munmap(mm, newbrk, oldbrk-newbrk, &uf)) |
1da177e4 LT |
229 | goto set_brk; |
230 | goto out; | |
231 | } | |
232 | ||
1da177e4 | 233 | /* Check against existing mmap mappings. */ |
1be7107f HD |
234 | next = find_vma(mm, oldbrk); |
235 | if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) | |
1da177e4 LT |
236 | goto out; |
237 | ||
238 | /* Ok, looks good - let it rip. */ | |
897ab3e0 | 239 | if (do_brk(oldbrk, newbrk-oldbrk, &uf) < 0) |
1da177e4 | 240 | goto out; |
128557ff | 241 | |
1da177e4 LT |
242 | set_brk: |
243 | mm->brk = brk; | |
128557ff ML |
244 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
245 | up_write(&mm->mmap_sem); | |
897ab3e0 | 246 | userfaultfd_unmap_complete(mm, &uf); |
128557ff ML |
247 | if (populate) |
248 | mm_populate(oldbrk, newbrk - oldbrk); | |
249 | return brk; | |
250 | ||
1da177e4 LT |
251 | out: |
252 | retval = mm->brk; | |
253 | up_write(&mm->mmap_sem); | |
254 | return retval; | |
255 | } | |
256 | ||
d3737187 ML |
257 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
258 | { | |
1be7107f HD |
259 | unsigned long max, prev_end, subtree_gap; |
260 | ||
261 | /* | |
262 | * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we | |
263 | * allow two stack_guard_gaps between them here, and when choosing | |
264 | * an unmapped area; whereas when expanding we only require one. | |
265 | * That's a little inconsistent, but keeps the code here simpler. | |
266 | */ | |
267 | max = vm_start_gap(vma); | |
268 | if (vma->vm_prev) { | |
269 | prev_end = vm_end_gap(vma->vm_prev); | |
270 | if (max > prev_end) | |
271 | max -= prev_end; | |
272 | else | |
273 | max = 0; | |
274 | } | |
d3737187 ML |
275 | if (vma->vm_rb.rb_left) { |
276 | subtree_gap = rb_entry(vma->vm_rb.rb_left, | |
277 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
278 | if (subtree_gap > max) | |
279 | max = subtree_gap; | |
280 | } | |
281 | if (vma->vm_rb.rb_right) { | |
282 | subtree_gap = rb_entry(vma->vm_rb.rb_right, | |
283 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
284 | if (subtree_gap > max) | |
285 | max = subtree_gap; | |
286 | } | |
287 | return max; | |
288 | } | |
289 | ||
ed8ea815 | 290 | #ifdef CONFIG_DEBUG_VM_RB |
acf128d0 | 291 | static int browse_rb(struct mm_struct *mm) |
1da177e4 | 292 | { |
acf128d0 | 293 | struct rb_root *root = &mm->mm_rb; |
5a0768f6 | 294 | int i = 0, j, bug = 0; |
1da177e4 LT |
295 | struct rb_node *nd, *pn = NULL; |
296 | unsigned long prev = 0, pend = 0; | |
297 | ||
298 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
299 | struct vm_area_struct *vma; | |
300 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
5a0768f6 | 301 | if (vma->vm_start < prev) { |
ff26f70f AM |
302 | pr_emerg("vm_start %lx < prev %lx\n", |
303 | vma->vm_start, prev); | |
5a0768f6 ML |
304 | bug = 1; |
305 | } | |
306 | if (vma->vm_start < pend) { | |
ff26f70f AM |
307 | pr_emerg("vm_start %lx < pend %lx\n", |
308 | vma->vm_start, pend); | |
5a0768f6 ML |
309 | bug = 1; |
310 | } | |
311 | if (vma->vm_start > vma->vm_end) { | |
ff26f70f AM |
312 | pr_emerg("vm_start %lx > vm_end %lx\n", |
313 | vma->vm_start, vma->vm_end); | |
5a0768f6 ML |
314 | bug = 1; |
315 | } | |
acf128d0 | 316 | spin_lock(&mm->page_table_lock); |
5a0768f6 | 317 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { |
8542bdfc | 318 | pr_emerg("free gap %lx, correct %lx\n", |
5a0768f6 ML |
319 | vma->rb_subtree_gap, |
320 | vma_compute_subtree_gap(vma)); | |
321 | bug = 1; | |
322 | } | |
acf128d0 | 323 | spin_unlock(&mm->page_table_lock); |
1da177e4 LT |
324 | i++; |
325 | pn = nd; | |
d1af65d1 DM |
326 | prev = vma->vm_start; |
327 | pend = vma->vm_end; | |
1da177e4 LT |
328 | } |
329 | j = 0; | |
5a0768f6 | 330 | for (nd = pn; nd; nd = rb_prev(nd)) |
1da177e4 | 331 | j++; |
5a0768f6 | 332 | if (i != j) { |
8542bdfc | 333 | pr_emerg("backwards %d, forwards %d\n", j, i); |
5a0768f6 | 334 | bug = 1; |
1da177e4 | 335 | } |
5a0768f6 | 336 | return bug ? -1 : i; |
1da177e4 LT |
337 | } |
338 | ||
d3737187 ML |
339 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
340 | { | |
341 | struct rb_node *nd; | |
342 | ||
343 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
344 | struct vm_area_struct *vma; | |
345 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
96dad67f SL |
346 | VM_BUG_ON_VMA(vma != ignore && |
347 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma), | |
348 | vma); | |
1da177e4 | 349 | } |
1da177e4 LT |
350 | } |
351 | ||
eafd4dc4 | 352 | static void validate_mm(struct mm_struct *mm) |
1da177e4 LT |
353 | { |
354 | int bug = 0; | |
355 | int i = 0; | |
5a0768f6 | 356 | unsigned long highest_address = 0; |
ed8ea815 | 357 | struct vm_area_struct *vma = mm->mmap; |
ff26f70f | 358 | |
ed8ea815 | 359 | while (vma) { |
12352d3c | 360 | struct anon_vma *anon_vma = vma->anon_vma; |
ed8ea815 | 361 | struct anon_vma_chain *avc; |
ff26f70f | 362 | |
12352d3c KK |
363 | if (anon_vma) { |
364 | anon_vma_lock_read(anon_vma); | |
365 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
366 | anon_vma_interval_tree_verify(avc); | |
367 | anon_vma_unlock_read(anon_vma); | |
368 | } | |
369 | ||
1be7107f | 370 | highest_address = vm_end_gap(vma); |
ed8ea815 | 371 | vma = vma->vm_next; |
1da177e4 LT |
372 | i++; |
373 | } | |
5a0768f6 | 374 | if (i != mm->map_count) { |
8542bdfc | 375 | pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); |
5a0768f6 ML |
376 | bug = 1; |
377 | } | |
378 | if (highest_address != mm->highest_vm_end) { | |
8542bdfc | 379 | pr_emerg("mm->highest_vm_end %lx, found %lx\n", |
ff26f70f | 380 | mm->highest_vm_end, highest_address); |
5a0768f6 ML |
381 | bug = 1; |
382 | } | |
acf128d0 | 383 | i = browse_rb(mm); |
5a0768f6 | 384 | if (i != mm->map_count) { |
ff26f70f AM |
385 | if (i != -1) |
386 | pr_emerg("map_count %d rb %d\n", mm->map_count, i); | |
5a0768f6 ML |
387 | bug = 1; |
388 | } | |
96dad67f | 389 | VM_BUG_ON_MM(bug, mm); |
1da177e4 LT |
390 | } |
391 | #else | |
d3737187 | 392 | #define validate_mm_rb(root, ignore) do { } while (0) |
1da177e4 LT |
393 | #define validate_mm(mm) do { } while (0) |
394 | #endif | |
395 | ||
d3737187 ML |
396 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
397 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) | |
398 | ||
399 | /* | |
400 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or | |
401 | * vma->vm_prev->vm_end values changed, without modifying the vma's position | |
402 | * in the rbtree. | |
403 | */ | |
404 | static void vma_gap_update(struct vm_area_struct *vma) | |
405 | { | |
406 | /* | |
407 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback | |
408 | * function that does exacltly what we want. | |
409 | */ | |
410 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); | |
411 | } | |
412 | ||
413 | static inline void vma_rb_insert(struct vm_area_struct *vma, | |
414 | struct rb_root *root) | |
415 | { | |
416 | /* All rb_subtree_gap values must be consistent prior to insertion */ | |
417 | validate_mm_rb(root, NULL); | |
418 | ||
419 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
420 | } | |
421 | ||
8f26e0b1 | 422 | static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) |
d3737187 | 423 | { |
d3737187 ML |
424 | /* |
425 | * Note rb_erase_augmented is a fairly large inline function, | |
426 | * so make sure we instantiate it only once with our desired | |
427 | * augmented rbtree callbacks. | |
428 | */ | |
429 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
430 | } | |
431 | ||
8f26e0b1 AA |
432 | static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, |
433 | struct rb_root *root, | |
434 | struct vm_area_struct *ignore) | |
435 | { | |
436 | /* | |
437 | * All rb_subtree_gap values must be consistent prior to erase, | |
438 | * with the possible exception of the "next" vma being erased if | |
439 | * next->vm_start was reduced. | |
440 | */ | |
441 | validate_mm_rb(root, ignore); | |
442 | ||
443 | __vma_rb_erase(vma, root); | |
444 | } | |
445 | ||
446 | static __always_inline void vma_rb_erase(struct vm_area_struct *vma, | |
447 | struct rb_root *root) | |
448 | { | |
449 | /* | |
450 | * All rb_subtree_gap values must be consistent prior to erase, | |
451 | * with the possible exception of the vma being erased. | |
452 | */ | |
453 | validate_mm_rb(root, vma); | |
454 | ||
455 | __vma_rb_erase(vma, root); | |
456 | } | |
457 | ||
bf181b9f ML |
458 | /* |
459 | * vma has some anon_vma assigned, and is already inserted on that | |
460 | * anon_vma's interval trees. | |
461 | * | |
462 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
463 | * vma must be removed from the anon_vma's interval trees using | |
464 | * anon_vma_interval_tree_pre_update_vma(). | |
465 | * | |
466 | * After the update, the vma will be reinserted using | |
467 | * anon_vma_interval_tree_post_update_vma(). | |
468 | * | |
469 | * The entire update must be protected by exclusive mmap_sem and by | |
470 | * the root anon_vma's mutex. | |
471 | */ | |
472 | static inline void | |
473 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
474 | { | |
475 | struct anon_vma_chain *avc; | |
476 | ||
477 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
478 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
479 | } | |
480 | ||
481 | static inline void | |
482 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
483 | { | |
484 | struct anon_vma_chain *avc; | |
485 | ||
486 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
487 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
488 | } | |
489 | ||
6597d783 HD |
490 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
491 | unsigned long end, struct vm_area_struct **pprev, | |
492 | struct rb_node ***rb_link, struct rb_node **rb_parent) | |
1da177e4 | 493 | { |
6597d783 | 494 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
1da177e4 LT |
495 | |
496 | __rb_link = &mm->mm_rb.rb_node; | |
497 | rb_prev = __rb_parent = NULL; | |
1da177e4 LT |
498 | |
499 | while (*__rb_link) { | |
500 | struct vm_area_struct *vma_tmp; | |
501 | ||
502 | __rb_parent = *__rb_link; | |
503 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
504 | ||
505 | if (vma_tmp->vm_end > addr) { | |
6597d783 HD |
506 | /* Fail if an existing vma overlaps the area */ |
507 | if (vma_tmp->vm_start < end) | |
508 | return -ENOMEM; | |
1da177e4 LT |
509 | __rb_link = &__rb_parent->rb_left; |
510 | } else { | |
511 | rb_prev = __rb_parent; | |
512 | __rb_link = &__rb_parent->rb_right; | |
513 | } | |
514 | } | |
515 | ||
516 | *pprev = NULL; | |
517 | if (rb_prev) | |
518 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
519 | *rb_link = __rb_link; | |
520 | *rb_parent = __rb_parent; | |
6597d783 | 521 | return 0; |
1da177e4 LT |
522 | } |
523 | ||
e8420a8e CH |
524 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
525 | unsigned long addr, unsigned long end) | |
526 | { | |
527 | unsigned long nr_pages = 0; | |
528 | struct vm_area_struct *vma; | |
529 | ||
530 | /* Find first overlaping mapping */ | |
531 | vma = find_vma_intersection(mm, addr, end); | |
532 | if (!vma) | |
533 | return 0; | |
534 | ||
535 | nr_pages = (min(end, vma->vm_end) - | |
536 | max(addr, vma->vm_start)) >> PAGE_SHIFT; | |
537 | ||
538 | /* Iterate over the rest of the overlaps */ | |
539 | for (vma = vma->vm_next; vma; vma = vma->vm_next) { | |
540 | unsigned long overlap_len; | |
541 | ||
542 | if (vma->vm_start > end) | |
543 | break; | |
544 | ||
545 | overlap_len = min(end, vma->vm_end) - vma->vm_start; | |
546 | nr_pages += overlap_len >> PAGE_SHIFT; | |
547 | } | |
548 | ||
549 | return nr_pages; | |
550 | } | |
551 | ||
1da177e4 LT |
552 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
553 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
554 | { | |
d3737187 ML |
555 | /* Update tracking information for the gap following the new vma. */ |
556 | if (vma->vm_next) | |
557 | vma_gap_update(vma->vm_next); | |
558 | else | |
1be7107f | 559 | mm->highest_vm_end = vm_end_gap(vma); |
d3737187 ML |
560 | |
561 | /* | |
562 | * vma->vm_prev wasn't known when we followed the rbtree to find the | |
563 | * correct insertion point for that vma. As a result, we could not | |
564 | * update the vma vm_rb parents rb_subtree_gap values on the way down. | |
565 | * So, we first insert the vma with a zero rb_subtree_gap value | |
566 | * (to be consistent with what we did on the way down), and then | |
567 | * immediately update the gap to the correct value. Finally we | |
568 | * rebalance the rbtree after all augmented values have been set. | |
569 | */ | |
1da177e4 | 570 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
d3737187 ML |
571 | vma->rb_subtree_gap = 0; |
572 | vma_gap_update(vma); | |
573 | vma_rb_insert(vma, &mm->mm_rb); | |
1da177e4 LT |
574 | } |
575 | ||
cb8f488c | 576 | static void __vma_link_file(struct vm_area_struct *vma) |
1da177e4 | 577 | { |
48aae425 | 578 | struct file *file; |
1da177e4 LT |
579 | |
580 | file = vma->vm_file; | |
581 | if (file) { | |
582 | struct address_space *mapping = file->f_mapping; | |
583 | ||
584 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 585 | atomic_dec(&file_inode(file)->i_writecount); |
1da177e4 | 586 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 587 | atomic_inc(&mapping->i_mmap_writable); |
1da177e4 LT |
588 | |
589 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 590 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 LT |
591 | flush_dcache_mmap_unlock(mapping); |
592 | } | |
593 | } | |
594 | ||
595 | static void | |
596 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
597 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
598 | struct rb_node *rb_parent) | |
599 | { | |
600 | __vma_link_list(mm, vma, prev, rb_parent); | |
601 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
1da177e4 LT |
602 | } |
603 | ||
604 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
605 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
606 | struct rb_node *rb_parent) | |
607 | { | |
608 | struct address_space *mapping = NULL; | |
609 | ||
64ac4940 | 610 | if (vma->vm_file) { |
1da177e4 | 611 | mapping = vma->vm_file->f_mapping; |
83cde9e8 | 612 | i_mmap_lock_write(mapping); |
64ac4940 | 613 | } |
1da177e4 LT |
614 | |
615 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
616 | __vma_link_file(vma); | |
617 | ||
1da177e4 | 618 | if (mapping) |
83cde9e8 | 619 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
620 | |
621 | mm->map_count++; | |
622 | validate_mm(mm); | |
623 | } | |
624 | ||
625 | /* | |
88f6b4c3 | 626 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
6b2dbba8 | 627 | * mm's list and rbtree. It has already been inserted into the interval tree. |
1da177e4 | 628 | */ |
48aae425 | 629 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 630 | { |
6597d783 | 631 | struct vm_area_struct *prev; |
48aae425 | 632 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 633 | |
6597d783 HD |
634 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
635 | &prev, &rb_link, &rb_parent)) | |
636 | BUG(); | |
1da177e4 LT |
637 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
638 | mm->map_count++; | |
639 | } | |
640 | ||
e86f15ee AA |
641 | static __always_inline void __vma_unlink_common(struct mm_struct *mm, |
642 | struct vm_area_struct *vma, | |
643 | struct vm_area_struct *prev, | |
8f26e0b1 AA |
644 | bool has_prev, |
645 | struct vm_area_struct *ignore) | |
1da177e4 | 646 | { |
d3737187 | 647 | struct vm_area_struct *next; |
297c5eee | 648 | |
8f26e0b1 | 649 | vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); |
e86f15ee AA |
650 | next = vma->vm_next; |
651 | if (has_prev) | |
652 | prev->vm_next = next; | |
653 | else { | |
654 | prev = vma->vm_prev; | |
655 | if (prev) | |
656 | prev->vm_next = next; | |
657 | else | |
658 | mm->mmap = next; | |
659 | } | |
297c5eee LT |
660 | if (next) |
661 | next->vm_prev = prev; | |
615d6e87 DB |
662 | |
663 | /* Kill the cache */ | |
664 | vmacache_invalidate(mm); | |
1da177e4 LT |
665 | } |
666 | ||
e86f15ee AA |
667 | static inline void __vma_unlink_prev(struct mm_struct *mm, |
668 | struct vm_area_struct *vma, | |
669 | struct vm_area_struct *prev) | |
670 | { | |
8f26e0b1 | 671 | __vma_unlink_common(mm, vma, prev, true, vma); |
e86f15ee AA |
672 | } |
673 | ||
1da177e4 LT |
674 | /* |
675 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
676 | * is already present in an i_mmap tree without adjusting the tree. | |
677 | * The following helper function should be used when such adjustments | |
678 | * are necessary. The "insert" vma (if any) is to be inserted | |
679 | * before we drop the necessary locks. | |
680 | */ | |
e86f15ee AA |
681 | int __vma_adjust(struct vm_area_struct *vma, unsigned long start, |
682 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, | |
683 | struct vm_area_struct *expand) | |
1da177e4 LT |
684 | { |
685 | struct mm_struct *mm = vma->vm_mm; | |
e86f15ee | 686 | struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; |
1da177e4 | 687 | struct address_space *mapping = NULL; |
f808c13f | 688 | struct rb_root_cached *root = NULL; |
012f1800 | 689 | struct anon_vma *anon_vma = NULL; |
1da177e4 | 690 | struct file *file = vma->vm_file; |
d3737187 | 691 | bool start_changed = false, end_changed = false; |
1da177e4 LT |
692 | long adjust_next = 0; |
693 | int remove_next = 0; | |
694 | ||
695 | if (next && !insert) { | |
734537c9 | 696 | struct vm_area_struct *exporter = NULL, *importer = NULL; |
287d97ac | 697 | |
1da177e4 LT |
698 | if (end >= next->vm_end) { |
699 | /* | |
700 | * vma expands, overlapping all the next, and | |
701 | * perhaps the one after too (mprotect case 6). | |
86d12e47 | 702 | * The only other cases that gets here are |
e86f15ee | 703 | * case 1, case 7 and case 8. |
1da177e4 | 704 | */ |
e86f15ee AA |
705 | if (next == expand) { |
706 | /* | |
707 | * The only case where we don't expand "vma" | |
708 | * and we expand "next" instead is case 8. | |
709 | */ | |
710 | VM_WARN_ON(end != next->vm_end); | |
711 | /* | |
712 | * remove_next == 3 means we're | |
713 | * removing "vma" and that to do so we | |
714 | * swapped "vma" and "next". | |
715 | */ | |
716 | remove_next = 3; | |
717 | VM_WARN_ON(file != next->vm_file); | |
718 | swap(vma, next); | |
719 | } else { | |
720 | VM_WARN_ON(expand != vma); | |
721 | /* | |
722 | * case 1, 6, 7, remove_next == 2 is case 6, | |
723 | * remove_next == 1 is case 1 or 7. | |
724 | */ | |
725 | remove_next = 1 + (end > next->vm_end); | |
726 | VM_WARN_ON(remove_next == 2 && | |
727 | end != next->vm_next->vm_end); | |
728 | VM_WARN_ON(remove_next == 1 && | |
729 | end != next->vm_end); | |
730 | /* trim end to next, for case 6 first pass */ | |
731 | end = next->vm_end; | |
732 | } | |
733 | ||
287d97ac | 734 | exporter = next; |
1da177e4 | 735 | importer = vma; |
734537c9 KS |
736 | |
737 | /* | |
738 | * If next doesn't have anon_vma, import from vma after | |
739 | * next, if the vma overlaps with it. | |
740 | */ | |
97a42cd4 | 741 | if (remove_next == 2 && !next->anon_vma) |
734537c9 KS |
742 | exporter = next->vm_next; |
743 | ||
1da177e4 LT |
744 | } else if (end > next->vm_start) { |
745 | /* | |
746 | * vma expands, overlapping part of the next: | |
747 | * mprotect case 5 shifting the boundary up. | |
748 | */ | |
749 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
287d97ac | 750 | exporter = next; |
1da177e4 | 751 | importer = vma; |
e86f15ee | 752 | VM_WARN_ON(expand != importer); |
1da177e4 LT |
753 | } else if (end < vma->vm_end) { |
754 | /* | |
755 | * vma shrinks, and !insert tells it's not | |
756 | * split_vma inserting another: so it must be | |
757 | * mprotect case 4 shifting the boundary down. | |
758 | */ | |
cc71aba3 | 759 | adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); |
287d97ac | 760 | exporter = vma; |
1da177e4 | 761 | importer = next; |
e86f15ee | 762 | VM_WARN_ON(expand != importer); |
1da177e4 | 763 | } |
1da177e4 | 764 | |
5beb4930 RR |
765 | /* |
766 | * Easily overlooked: when mprotect shifts the boundary, | |
767 | * make sure the expanding vma has anon_vma set if the | |
768 | * shrinking vma had, to cover any anon pages imported. | |
769 | */ | |
287d97ac | 770 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
c4ea95d7 DF |
771 | int error; |
772 | ||
b800c91a | 773 | importer->anon_vma = exporter->anon_vma; |
c4ea95d7 | 774 | error = anon_vma_clone(importer, exporter); |
3fe89b3e | 775 | if (error) |
c4ea95d7 | 776 | return error; |
5beb4930 RR |
777 | } |
778 | } | |
734537c9 | 779 | again: |
e86f15ee | 780 | vma_adjust_trans_huge(orig_vma, start, end, adjust_next); |
37f9f559 | 781 | |
1da177e4 LT |
782 | if (file) { |
783 | mapping = file->f_mapping; | |
27ba0644 KS |
784 | root = &mapping->i_mmap; |
785 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); | |
682968e0 | 786 | |
27ba0644 KS |
787 | if (adjust_next) |
788 | uprobe_munmap(next, next->vm_start, next->vm_end); | |
682968e0 | 789 | |
83cde9e8 | 790 | i_mmap_lock_write(mapping); |
1da177e4 | 791 | if (insert) { |
1da177e4 | 792 | /* |
6b2dbba8 | 793 | * Put into interval tree now, so instantiated pages |
1da177e4 LT |
794 | * are visible to arm/parisc __flush_dcache_page |
795 | * throughout; but we cannot insert into address | |
796 | * space until vma start or end is updated. | |
797 | */ | |
798 | __vma_link_file(insert); | |
799 | } | |
800 | } | |
801 | ||
bf181b9f ML |
802 | anon_vma = vma->anon_vma; |
803 | if (!anon_vma && adjust_next) | |
804 | anon_vma = next->anon_vma; | |
805 | if (anon_vma) { | |
e86f15ee AA |
806 | VM_WARN_ON(adjust_next && next->anon_vma && |
807 | anon_vma != next->anon_vma); | |
4fc3f1d6 | 808 | anon_vma_lock_write(anon_vma); |
bf181b9f ML |
809 | anon_vma_interval_tree_pre_update_vma(vma); |
810 | if (adjust_next) | |
811 | anon_vma_interval_tree_pre_update_vma(next); | |
812 | } | |
012f1800 | 813 | |
1da177e4 LT |
814 | if (root) { |
815 | flush_dcache_mmap_lock(mapping); | |
6b2dbba8 | 816 | vma_interval_tree_remove(vma, root); |
1da177e4 | 817 | if (adjust_next) |
6b2dbba8 | 818 | vma_interval_tree_remove(next, root); |
1da177e4 LT |
819 | } |
820 | ||
d3737187 ML |
821 | if (start != vma->vm_start) { |
822 | vma->vm_start = start; | |
823 | start_changed = true; | |
824 | } | |
825 | if (end != vma->vm_end) { | |
826 | vma->vm_end = end; | |
827 | end_changed = true; | |
828 | } | |
1da177e4 LT |
829 | vma->vm_pgoff = pgoff; |
830 | if (adjust_next) { | |
831 | next->vm_start += adjust_next << PAGE_SHIFT; | |
832 | next->vm_pgoff += adjust_next; | |
833 | } | |
834 | ||
835 | if (root) { | |
836 | if (adjust_next) | |
6b2dbba8 ML |
837 | vma_interval_tree_insert(next, root); |
838 | vma_interval_tree_insert(vma, root); | |
1da177e4 LT |
839 | flush_dcache_mmap_unlock(mapping); |
840 | } | |
841 | ||
842 | if (remove_next) { | |
843 | /* | |
844 | * vma_merge has merged next into vma, and needs | |
845 | * us to remove next before dropping the locks. | |
846 | */ | |
e86f15ee AA |
847 | if (remove_next != 3) |
848 | __vma_unlink_prev(mm, next, vma); | |
849 | else | |
8f26e0b1 AA |
850 | /* |
851 | * vma is not before next if they've been | |
852 | * swapped. | |
853 | * | |
854 | * pre-swap() next->vm_start was reduced so | |
855 | * tell validate_mm_rb to ignore pre-swap() | |
856 | * "next" (which is stored in post-swap() | |
857 | * "vma"). | |
858 | */ | |
859 | __vma_unlink_common(mm, next, NULL, false, vma); | |
1da177e4 LT |
860 | if (file) |
861 | __remove_shared_vm_struct(next, file, mapping); | |
1da177e4 LT |
862 | } else if (insert) { |
863 | /* | |
864 | * split_vma has split insert from vma, and needs | |
865 | * us to insert it before dropping the locks | |
866 | * (it may either follow vma or precede it). | |
867 | */ | |
868 | __insert_vm_struct(mm, insert); | |
d3737187 ML |
869 | } else { |
870 | if (start_changed) | |
871 | vma_gap_update(vma); | |
872 | if (end_changed) { | |
873 | if (!next) | |
1be7107f | 874 | mm->highest_vm_end = vm_end_gap(vma); |
d3737187 ML |
875 | else if (!adjust_next) |
876 | vma_gap_update(next); | |
877 | } | |
1da177e4 LT |
878 | } |
879 | ||
bf181b9f ML |
880 | if (anon_vma) { |
881 | anon_vma_interval_tree_post_update_vma(vma); | |
882 | if (adjust_next) | |
883 | anon_vma_interval_tree_post_update_vma(next); | |
08b52706 | 884 | anon_vma_unlock_write(anon_vma); |
bf181b9f | 885 | } |
1da177e4 | 886 | if (mapping) |
83cde9e8 | 887 | i_mmap_unlock_write(mapping); |
1da177e4 | 888 | |
2b144498 | 889 | if (root) { |
7b2d81d4 | 890 | uprobe_mmap(vma); |
2b144498 SD |
891 | |
892 | if (adjust_next) | |
7b2d81d4 | 893 | uprobe_mmap(next); |
2b144498 SD |
894 | } |
895 | ||
1da177e4 | 896 | if (remove_next) { |
925d1c40 | 897 | if (file) { |
cbc91f71 | 898 | uprobe_munmap(next, next->vm_start, next->vm_end); |
1da177e4 | 899 | fput(file); |
925d1c40 | 900 | } |
5beb4930 RR |
901 | if (next->anon_vma) |
902 | anon_vma_merge(vma, next); | |
1da177e4 | 903 | mm->map_count--; |
3964acd0 | 904 | mpol_put(vma_policy(next)); |
1da177e4 LT |
905 | kmem_cache_free(vm_area_cachep, next); |
906 | /* | |
907 | * In mprotect's case 6 (see comments on vma_merge), | |
908 | * we must remove another next too. It would clutter | |
909 | * up the code too much to do both in one go. | |
910 | */ | |
e86f15ee AA |
911 | if (remove_next != 3) { |
912 | /* | |
913 | * If "next" was removed and vma->vm_end was | |
914 | * expanded (up) over it, in turn | |
915 | * "next->vm_prev->vm_end" changed and the | |
916 | * "vma->vm_next" gap must be updated. | |
917 | */ | |
918 | next = vma->vm_next; | |
919 | } else { | |
920 | /* | |
921 | * For the scope of the comment "next" and | |
922 | * "vma" considered pre-swap(): if "vma" was | |
923 | * removed, next->vm_start was expanded (down) | |
924 | * over it and the "next" gap must be updated. | |
925 | * Because of the swap() the post-swap() "vma" | |
926 | * actually points to pre-swap() "next" | |
927 | * (post-swap() "next" as opposed is now a | |
928 | * dangling pointer). | |
929 | */ | |
930 | next = vma; | |
931 | } | |
734537c9 KS |
932 | if (remove_next == 2) { |
933 | remove_next = 1; | |
934 | end = next->vm_end; | |
1da177e4 | 935 | goto again; |
734537c9 | 936 | } |
d3737187 ML |
937 | else if (next) |
938 | vma_gap_update(next); | |
fb8c41e9 AA |
939 | else { |
940 | /* | |
941 | * If remove_next == 2 we obviously can't | |
942 | * reach this path. | |
943 | * | |
944 | * If remove_next == 3 we can't reach this | |
945 | * path because pre-swap() next is always not | |
946 | * NULL. pre-swap() "next" is not being | |
947 | * removed and its next->vm_end is not altered | |
948 | * (and furthermore "end" already matches | |
949 | * next->vm_end in remove_next == 3). | |
950 | * | |
951 | * We reach this only in the remove_next == 1 | |
952 | * case if the "next" vma that was removed was | |
953 | * the highest vma of the mm. However in such | |
954 | * case next->vm_end == "end" and the extended | |
955 | * "vma" has vma->vm_end == next->vm_end so | |
956 | * mm->highest_vm_end doesn't need any update | |
957 | * in remove_next == 1 case. | |
958 | */ | |
1be7107f | 959 | VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); |
fb8c41e9 | 960 | } |
1da177e4 | 961 | } |
2b144498 | 962 | if (insert && file) |
7b2d81d4 | 963 | uprobe_mmap(insert); |
1da177e4 LT |
964 | |
965 | validate_mm(mm); | |
5beb4930 RR |
966 | |
967 | return 0; | |
1da177e4 LT |
968 | } |
969 | ||
970 | /* | |
971 | * If the vma has a ->close operation then the driver probably needs to release | |
972 | * per-vma resources, so we don't attempt to merge those. | |
973 | */ | |
1da177e4 | 974 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
19a809af AA |
975 | struct file *file, unsigned long vm_flags, |
976 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 977 | { |
34228d47 CG |
978 | /* |
979 | * VM_SOFTDIRTY should not prevent from VMA merging, if we | |
980 | * match the flags but dirty bit -- the caller should mark | |
981 | * merged VMA as dirty. If dirty bit won't be excluded from | |
982 | * comparison, we increase pressue on the memory system forcing | |
983 | * the kernel to generate new VMAs when old one could be | |
984 | * extended instead. | |
985 | */ | |
986 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) | |
1da177e4 LT |
987 | return 0; |
988 | if (vma->vm_file != file) | |
989 | return 0; | |
990 | if (vma->vm_ops && vma->vm_ops->close) | |
991 | return 0; | |
19a809af AA |
992 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) |
993 | return 0; | |
1da177e4 LT |
994 | return 1; |
995 | } | |
996 | ||
997 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
965f55de SL |
998 | struct anon_vma *anon_vma2, |
999 | struct vm_area_struct *vma) | |
1da177e4 | 1000 | { |
965f55de SL |
1001 | /* |
1002 | * The list_is_singular() test is to avoid merging VMA cloned from | |
1003 | * parents. This can improve scalability caused by anon_vma lock. | |
1004 | */ | |
1005 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
1006 | list_is_singular(&vma->anon_vma_chain))) | |
1007 | return 1; | |
1008 | return anon_vma1 == anon_vma2; | |
1da177e4 LT |
1009 | } |
1010 | ||
1011 | /* | |
1012 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
1013 | * in front of (at a lower virtual address and file offset than) the vma. | |
1014 | * | |
1015 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
1016 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
1017 | * | |
1018 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
1019 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
1020 | * wrap, nor mmaps which cover the final page at index -1UL. | |
1021 | */ | |
1022 | static int | |
1023 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
1024 | struct anon_vma *anon_vma, struct file *file, |
1025 | pgoff_t vm_pgoff, | |
1026 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 1027 | { |
19a809af | 1028 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 1029 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
1030 | if (vma->vm_pgoff == vm_pgoff) |
1031 | return 1; | |
1032 | } | |
1033 | return 0; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
1038 | * beyond (at a higher virtual address and file offset than) the vma. | |
1039 | * | |
1040 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
1041 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
1042 | */ | |
1043 | static int | |
1044 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
19a809af AA |
1045 | struct anon_vma *anon_vma, struct file *file, |
1046 | pgoff_t vm_pgoff, | |
1047 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 | 1048 | { |
19a809af | 1049 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && |
965f55de | 1050 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 1051 | pgoff_t vm_pglen; |
d6e93217 | 1052 | vm_pglen = vma_pages(vma); |
1da177e4 LT |
1053 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
1054 | return 1; | |
1055 | } | |
1056 | return 0; | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
1061 | * whether that can be merged with its predecessor or its successor. | |
1062 | * Or both (it neatly fills a hole). | |
1063 | * | |
1064 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
1065 | * certain not to be mapped by the time vma_merge is called; but when | |
1066 | * called for mprotect, it is certain to be already mapped (either at | |
1067 | * an offset within prev, or at the start of next), and the flags of | |
1068 | * this area are about to be changed to vm_flags - and the no-change | |
1069 | * case has already been eliminated. | |
1070 | * | |
1071 | * The following mprotect cases have to be considered, where AAAA is | |
1072 | * the area passed down from mprotect_fixup, never extending beyond one | |
1073 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
1074 | * | |
1075 | * AAAA AAAA AAAA AAAA | |
1076 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
1077 | * cannot merge might become might become might become | |
1078 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
1079 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
e86f15ee | 1080 | * mremap move: PPPPXXXXXXXX 8 |
1da177e4 LT |
1081 | * AAAA |
1082 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
1083 | * might become case 1 below case 2 below case 3 below | |
1084 | * | |
e86f15ee AA |
1085 | * It is important for case 8 that the the vma NNNN overlapping the |
1086 | * region AAAA is never going to extended over XXXX. Instead XXXX must | |
1087 | * be extended in region AAAA and NNNN must be removed. This way in | |
1088 | * all cases where vma_merge succeeds, the moment vma_adjust drops the | |
1089 | * rmap_locks, the properties of the merged vma will be already | |
1090 | * correct for the whole merged range. Some of those properties like | |
1091 | * vm_page_prot/vm_flags may be accessed by rmap_walks and they must | |
1092 | * be correct for the whole merged range immediately after the | |
1093 | * rmap_locks are released. Otherwise if XXXX would be removed and | |
1094 | * NNNN would be extended over the XXXX range, remove_migration_ptes | |
1095 | * or other rmap walkers (if working on addresses beyond the "end" | |
1096 | * parameter) may establish ptes with the wrong permissions of NNNN | |
1097 | * instead of the right permissions of XXXX. | |
1da177e4 LT |
1098 | */ |
1099 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
1100 | struct vm_area_struct *prev, unsigned long addr, | |
1101 | unsigned long end, unsigned long vm_flags, | |
cc71aba3 | 1102 | struct anon_vma *anon_vma, struct file *file, |
19a809af AA |
1103 | pgoff_t pgoff, struct mempolicy *policy, |
1104 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx) | |
1da177e4 LT |
1105 | { |
1106 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
1107 | struct vm_area_struct *area, *next; | |
5beb4930 | 1108 | int err; |
1da177e4 LT |
1109 | |
1110 | /* | |
1111 | * We later require that vma->vm_flags == vm_flags, | |
1112 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
1113 | */ | |
1114 | if (vm_flags & VM_SPECIAL) | |
1115 | return NULL; | |
1116 | ||
1117 | if (prev) | |
1118 | next = prev->vm_next; | |
1119 | else | |
1120 | next = mm->mmap; | |
1121 | area = next; | |
e86f15ee | 1122 | if (area && area->vm_end == end) /* cases 6, 7, 8 */ |
1da177e4 LT |
1123 | next = next->vm_next; |
1124 | ||
e86f15ee AA |
1125 | /* verify some invariant that must be enforced by the caller */ |
1126 | VM_WARN_ON(prev && addr <= prev->vm_start); | |
1127 | VM_WARN_ON(area && end > area->vm_end); | |
1128 | VM_WARN_ON(addr >= end); | |
1129 | ||
1da177e4 LT |
1130 | /* |
1131 | * Can it merge with the predecessor? | |
1132 | */ | |
1133 | if (prev && prev->vm_end == addr && | |
cc71aba3 | 1134 | mpol_equal(vma_policy(prev), policy) && |
1da177e4 | 1135 | can_vma_merge_after(prev, vm_flags, |
19a809af AA |
1136 | anon_vma, file, pgoff, |
1137 | vm_userfaultfd_ctx)) { | |
1da177e4 LT |
1138 | /* |
1139 | * OK, it can. Can we now merge in the successor as well? | |
1140 | */ | |
1141 | if (next && end == next->vm_start && | |
1142 | mpol_equal(policy, vma_policy(next)) && | |
1143 | can_vma_merge_before(next, vm_flags, | |
19a809af AA |
1144 | anon_vma, file, |
1145 | pgoff+pglen, | |
1146 | vm_userfaultfd_ctx) && | |
1da177e4 | 1147 | is_mergeable_anon_vma(prev->anon_vma, |
965f55de | 1148 | next->anon_vma, NULL)) { |
1da177e4 | 1149 | /* cases 1, 6 */ |
e86f15ee AA |
1150 | err = __vma_adjust(prev, prev->vm_start, |
1151 | next->vm_end, prev->vm_pgoff, NULL, | |
1152 | prev); | |
1da177e4 | 1153 | } else /* cases 2, 5, 7 */ |
e86f15ee AA |
1154 | err = __vma_adjust(prev, prev->vm_start, |
1155 | end, prev->vm_pgoff, NULL, prev); | |
5beb4930 RR |
1156 | if (err) |
1157 | return NULL; | |
6d50e60c | 1158 | khugepaged_enter_vma_merge(prev, vm_flags); |
1da177e4 LT |
1159 | return prev; |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Can this new request be merged in front of next? | |
1164 | */ | |
1165 | if (next && end == next->vm_start && | |
cc71aba3 | 1166 | mpol_equal(policy, vma_policy(next)) && |
1da177e4 | 1167 | can_vma_merge_before(next, vm_flags, |
19a809af AA |
1168 | anon_vma, file, pgoff+pglen, |
1169 | vm_userfaultfd_ctx)) { | |
1da177e4 | 1170 | if (prev && addr < prev->vm_end) /* case 4 */ |
e86f15ee AA |
1171 | err = __vma_adjust(prev, prev->vm_start, |
1172 | addr, prev->vm_pgoff, NULL, next); | |
1173 | else { /* cases 3, 8 */ | |
1174 | err = __vma_adjust(area, addr, next->vm_end, | |
1175 | next->vm_pgoff - pglen, NULL, next); | |
1176 | /* | |
1177 | * In case 3 area is already equal to next and | |
1178 | * this is a noop, but in case 8 "area" has | |
1179 | * been removed and next was expanded over it. | |
1180 | */ | |
1181 | area = next; | |
1182 | } | |
5beb4930 RR |
1183 | if (err) |
1184 | return NULL; | |
6d50e60c | 1185 | khugepaged_enter_vma_merge(area, vm_flags); |
1da177e4 LT |
1186 | return area; |
1187 | } | |
1188 | ||
1189 | return NULL; | |
1190 | } | |
1191 | ||
d0e9fe17 LT |
1192 | /* |
1193 | * Rough compatbility check to quickly see if it's even worth looking | |
1194 | * at sharing an anon_vma. | |
1195 | * | |
1196 | * They need to have the same vm_file, and the flags can only differ | |
1197 | * in things that mprotect may change. | |
1198 | * | |
1199 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1200 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1201 | * there is a vm_ops->close() function, because that indicates that the | |
1202 | * driver is doing some kind of reference counting. But that doesn't | |
1203 | * really matter for the anon_vma sharing case. | |
1204 | */ | |
1205 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1206 | { | |
1207 | return a->vm_end == b->vm_start && | |
1208 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1209 | a->vm_file == b->vm_file && | |
34228d47 | 1210 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && |
d0e9fe17 LT |
1211 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1216 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1217 | * the same as 'old', the other will be the new one that is trying | |
1218 | * to share the anon_vma. | |
1219 | * | |
1220 | * NOTE! This runs with mm_sem held for reading, so it is possible that | |
1221 | * the anon_vma of 'old' is concurrently in the process of being set up | |
1222 | * by another page fault trying to merge _that_. But that's ok: if it | |
1223 | * is being set up, that automatically means that it will be a singleton | |
1224 | * acceptable for merging, so we can do all of this optimistically. But | |
4db0c3c2 | 1225 | * we do that READ_ONCE() to make sure that we never re-load the pointer. |
d0e9fe17 LT |
1226 | * |
1227 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1228 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1229 | * is to return an anon_vma that is "complex" due to having gone through | |
1230 | * a fork). | |
1231 | * | |
1232 | * We also make sure that the two vma's are compatible (adjacent, | |
1233 | * and with the same memory policies). That's all stable, even with just | |
1234 | * a read lock on the mm_sem. | |
1235 | */ | |
1236 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1237 | { | |
1238 | if (anon_vma_compatible(a, b)) { | |
4db0c3c2 | 1239 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); |
d0e9fe17 LT |
1240 | |
1241 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1242 | return anon_vma; | |
1243 | } | |
1244 | return NULL; | |
1245 | } | |
1246 | ||
1da177e4 LT |
1247 | /* |
1248 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1249 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1250 | * to allocate a new anon_vma. It checks because a repetitive | |
1251 | * sequence of mprotects and faults may otherwise lead to distinct | |
1252 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1253 | * mprotect. | |
1254 | */ | |
1255 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1256 | { | |
d0e9fe17 | 1257 | struct anon_vma *anon_vma; |
1da177e4 | 1258 | struct vm_area_struct *near; |
1da177e4 LT |
1259 | |
1260 | near = vma->vm_next; | |
1261 | if (!near) | |
1262 | goto try_prev; | |
1263 | ||
d0e9fe17 LT |
1264 | anon_vma = reusable_anon_vma(near, vma, near); |
1265 | if (anon_vma) | |
1266 | return anon_vma; | |
1da177e4 | 1267 | try_prev: |
9be34c9d | 1268 | near = vma->vm_prev; |
1da177e4 LT |
1269 | if (!near) |
1270 | goto none; | |
1271 | ||
d0e9fe17 LT |
1272 | anon_vma = reusable_anon_vma(near, near, vma); |
1273 | if (anon_vma) | |
1274 | return anon_vma; | |
1da177e4 LT |
1275 | none: |
1276 | /* | |
1277 | * There's no absolute need to look only at touching neighbours: | |
1278 | * we could search further afield for "compatible" anon_vmas. | |
1279 | * But it would probably just be a waste of time searching, | |
1280 | * or lead to too many vmas hanging off the same anon_vma. | |
1281 | * We're trying to allow mprotect remerging later on, | |
1282 | * not trying to minimize memory used for anon_vmas. | |
1283 | */ | |
1284 | return NULL; | |
1285 | } | |
1286 | ||
40401530 AV |
1287 | /* |
1288 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1289 | * low as possible but still greater than mmap_min_addr | |
1290 | */ | |
1291 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1292 | { | |
1293 | hint &= PAGE_MASK; | |
1294 | if (((void *)hint != NULL) && | |
1295 | (hint < mmap_min_addr)) | |
1296 | return PAGE_ALIGN(mmap_min_addr); | |
1297 | return hint; | |
1298 | } | |
1299 | ||
363ee17f DB |
1300 | static inline int mlock_future_check(struct mm_struct *mm, |
1301 | unsigned long flags, | |
1302 | unsigned long len) | |
1303 | { | |
1304 | unsigned long locked, lock_limit; | |
1305 | ||
1306 | /* mlock MCL_FUTURE? */ | |
1307 | if (flags & VM_LOCKED) { | |
1308 | locked = len >> PAGE_SHIFT; | |
1309 | locked += mm->locked_vm; | |
1310 | lock_limit = rlimit(RLIMIT_MEMLOCK); | |
1311 | lock_limit >>= PAGE_SHIFT; | |
1312 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
1313 | return -EAGAIN; | |
1314 | } | |
1315 | return 0; | |
1316 | } | |
1317 | ||
1da177e4 | 1318 | /* |
27f5de79 | 1319 | * The caller must hold down_write(¤t->mm->mmap_sem). |
1da177e4 | 1320 | */ |
1fcfd8db | 1321 | unsigned long do_mmap(struct file *file, unsigned long addr, |
1da177e4 | 1322 | unsigned long len, unsigned long prot, |
1fcfd8db | 1323 | unsigned long flags, vm_flags_t vm_flags, |
897ab3e0 MR |
1324 | unsigned long pgoff, unsigned long *populate, |
1325 | struct list_head *uf) | |
1da177e4 | 1326 | { |
cc71aba3 | 1327 | struct mm_struct *mm = current->mm; |
62b5f7d0 | 1328 | int pkey = 0; |
1da177e4 | 1329 | |
41badc15 | 1330 | *populate = 0; |
bebeb3d6 | 1331 | |
e37609bb PK |
1332 | if (!len) |
1333 | return -EINVAL; | |
1334 | ||
1da177e4 LT |
1335 | /* |
1336 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1337 | * | |
1338 | * (the exception is when the underlying filesystem is noexec | |
1339 | * mounted, in which case we dont add PROT_EXEC.) | |
1340 | */ | |
1341 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
90f8572b | 1342 | if (!(file && path_noexec(&file->f_path))) |
1da177e4 LT |
1343 | prot |= PROT_EXEC; |
1344 | ||
7cd94146 EP |
1345 | if (!(flags & MAP_FIXED)) |
1346 | addr = round_hint_to_min(addr); | |
1347 | ||
1da177e4 LT |
1348 | /* Careful about overflows.. */ |
1349 | len = PAGE_ALIGN(len); | |
9206de95 | 1350 | if (!len) |
1da177e4 LT |
1351 | return -ENOMEM; |
1352 | ||
1353 | /* offset overflow? */ | |
1354 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
cc71aba3 | 1355 | return -EOVERFLOW; |
1da177e4 LT |
1356 | |
1357 | /* Too many mappings? */ | |
1358 | if (mm->map_count > sysctl_max_map_count) | |
1359 | return -ENOMEM; | |
1360 | ||
1361 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1362 | * that it represents a valid section of the address space. | |
1363 | */ | |
1364 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
de1741a1 | 1365 | if (offset_in_page(addr)) |
1da177e4 LT |
1366 | return addr; |
1367 | ||
62b5f7d0 DH |
1368 | if (prot == PROT_EXEC) { |
1369 | pkey = execute_only_pkey(mm); | |
1370 | if (pkey < 0) | |
1371 | pkey = 0; | |
1372 | } | |
1373 | ||
1da177e4 LT |
1374 | /* Do simple checking here so the lower-level routines won't have |
1375 | * to. we assume access permissions have been handled by the open | |
1376 | * of the memory object, so we don't do any here. | |
1377 | */ | |
62b5f7d0 | 1378 | vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | |
1da177e4 LT |
1379 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1380 | ||
cdf7b341 | 1381 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1382 | if (!can_do_mlock()) |
1383 | return -EPERM; | |
ba470de4 | 1384 | |
363ee17f DB |
1385 | if (mlock_future_check(mm, vm_flags, len)) |
1386 | return -EAGAIN; | |
1da177e4 | 1387 | |
1da177e4 | 1388 | if (file) { |
077bf22b | 1389 | struct inode *inode = file_inode(file); |
1c972597 DW |
1390 | unsigned long flags_mask; |
1391 | ||
1392 | flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; | |
077bf22b | 1393 | |
1da177e4 LT |
1394 | switch (flags & MAP_TYPE) { |
1395 | case MAP_SHARED: | |
1c972597 DW |
1396 | /* |
1397 | * Force use of MAP_SHARED_VALIDATE with non-legacy | |
1398 | * flags. E.g. MAP_SYNC is dangerous to use with | |
1399 | * MAP_SHARED as you don't know which consistency model | |
1400 | * you will get. We silently ignore unsupported flags | |
1401 | * with MAP_SHARED to preserve backward compatibility. | |
1402 | */ | |
1403 | flags &= LEGACY_MAP_MASK; | |
1404 | /* fall through */ | |
1405 | case MAP_SHARED_VALIDATE: | |
1406 | if (flags & ~flags_mask) | |
1407 | return -EOPNOTSUPP; | |
1da177e4 LT |
1408 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) |
1409 | return -EACCES; | |
1410 | ||
1411 | /* | |
1412 | * Make sure we don't allow writing to an append-only | |
1413 | * file.. | |
1414 | */ | |
1415 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1416 | return -EACCES; | |
1417 | ||
1418 | /* | |
1419 | * Make sure there are no mandatory locks on the file. | |
1420 | */ | |
d7a06983 | 1421 | if (locks_verify_locked(file)) |
1da177e4 LT |
1422 | return -EAGAIN; |
1423 | ||
1424 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
1425 | if (!(file->f_mode & FMODE_WRITE)) | |
1426 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
1427 | ||
1428 | /* fall through */ | |
1429 | case MAP_PRIVATE: | |
1430 | if (!(file->f_mode & FMODE_READ)) | |
1431 | return -EACCES; | |
90f8572b | 1432 | if (path_noexec(&file->f_path)) { |
80c5606c LT |
1433 | if (vm_flags & VM_EXEC) |
1434 | return -EPERM; | |
1435 | vm_flags &= ~VM_MAYEXEC; | |
1436 | } | |
80c5606c | 1437 | |
72c2d531 | 1438 | if (!file->f_op->mmap) |
80c5606c | 1439 | return -ENODEV; |
b2c56e4f ON |
1440 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1441 | return -EINVAL; | |
1da177e4 LT |
1442 | break; |
1443 | ||
1444 | default: | |
1445 | return -EINVAL; | |
1446 | } | |
1447 | } else { | |
1448 | switch (flags & MAP_TYPE) { | |
1449 | case MAP_SHARED: | |
b2c56e4f ON |
1450 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1451 | return -EINVAL; | |
ce363942 TH |
1452 | /* |
1453 | * Ignore pgoff. | |
1454 | */ | |
1455 | pgoff = 0; | |
1da177e4 LT |
1456 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1457 | break; | |
1458 | case MAP_PRIVATE: | |
1459 | /* | |
1460 | * Set pgoff according to addr for anon_vma. | |
1461 | */ | |
1462 | pgoff = addr >> PAGE_SHIFT; | |
1463 | break; | |
1464 | default: | |
1465 | return -EINVAL; | |
1466 | } | |
1467 | } | |
1468 | ||
c22c0d63 ML |
1469 | /* |
1470 | * Set 'VM_NORESERVE' if we should not account for the | |
1471 | * memory use of this mapping. | |
1472 | */ | |
1473 | if (flags & MAP_NORESERVE) { | |
1474 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1475 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1476 | vm_flags |= VM_NORESERVE; | |
1477 | ||
1478 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1479 | if (file && is_file_hugepages(file)) | |
1480 | vm_flags |= VM_NORESERVE; | |
1481 | } | |
1482 | ||
897ab3e0 | 1483 | addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); |
09a9f1d2 ML |
1484 | if (!IS_ERR_VALUE(addr) && |
1485 | ((vm_flags & VM_LOCKED) || | |
1486 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1487 | *populate = len; |
bebeb3d6 | 1488 | return addr; |
0165ab44 | 1489 | } |
6be5ceb0 | 1490 | |
a90f590a DB |
1491 | unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, |
1492 | unsigned long prot, unsigned long flags, | |
1493 | unsigned long fd, unsigned long pgoff) | |
66f0dc48 HD |
1494 | { |
1495 | struct file *file = NULL; | |
1e3ee14b | 1496 | unsigned long retval; |
66f0dc48 HD |
1497 | |
1498 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1499 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1500 | file = fget(fd); |
1501 | if (!file) | |
1e3ee14b | 1502 | return -EBADF; |
af73e4d9 NH |
1503 | if (is_file_hugepages(file)) |
1504 | len = ALIGN(len, huge_page_size(hstate_file(file))); | |
493af578 JE |
1505 | retval = -EINVAL; |
1506 | if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) | |
1507 | goto out_fput; | |
66f0dc48 HD |
1508 | } else if (flags & MAP_HUGETLB) { |
1509 | struct user_struct *user = NULL; | |
c103a4dc | 1510 | struct hstate *hs; |
af73e4d9 | 1511 | |
20ac2893 | 1512 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); |
091d0d55 LZ |
1513 | if (!hs) |
1514 | return -EINVAL; | |
1515 | ||
1516 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1517 | /* |
1518 | * VM_NORESERVE is used because the reservations will be | |
1519 | * taken when vm_ops->mmap() is called | |
1520 | * A dummy user value is used because we are not locking | |
1521 | * memory so no accounting is necessary | |
1522 | */ | |
af73e4d9 | 1523 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f AK |
1524 | VM_NORESERVE, |
1525 | &user, HUGETLB_ANONHUGE_INODE, | |
1526 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); | |
66f0dc48 HD |
1527 | if (IS_ERR(file)) |
1528 | return PTR_ERR(file); | |
1529 | } | |
1530 | ||
1531 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1532 | ||
9fbeb5ab | 1533 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
493af578 | 1534 | out_fput: |
66f0dc48 HD |
1535 | if (file) |
1536 | fput(file); | |
66f0dc48 HD |
1537 | return retval; |
1538 | } | |
1539 | ||
a90f590a DB |
1540 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1541 | unsigned long, prot, unsigned long, flags, | |
1542 | unsigned long, fd, unsigned long, pgoff) | |
1543 | { | |
1544 | return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); | |
1545 | } | |
1546 | ||
a4679373 CH |
1547 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1548 | struct mmap_arg_struct { | |
1549 | unsigned long addr; | |
1550 | unsigned long len; | |
1551 | unsigned long prot; | |
1552 | unsigned long flags; | |
1553 | unsigned long fd; | |
1554 | unsigned long offset; | |
1555 | }; | |
1556 | ||
1557 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1558 | { | |
1559 | struct mmap_arg_struct a; | |
1560 | ||
1561 | if (copy_from_user(&a, arg, sizeof(a))) | |
1562 | return -EFAULT; | |
de1741a1 | 1563 | if (offset_in_page(a.offset)) |
a4679373 CH |
1564 | return -EINVAL; |
1565 | ||
a90f590a DB |
1566 | return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, |
1567 | a.offset >> PAGE_SHIFT); | |
a4679373 CH |
1568 | } |
1569 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1570 | ||
4e950f6f AD |
1571 | /* |
1572 | * Some shared mappigns will want the pages marked read-only | |
1573 | * to track write events. If so, we'll downgrade vm_page_prot | |
1574 | * to the private version (using protection_map[] without the | |
1575 | * VM_SHARED bit). | |
1576 | */ | |
6d2329f8 | 1577 | int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) |
4e950f6f | 1578 | { |
ca16d140 | 1579 | vm_flags_t vm_flags = vma->vm_flags; |
8a04446a | 1580 | const struct vm_operations_struct *vm_ops = vma->vm_ops; |
4e950f6f AD |
1581 | |
1582 | /* If it was private or non-writable, the write bit is already clear */ | |
1583 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1584 | return 0; | |
1585 | ||
1586 | /* The backer wishes to know when pages are first written to? */ | |
8a04446a | 1587 | if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) |
4e950f6f AD |
1588 | return 1; |
1589 | ||
64e45507 PF |
1590 | /* The open routine did something to the protections that pgprot_modify |
1591 | * won't preserve? */ | |
6d2329f8 AA |
1592 | if (pgprot_val(vm_page_prot) != |
1593 | pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) | |
4e950f6f AD |
1594 | return 0; |
1595 | ||
64e45507 PF |
1596 | /* Do we need to track softdirty? */ |
1597 | if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) | |
1598 | return 1; | |
1599 | ||
4e950f6f | 1600 | /* Specialty mapping? */ |
4b6e1e37 | 1601 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1602 | return 0; |
1603 | ||
1604 | /* Can the mapping track the dirty pages? */ | |
1605 | return vma->vm_file && vma->vm_file->f_mapping && | |
1606 | mapping_cap_account_dirty(vma->vm_file->f_mapping); | |
1607 | } | |
1608 | ||
fc8744ad LT |
1609 | /* |
1610 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1611 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1612 | */ |
ca16d140 | 1613 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1614 | { |
5a6fe125 MG |
1615 | /* |
1616 | * hugetlb has its own accounting separate from the core VM | |
1617 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1618 | */ | |
1619 | if (file && is_file_hugepages(file)) | |
1620 | return 0; | |
1621 | ||
fc8744ad LT |
1622 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1623 | } | |
1624 | ||
0165ab44 | 1625 | unsigned long mmap_region(struct file *file, unsigned long addr, |
897ab3e0 MR |
1626 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, |
1627 | struct list_head *uf) | |
0165ab44 MS |
1628 | { |
1629 | struct mm_struct *mm = current->mm; | |
1630 | struct vm_area_struct *vma, *prev; | |
0165ab44 MS |
1631 | int error; |
1632 | struct rb_node **rb_link, *rb_parent; | |
1633 | unsigned long charged = 0; | |
0165ab44 | 1634 | |
e8420a8e | 1635 | /* Check against address space limit. */ |
84638335 | 1636 | if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { |
e8420a8e CH |
1637 | unsigned long nr_pages; |
1638 | ||
1639 | /* | |
1640 | * MAP_FIXED may remove pages of mappings that intersects with | |
1641 | * requested mapping. Account for the pages it would unmap. | |
1642 | */ | |
e8420a8e CH |
1643 | nr_pages = count_vma_pages_range(mm, addr, addr + len); |
1644 | ||
84638335 KK |
1645 | if (!may_expand_vm(mm, vm_flags, |
1646 | (len >> PAGE_SHIFT) - nr_pages)) | |
e8420a8e CH |
1647 | return -ENOMEM; |
1648 | } | |
1649 | ||
1da177e4 | 1650 | /* Clear old maps */ |
9fcd1457 RV |
1651 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
1652 | &rb_parent)) { | |
897ab3e0 | 1653 | if (do_munmap(mm, addr, len, uf)) |
1da177e4 | 1654 | return -ENOMEM; |
1da177e4 LT |
1655 | } |
1656 | ||
fc8744ad LT |
1657 | /* |
1658 | * Private writable mapping: check memory availability | |
1659 | */ | |
5a6fe125 | 1660 | if (accountable_mapping(file, vm_flags)) { |
fc8744ad | 1661 | charged = len >> PAGE_SHIFT; |
191c5424 | 1662 | if (security_vm_enough_memory_mm(mm, charged)) |
fc8744ad LT |
1663 | return -ENOMEM; |
1664 | vm_flags |= VM_ACCOUNT; | |
1da177e4 LT |
1665 | } |
1666 | ||
1667 | /* | |
de33c8db | 1668 | * Can we just expand an old mapping? |
1da177e4 | 1669 | */ |
19a809af AA |
1670 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, |
1671 | NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); | |
de33c8db LT |
1672 | if (vma) |
1673 | goto out; | |
1da177e4 LT |
1674 | |
1675 | /* | |
1676 | * Determine the object being mapped and call the appropriate | |
1677 | * specific mapper. the address has already been validated, but | |
1678 | * not unmapped, but the maps are removed from the list. | |
1679 | */ | |
c5e3b83e | 1680 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1681 | if (!vma) { |
1682 | error = -ENOMEM; | |
1683 | goto unacct_error; | |
1684 | } | |
1da177e4 LT |
1685 | |
1686 | vma->vm_mm = mm; | |
1687 | vma->vm_start = addr; | |
1688 | vma->vm_end = addr + len; | |
1689 | vma->vm_flags = vm_flags; | |
3ed75eb8 | 1690 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
1da177e4 | 1691 | vma->vm_pgoff = pgoff; |
5beb4930 | 1692 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
1693 | |
1694 | if (file) { | |
1da177e4 LT |
1695 | if (vm_flags & VM_DENYWRITE) { |
1696 | error = deny_write_access(file); | |
1697 | if (error) | |
1698 | goto free_vma; | |
1da177e4 | 1699 | } |
4bb5f5d9 DH |
1700 | if (vm_flags & VM_SHARED) { |
1701 | error = mapping_map_writable(file->f_mapping); | |
1702 | if (error) | |
1703 | goto allow_write_and_free_vma; | |
1704 | } | |
1705 | ||
1706 | /* ->mmap() can change vma->vm_file, but must guarantee that | |
1707 | * vma_link() below can deny write-access if VM_DENYWRITE is set | |
1708 | * and map writably if VM_SHARED is set. This usually means the | |
1709 | * new file must not have been exposed to user-space, yet. | |
1710 | */ | |
cb0942b8 | 1711 | vma->vm_file = get_file(file); |
f74ac015 | 1712 | error = call_mmap(file, vma); |
1da177e4 LT |
1713 | if (error) |
1714 | goto unmap_and_free_vma; | |
f8dbf0a7 HS |
1715 | |
1716 | /* Can addr have changed?? | |
1717 | * | |
1718 | * Answer: Yes, several device drivers can do it in their | |
1719 | * f_op->mmap method. -DaveM | |
2897b4d2 JK |
1720 | * Bug: If addr is changed, prev, rb_link, rb_parent should |
1721 | * be updated for vma_link() | |
f8dbf0a7 | 1722 | */ |
2897b4d2 JK |
1723 | WARN_ON_ONCE(addr != vma->vm_start); |
1724 | ||
f8dbf0a7 | 1725 | addr = vma->vm_start; |
f8dbf0a7 | 1726 | vm_flags = vma->vm_flags; |
1da177e4 LT |
1727 | } else if (vm_flags & VM_SHARED) { |
1728 | error = shmem_zero_setup(vma); | |
1729 | if (error) | |
1730 | goto free_vma; | |
1731 | } | |
1732 | ||
de33c8db | 1733 | vma_link(mm, vma, prev, rb_link, rb_parent); |
4d3d5b41 | 1734 | /* Once vma denies write, undo our temporary denial count */ |
4bb5f5d9 DH |
1735 | if (file) { |
1736 | if (vm_flags & VM_SHARED) | |
1737 | mapping_unmap_writable(file->f_mapping); | |
1738 | if (vm_flags & VM_DENYWRITE) | |
1739 | allow_write_access(file); | |
1740 | } | |
e8686772 | 1741 | file = vma->vm_file; |
4d3d5b41 | 1742 | out: |
cdd6c482 | 1743 | perf_event_mmap(vma); |
0a4a9391 | 1744 | |
84638335 | 1745 | vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); |
1da177e4 | 1746 | if (vm_flags & VM_LOCKED) { |
bebeb3d6 ML |
1747 | if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || |
1748 | vma == get_gate_vma(current->mm))) | |
06f9d8c2 | 1749 | mm->locked_vm += (len >> PAGE_SHIFT); |
bebeb3d6 | 1750 | else |
de60f5f1 | 1751 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
bebeb3d6 | 1752 | } |
2b144498 | 1753 | |
c7a3a88c ON |
1754 | if (file) |
1755 | uprobe_mmap(vma); | |
2b144498 | 1756 | |
d9104d1c CG |
1757 | /* |
1758 | * New (or expanded) vma always get soft dirty status. | |
1759 | * Otherwise user-space soft-dirty page tracker won't | |
1760 | * be able to distinguish situation when vma area unmapped, | |
1761 | * then new mapped in-place (which must be aimed as | |
1762 | * a completely new data area). | |
1763 | */ | |
1764 | vma->vm_flags |= VM_SOFTDIRTY; | |
1765 | ||
64e45507 PF |
1766 | vma_set_page_prot(vma); |
1767 | ||
1da177e4 LT |
1768 | return addr; |
1769 | ||
1770 | unmap_and_free_vma: | |
1da177e4 LT |
1771 | vma->vm_file = NULL; |
1772 | fput(file); | |
1773 | ||
1774 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1775 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1776 | charged = 0; | |
4bb5f5d9 DH |
1777 | if (vm_flags & VM_SHARED) |
1778 | mapping_unmap_writable(file->f_mapping); | |
1779 | allow_write_and_free_vma: | |
1780 | if (vm_flags & VM_DENYWRITE) | |
1781 | allow_write_access(file); | |
1da177e4 LT |
1782 | free_vma: |
1783 | kmem_cache_free(vm_area_cachep, vma); | |
1784 | unacct_error: | |
1785 | if (charged) | |
1786 | vm_unacct_memory(charged); | |
1787 | return error; | |
1788 | } | |
1789 | ||
db4fbfb9 ML |
1790 | unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
1791 | { | |
1792 | /* | |
1793 | * We implement the search by looking for an rbtree node that | |
1794 | * immediately follows a suitable gap. That is, | |
1795 | * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; | |
1796 | * - gap_end = vma->vm_start >= info->low_limit + length; | |
1797 | * - gap_end - gap_start >= length | |
1798 | */ | |
1799 | ||
1800 | struct mm_struct *mm = current->mm; | |
1801 | struct vm_area_struct *vma; | |
1802 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1803 | ||
1804 | /* Adjust search length to account for worst case alignment overhead */ | |
1805 | length = info->length + info->align_mask; | |
1806 | if (length < info->length) | |
1807 | return -ENOMEM; | |
1808 | ||
1809 | /* Adjust search limits by the desired length */ | |
1810 | if (info->high_limit < length) | |
1811 | return -ENOMEM; | |
1812 | high_limit = info->high_limit - length; | |
1813 | ||
1814 | if (info->low_limit > high_limit) | |
1815 | return -ENOMEM; | |
1816 | low_limit = info->low_limit + length; | |
1817 | ||
1818 | /* Check if rbtree root looks promising */ | |
1819 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1820 | goto check_highest; | |
1821 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1822 | if (vma->rb_subtree_gap < length) | |
1823 | goto check_highest; | |
1824 | ||
1825 | while (true) { | |
1826 | /* Visit left subtree if it looks promising */ | |
1be7107f | 1827 | gap_end = vm_start_gap(vma); |
db4fbfb9 ML |
1828 | if (gap_end >= low_limit && vma->vm_rb.rb_left) { |
1829 | struct vm_area_struct *left = | |
1830 | rb_entry(vma->vm_rb.rb_left, | |
1831 | struct vm_area_struct, vm_rb); | |
1832 | if (left->rb_subtree_gap >= length) { | |
1833 | vma = left; | |
1834 | continue; | |
1835 | } | |
1836 | } | |
1837 | ||
1be7107f | 1838 | gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; |
db4fbfb9 ML |
1839 | check_current: |
1840 | /* Check if current node has a suitable gap */ | |
1841 | if (gap_start > high_limit) | |
1842 | return -ENOMEM; | |
f4cb767d HD |
1843 | if (gap_end >= low_limit && |
1844 | gap_end > gap_start && gap_end - gap_start >= length) | |
db4fbfb9 ML |
1845 | goto found; |
1846 | ||
1847 | /* Visit right subtree if it looks promising */ | |
1848 | if (vma->vm_rb.rb_right) { | |
1849 | struct vm_area_struct *right = | |
1850 | rb_entry(vma->vm_rb.rb_right, | |
1851 | struct vm_area_struct, vm_rb); | |
1852 | if (right->rb_subtree_gap >= length) { | |
1853 | vma = right; | |
1854 | continue; | |
1855 | } | |
1856 | } | |
1857 | ||
1858 | /* Go back up the rbtree to find next candidate node */ | |
1859 | while (true) { | |
1860 | struct rb_node *prev = &vma->vm_rb; | |
1861 | if (!rb_parent(prev)) | |
1862 | goto check_highest; | |
1863 | vma = rb_entry(rb_parent(prev), | |
1864 | struct vm_area_struct, vm_rb); | |
1865 | if (prev == vma->vm_rb.rb_left) { | |
1be7107f HD |
1866 | gap_start = vm_end_gap(vma->vm_prev); |
1867 | gap_end = vm_start_gap(vma); | |
db4fbfb9 ML |
1868 | goto check_current; |
1869 | } | |
1870 | } | |
1871 | } | |
1872 | ||
1873 | check_highest: | |
1874 | /* Check highest gap, which does not precede any rbtree node */ | |
1875 | gap_start = mm->highest_vm_end; | |
1876 | gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ | |
1877 | if (gap_start > high_limit) | |
1878 | return -ENOMEM; | |
1879 | ||
1880 | found: | |
1881 | /* We found a suitable gap. Clip it with the original low_limit. */ | |
1882 | if (gap_start < info->low_limit) | |
1883 | gap_start = info->low_limit; | |
1884 | ||
1885 | /* Adjust gap address to the desired alignment */ | |
1886 | gap_start += (info->align_offset - gap_start) & info->align_mask; | |
1887 | ||
1888 | VM_BUG_ON(gap_start + info->length > info->high_limit); | |
1889 | VM_BUG_ON(gap_start + info->length > gap_end); | |
1890 | return gap_start; | |
1891 | } | |
1892 | ||
1893 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) | |
1894 | { | |
1895 | struct mm_struct *mm = current->mm; | |
1896 | struct vm_area_struct *vma; | |
1897 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1898 | ||
1899 | /* Adjust search length to account for worst case alignment overhead */ | |
1900 | length = info->length + info->align_mask; | |
1901 | if (length < info->length) | |
1902 | return -ENOMEM; | |
1903 | ||
1904 | /* | |
1905 | * Adjust search limits by the desired length. | |
1906 | * See implementation comment at top of unmapped_area(). | |
1907 | */ | |
1908 | gap_end = info->high_limit; | |
1909 | if (gap_end < length) | |
1910 | return -ENOMEM; | |
1911 | high_limit = gap_end - length; | |
1912 | ||
1913 | if (info->low_limit > high_limit) | |
1914 | return -ENOMEM; | |
1915 | low_limit = info->low_limit + length; | |
1916 | ||
1917 | /* Check highest gap, which does not precede any rbtree node */ | |
1918 | gap_start = mm->highest_vm_end; | |
1919 | if (gap_start <= high_limit) | |
1920 | goto found_highest; | |
1921 | ||
1922 | /* Check if rbtree root looks promising */ | |
1923 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1924 | return -ENOMEM; | |
1925 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1926 | if (vma->rb_subtree_gap < length) | |
1927 | return -ENOMEM; | |
1928 | ||
1929 | while (true) { | |
1930 | /* Visit right subtree if it looks promising */ | |
1be7107f | 1931 | gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; |
db4fbfb9 ML |
1932 | if (gap_start <= high_limit && vma->vm_rb.rb_right) { |
1933 | struct vm_area_struct *right = | |
1934 | rb_entry(vma->vm_rb.rb_right, | |
1935 | struct vm_area_struct, vm_rb); | |
1936 | if (right->rb_subtree_gap >= length) { | |
1937 | vma = right; | |
1938 | continue; | |
1939 | } | |
1940 | } | |
1941 | ||
1942 | check_current: | |
1943 | /* Check if current node has a suitable gap */ | |
1be7107f | 1944 | gap_end = vm_start_gap(vma); |
db4fbfb9 ML |
1945 | if (gap_end < low_limit) |
1946 | return -ENOMEM; | |
f4cb767d HD |
1947 | if (gap_start <= high_limit && |
1948 | gap_end > gap_start && gap_end - gap_start >= length) | |
db4fbfb9 ML |
1949 | goto found; |
1950 | ||
1951 | /* Visit left subtree if it looks promising */ | |
1952 | if (vma->vm_rb.rb_left) { | |
1953 | struct vm_area_struct *left = | |
1954 | rb_entry(vma->vm_rb.rb_left, | |
1955 | struct vm_area_struct, vm_rb); | |
1956 | if (left->rb_subtree_gap >= length) { | |
1957 | vma = left; | |
1958 | continue; | |
1959 | } | |
1960 | } | |
1961 | ||
1962 | /* Go back up the rbtree to find next candidate node */ | |
1963 | while (true) { | |
1964 | struct rb_node *prev = &vma->vm_rb; | |
1965 | if (!rb_parent(prev)) | |
1966 | return -ENOMEM; | |
1967 | vma = rb_entry(rb_parent(prev), | |
1968 | struct vm_area_struct, vm_rb); | |
1969 | if (prev == vma->vm_rb.rb_right) { | |
1970 | gap_start = vma->vm_prev ? | |
1be7107f | 1971 | vm_end_gap(vma->vm_prev) : 0; |
db4fbfb9 ML |
1972 | goto check_current; |
1973 | } | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | found: | |
1978 | /* We found a suitable gap. Clip it with the original high_limit. */ | |
1979 | if (gap_end > info->high_limit) | |
1980 | gap_end = info->high_limit; | |
1981 | ||
1982 | found_highest: | |
1983 | /* Compute highest gap address at the desired alignment */ | |
1984 | gap_end -= info->length; | |
1985 | gap_end -= (gap_end - info->align_offset) & info->align_mask; | |
1986 | ||
1987 | VM_BUG_ON(gap_end < info->low_limit); | |
1988 | VM_BUG_ON(gap_end < gap_start); | |
1989 | return gap_end; | |
1990 | } | |
1991 | ||
1da177e4 LT |
1992 | /* Get an address range which is currently unmapped. |
1993 | * For shmat() with addr=0. | |
1994 | * | |
1995 | * Ugly calling convention alert: | |
1996 | * Return value with the low bits set means error value, | |
1997 | * ie | |
1998 | * if (ret & ~PAGE_MASK) | |
1999 | * error = ret; | |
2000 | * | |
2001 | * This function "knows" that -ENOMEM has the bits set. | |
2002 | */ | |
2003 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
2004 | unsigned long | |
2005 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
2006 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
2007 | { | |
2008 | struct mm_struct *mm = current->mm; | |
1be7107f | 2009 | struct vm_area_struct *vma, *prev; |
db4fbfb9 | 2010 | struct vm_unmapped_area_info info; |
1da177e4 | 2011 | |
2afc745f | 2012 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
2013 | return -ENOMEM; |
2014 | ||
06abdfb4 BH |
2015 | if (flags & MAP_FIXED) |
2016 | return addr; | |
2017 | ||
1da177e4 LT |
2018 | if (addr) { |
2019 | addr = PAGE_ALIGN(addr); | |
1be7107f | 2020 | vma = find_vma_prev(mm, addr, &prev); |
2afc745f | 2021 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1be7107f HD |
2022 | (!vma || addr + len <= vm_start_gap(vma)) && |
2023 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
2024 | return addr; |
2025 | } | |
1da177e4 | 2026 | |
db4fbfb9 ML |
2027 | info.flags = 0; |
2028 | info.length = len; | |
4e99b021 | 2029 | info.low_limit = mm->mmap_base; |
db4fbfb9 ML |
2030 | info.high_limit = TASK_SIZE; |
2031 | info.align_mask = 0; | |
2032 | return vm_unmapped_area(&info); | |
1da177e4 | 2033 | } |
cc71aba3 | 2034 | #endif |
1da177e4 | 2035 | |
1da177e4 LT |
2036 | /* |
2037 | * This mmap-allocator allocates new areas top-down from below the | |
2038 | * stack's low limit (the base): | |
2039 | */ | |
2040 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
2041 | unsigned long | |
2042 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
2043 | const unsigned long len, const unsigned long pgoff, | |
2044 | const unsigned long flags) | |
2045 | { | |
1be7107f | 2046 | struct vm_area_struct *vma, *prev; |
1da177e4 | 2047 | struct mm_struct *mm = current->mm; |
db4fbfb9 ML |
2048 | unsigned long addr = addr0; |
2049 | struct vm_unmapped_area_info info; | |
1da177e4 LT |
2050 | |
2051 | /* requested length too big for entire address space */ | |
2afc745f | 2052 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
2053 | return -ENOMEM; |
2054 | ||
06abdfb4 BH |
2055 | if (flags & MAP_FIXED) |
2056 | return addr; | |
2057 | ||
1da177e4 LT |
2058 | /* requesting a specific address */ |
2059 | if (addr) { | |
2060 | addr = PAGE_ALIGN(addr); | |
1be7107f | 2061 | vma = find_vma_prev(mm, addr, &prev); |
2afc745f | 2062 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1be7107f HD |
2063 | (!vma || addr + len <= vm_start_gap(vma)) && |
2064 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
2065 | return addr; |
2066 | } | |
2067 | ||
db4fbfb9 ML |
2068 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
2069 | info.length = len; | |
2afc745f | 2070 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); |
db4fbfb9 ML |
2071 | info.high_limit = mm->mmap_base; |
2072 | info.align_mask = 0; | |
2073 | addr = vm_unmapped_area(&info); | |
b716ad95 | 2074 | |
1da177e4 LT |
2075 | /* |
2076 | * A failed mmap() very likely causes application failure, | |
2077 | * so fall back to the bottom-up function here. This scenario | |
2078 | * can happen with large stack limits and large mmap() | |
2079 | * allocations. | |
2080 | */ | |
de1741a1 | 2081 | if (offset_in_page(addr)) { |
db4fbfb9 ML |
2082 | VM_BUG_ON(addr != -ENOMEM); |
2083 | info.flags = 0; | |
2084 | info.low_limit = TASK_UNMAPPED_BASE; | |
2085 | info.high_limit = TASK_SIZE; | |
2086 | addr = vm_unmapped_area(&info); | |
2087 | } | |
1da177e4 LT |
2088 | |
2089 | return addr; | |
2090 | } | |
2091 | #endif | |
2092 | ||
1da177e4 LT |
2093 | unsigned long |
2094 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
2095 | unsigned long pgoff, unsigned long flags) | |
2096 | { | |
06abdfb4 BH |
2097 | unsigned long (*get_area)(struct file *, unsigned long, |
2098 | unsigned long, unsigned long, unsigned long); | |
2099 | ||
9206de95 AV |
2100 | unsigned long error = arch_mmap_check(addr, len, flags); |
2101 | if (error) | |
2102 | return error; | |
2103 | ||
2104 | /* Careful about overflows.. */ | |
2105 | if (len > TASK_SIZE) | |
2106 | return -ENOMEM; | |
2107 | ||
06abdfb4 | 2108 | get_area = current->mm->get_unmapped_area; |
c01d5b30 HD |
2109 | if (file) { |
2110 | if (file->f_op->get_unmapped_area) | |
2111 | get_area = file->f_op->get_unmapped_area; | |
2112 | } else if (flags & MAP_SHARED) { | |
2113 | /* | |
2114 | * mmap_region() will call shmem_zero_setup() to create a file, | |
2115 | * so use shmem's get_unmapped_area in case it can be huge. | |
2116 | * do_mmap_pgoff() will clear pgoff, so match alignment. | |
2117 | */ | |
2118 | pgoff = 0; | |
2119 | get_area = shmem_get_unmapped_area; | |
2120 | } | |
2121 | ||
06abdfb4 BH |
2122 | addr = get_area(file, addr, len, pgoff, flags); |
2123 | if (IS_ERR_VALUE(addr)) | |
2124 | return addr; | |
1da177e4 | 2125 | |
07ab67c8 LT |
2126 | if (addr > TASK_SIZE - len) |
2127 | return -ENOMEM; | |
de1741a1 | 2128 | if (offset_in_page(addr)) |
07ab67c8 | 2129 | return -EINVAL; |
06abdfb4 | 2130 | |
9ac4ed4b AV |
2131 | error = security_mmap_addr(addr); |
2132 | return error ? error : addr; | |
1da177e4 LT |
2133 | } |
2134 | ||
2135 | EXPORT_SYMBOL(get_unmapped_area); | |
2136 | ||
2137 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
48aae425 | 2138 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2139 | { |
615d6e87 DB |
2140 | struct rb_node *rb_node; |
2141 | struct vm_area_struct *vma; | |
1da177e4 | 2142 | |
841e31e5 | 2143 | /* Check the cache first. */ |
615d6e87 DB |
2144 | vma = vmacache_find(mm, addr); |
2145 | if (likely(vma)) | |
2146 | return vma; | |
841e31e5 | 2147 | |
615d6e87 | 2148 | rb_node = mm->mm_rb.rb_node; |
841e31e5 | 2149 | |
615d6e87 DB |
2150 | while (rb_node) { |
2151 | struct vm_area_struct *tmp; | |
2152 | ||
2153 | tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
2154 | ||
2155 | if (tmp->vm_end > addr) { | |
2156 | vma = tmp; | |
2157 | if (tmp->vm_start <= addr) | |
2158 | break; | |
2159 | rb_node = rb_node->rb_left; | |
2160 | } else | |
2161 | rb_node = rb_node->rb_right; | |
1da177e4 | 2162 | } |
615d6e87 DB |
2163 | |
2164 | if (vma) | |
2165 | vmacache_update(addr, vma); | |
1da177e4 LT |
2166 | return vma; |
2167 | } | |
2168 | ||
2169 | EXPORT_SYMBOL(find_vma); | |
2170 | ||
6bd4837d KM |
2171 | /* |
2172 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. | |
6bd4837d | 2173 | */ |
1da177e4 LT |
2174 | struct vm_area_struct * |
2175 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
2176 | struct vm_area_struct **pprev) | |
2177 | { | |
6bd4837d | 2178 | struct vm_area_struct *vma; |
1da177e4 | 2179 | |
6bd4837d | 2180 | vma = find_vma(mm, addr); |
83cd904d MP |
2181 | if (vma) { |
2182 | *pprev = vma->vm_prev; | |
2183 | } else { | |
2184 | struct rb_node *rb_node = mm->mm_rb.rb_node; | |
2185 | *pprev = NULL; | |
2186 | while (rb_node) { | |
2187 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
2188 | rb_node = rb_node->rb_right; | |
2189 | } | |
2190 | } | |
6bd4837d | 2191 | return vma; |
1da177e4 LT |
2192 | } |
2193 | ||
2194 | /* | |
2195 | * Verify that the stack growth is acceptable and | |
2196 | * update accounting. This is shared with both the | |
2197 | * grow-up and grow-down cases. | |
2198 | */ | |
1be7107f HD |
2199 | static int acct_stack_growth(struct vm_area_struct *vma, |
2200 | unsigned long size, unsigned long grow) | |
1da177e4 LT |
2201 | { |
2202 | struct mm_struct *mm = vma->vm_mm; | |
1be7107f | 2203 | unsigned long new_start; |
1da177e4 LT |
2204 | |
2205 | /* address space limit tests */ | |
84638335 | 2206 | if (!may_expand_vm(mm, vma->vm_flags, grow)) |
1da177e4 LT |
2207 | return -ENOMEM; |
2208 | ||
2209 | /* Stack limit test */ | |
24c79d8e | 2210 | if (size > rlimit(RLIMIT_STACK)) |
1da177e4 LT |
2211 | return -ENOMEM; |
2212 | ||
2213 | /* mlock limit tests */ | |
2214 | if (vma->vm_flags & VM_LOCKED) { | |
2215 | unsigned long locked; | |
2216 | unsigned long limit; | |
2217 | locked = mm->locked_vm + grow; | |
24c79d8e | 2218 | limit = rlimit(RLIMIT_MEMLOCK); |
59e99e5b | 2219 | limit >>= PAGE_SHIFT; |
1da177e4 LT |
2220 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
2221 | return -ENOMEM; | |
2222 | } | |
2223 | ||
0d59a01b AL |
2224 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
2225 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
2226 | vma->vm_end - size; | |
2227 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
2228 | return -EFAULT; | |
2229 | ||
1da177e4 LT |
2230 | /* |
2231 | * Overcommit.. This must be the final test, as it will | |
2232 | * update security statistics. | |
2233 | */ | |
05fa199d | 2234 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
2235 | return -ENOMEM; |
2236 | ||
1da177e4 LT |
2237 | return 0; |
2238 | } | |
2239 | ||
46dea3d0 | 2240 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 2241 | /* |
46dea3d0 HD |
2242 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
2243 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 2244 | */ |
46dea3d0 | 2245 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 2246 | { |
09357814 | 2247 | struct mm_struct *mm = vma->vm_mm; |
1be7107f HD |
2248 | struct vm_area_struct *next; |
2249 | unsigned long gap_addr; | |
12352d3c | 2250 | int error = 0; |
1da177e4 LT |
2251 | |
2252 | if (!(vma->vm_flags & VM_GROWSUP)) | |
2253 | return -EFAULT; | |
2254 | ||
bd726c90 | 2255 | /* Guard against exceeding limits of the address space. */ |
1be7107f | 2256 | address &= PAGE_MASK; |
37511fb5 | 2257 | if (address >= (TASK_SIZE & PAGE_MASK)) |
12352d3c | 2258 | return -ENOMEM; |
bd726c90 | 2259 | address += PAGE_SIZE; |
12352d3c | 2260 | |
1be7107f HD |
2261 | /* Enforce stack_guard_gap */ |
2262 | gap_addr = address + stack_guard_gap; | |
bd726c90 HD |
2263 | |
2264 | /* Guard against overflow */ | |
2265 | if (gap_addr < address || gap_addr > TASK_SIZE) | |
2266 | gap_addr = TASK_SIZE; | |
2267 | ||
1be7107f | 2268 | next = vma->vm_next; |
561b5e07 MH |
2269 | if (next && next->vm_start < gap_addr && |
2270 | (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) { | |
1be7107f HD |
2271 | if (!(next->vm_flags & VM_GROWSUP)) |
2272 | return -ENOMEM; | |
2273 | /* Check that both stack segments have the same anon_vma? */ | |
2274 | } | |
2275 | ||
12352d3c | 2276 | /* We must make sure the anon_vma is allocated. */ |
1da177e4 LT |
2277 | if (unlikely(anon_vma_prepare(vma))) |
2278 | return -ENOMEM; | |
1da177e4 LT |
2279 | |
2280 | /* | |
2281 | * vma->vm_start/vm_end cannot change under us because the caller | |
2282 | * is required to hold the mmap_sem in read mode. We need the | |
2283 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2284 | */ | |
12352d3c | 2285 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2286 | |
2287 | /* Somebody else might have raced and expanded it already */ | |
2288 | if (address > vma->vm_end) { | |
2289 | unsigned long size, grow; | |
2290 | ||
2291 | size = address - vma->vm_start; | |
2292 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
2293 | ||
42c36f63 HD |
2294 | error = -ENOMEM; |
2295 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
2296 | error = acct_stack_growth(vma, size, grow); | |
2297 | if (!error) { | |
4128997b ML |
2298 | /* |
2299 | * vma_gap_update() doesn't support concurrent | |
2300 | * updates, but we only hold a shared mmap_sem | |
2301 | * lock here, so we need to protect against | |
2302 | * concurrent vma expansions. | |
12352d3c | 2303 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2304 | * we don't guarantee that all growable vmas |
2305 | * in a mm share the same root anon vma. | |
2306 | * So, we reuse mm->page_table_lock to guard | |
2307 | * against concurrent vma expansions. | |
2308 | */ | |
09357814 | 2309 | spin_lock(&mm->page_table_lock); |
87e8827b | 2310 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2311 | mm->locked_vm += grow; |
84638335 | 2312 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2313 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 2314 | vma->vm_end = address; |
bf181b9f | 2315 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 ML |
2316 | if (vma->vm_next) |
2317 | vma_gap_update(vma->vm_next); | |
2318 | else | |
1be7107f | 2319 | mm->highest_vm_end = vm_end_gap(vma); |
09357814 | 2320 | spin_unlock(&mm->page_table_lock); |
4128997b | 2321 | |
42c36f63 HD |
2322 | perf_event_mmap(vma); |
2323 | } | |
3af9e859 | 2324 | } |
1da177e4 | 2325 | } |
12352d3c | 2326 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2327 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2328 | validate_mm(mm); |
1da177e4 LT |
2329 | return error; |
2330 | } | |
46dea3d0 HD |
2331 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2332 | ||
1da177e4 LT |
2333 | /* |
2334 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
2335 | */ | |
d05f3169 | 2336 | int expand_downwards(struct vm_area_struct *vma, |
b6a2fea3 | 2337 | unsigned long address) |
1da177e4 | 2338 | { |
09357814 | 2339 | struct mm_struct *mm = vma->vm_mm; |
1be7107f | 2340 | struct vm_area_struct *prev; |
1da177e4 LT |
2341 | int error; |
2342 | ||
8869477a | 2343 | address &= PAGE_MASK; |
e5467859 | 2344 | error = security_mmap_addr(address); |
8869477a EP |
2345 | if (error) |
2346 | return error; | |
2347 | ||
1be7107f | 2348 | /* Enforce stack_guard_gap */ |
1be7107f | 2349 | prev = vma->vm_prev; |
32e4e6d5 ON |
2350 | /* Check that both stack segments have the same anon_vma? */ |
2351 | if (prev && !(prev->vm_flags & VM_GROWSDOWN) && | |
561b5e07 | 2352 | (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) { |
32e4e6d5 | 2353 | if (address - prev->vm_end < stack_guard_gap) |
1be7107f | 2354 | return -ENOMEM; |
1be7107f HD |
2355 | } |
2356 | ||
12352d3c KK |
2357 | /* We must make sure the anon_vma is allocated. */ |
2358 | if (unlikely(anon_vma_prepare(vma))) | |
2359 | return -ENOMEM; | |
1da177e4 LT |
2360 | |
2361 | /* | |
2362 | * vma->vm_start/vm_end cannot change under us because the caller | |
2363 | * is required to hold the mmap_sem in read mode. We need the | |
2364 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2365 | */ | |
12352d3c | 2366 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2367 | |
2368 | /* Somebody else might have raced and expanded it already */ | |
2369 | if (address < vma->vm_start) { | |
2370 | unsigned long size, grow; | |
2371 | ||
2372 | size = vma->vm_end - address; | |
2373 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2374 | ||
a626ca6a LT |
2375 | error = -ENOMEM; |
2376 | if (grow <= vma->vm_pgoff) { | |
2377 | error = acct_stack_growth(vma, size, grow); | |
2378 | if (!error) { | |
4128997b ML |
2379 | /* |
2380 | * vma_gap_update() doesn't support concurrent | |
2381 | * updates, but we only hold a shared mmap_sem | |
2382 | * lock here, so we need to protect against | |
2383 | * concurrent vma expansions. | |
12352d3c | 2384 | * anon_vma_lock_write() doesn't help here, as |
4128997b ML |
2385 | * we don't guarantee that all growable vmas |
2386 | * in a mm share the same root anon vma. | |
2387 | * So, we reuse mm->page_table_lock to guard | |
2388 | * against concurrent vma expansions. | |
2389 | */ | |
09357814 | 2390 | spin_lock(&mm->page_table_lock); |
87e8827b | 2391 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2392 | mm->locked_vm += grow; |
84638335 | 2393 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2394 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2395 | vma->vm_start = address; |
2396 | vma->vm_pgoff -= grow; | |
bf181b9f | 2397 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 | 2398 | vma_gap_update(vma); |
09357814 | 2399 | spin_unlock(&mm->page_table_lock); |
4128997b | 2400 | |
a626ca6a LT |
2401 | perf_event_mmap(vma); |
2402 | } | |
1da177e4 LT |
2403 | } |
2404 | } | |
12352d3c | 2405 | anon_vma_unlock_write(vma->anon_vma); |
6d50e60c | 2406 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
09357814 | 2407 | validate_mm(mm); |
1da177e4 LT |
2408 | return error; |
2409 | } | |
2410 | ||
1be7107f HD |
2411 | /* enforced gap between the expanding stack and other mappings. */ |
2412 | unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; | |
2413 | ||
2414 | static int __init cmdline_parse_stack_guard_gap(char *p) | |
2415 | { | |
2416 | unsigned long val; | |
2417 | char *endptr; | |
2418 | ||
2419 | val = simple_strtoul(p, &endptr, 10); | |
2420 | if (!*endptr) | |
2421 | stack_guard_gap = val << PAGE_SHIFT; | |
2422 | ||
2423 | return 0; | |
2424 | } | |
2425 | __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); | |
2426 | ||
b6a2fea3 OW |
2427 | #ifdef CONFIG_STACK_GROWSUP |
2428 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2429 | { | |
2430 | return expand_upwards(vma, address); | |
2431 | } | |
2432 | ||
2433 | struct vm_area_struct * | |
2434 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2435 | { | |
2436 | struct vm_area_struct *vma, *prev; | |
2437 | ||
2438 | addr &= PAGE_MASK; | |
2439 | vma = find_vma_prev(mm, addr, &prev); | |
2440 | if (vma && (vma->vm_start <= addr)) | |
2441 | return vma; | |
1c127185 | 2442 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2443 | return NULL; |
cea10a19 | 2444 | if (prev->vm_flags & VM_LOCKED) |
fc05f566 | 2445 | populate_vma_page_range(prev, addr, prev->vm_end, NULL); |
b6a2fea3 OW |
2446 | return prev; |
2447 | } | |
2448 | #else | |
2449 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2450 | { | |
2451 | return expand_downwards(vma, address); | |
2452 | } | |
2453 | ||
1da177e4 | 2454 | struct vm_area_struct * |
cc71aba3 | 2455 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2456 | { |
cc71aba3 | 2457 | struct vm_area_struct *vma; |
1da177e4 LT |
2458 | unsigned long start; |
2459 | ||
2460 | addr &= PAGE_MASK; | |
cc71aba3 | 2461 | vma = find_vma(mm, addr); |
1da177e4 LT |
2462 | if (!vma) |
2463 | return NULL; | |
2464 | if (vma->vm_start <= addr) | |
2465 | return vma; | |
2466 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2467 | return NULL; | |
2468 | start = vma->vm_start; | |
2469 | if (expand_stack(vma, addr)) | |
2470 | return NULL; | |
cea10a19 | 2471 | if (vma->vm_flags & VM_LOCKED) |
fc05f566 | 2472 | populate_vma_page_range(vma, addr, start, NULL); |
1da177e4 LT |
2473 | return vma; |
2474 | } | |
2475 | #endif | |
2476 | ||
e1d6d01a JB |
2477 | EXPORT_SYMBOL_GPL(find_extend_vma); |
2478 | ||
1da177e4 | 2479 | /* |
2c0b3814 | 2480 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 2481 | * so release them, and do the vma updates. |
2c0b3814 HD |
2482 | * |
2483 | * Called with the mm semaphore held. | |
1da177e4 | 2484 | */ |
2c0b3814 | 2485 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2486 | { |
4f74d2c8 LT |
2487 | unsigned long nr_accounted = 0; |
2488 | ||
365e9c87 HD |
2489 | /* Update high watermark before we lower total_vm */ |
2490 | update_hiwater_vm(mm); | |
1da177e4 | 2491 | do { |
2c0b3814 HD |
2492 | long nrpages = vma_pages(vma); |
2493 | ||
4f74d2c8 LT |
2494 | if (vma->vm_flags & VM_ACCOUNT) |
2495 | nr_accounted += nrpages; | |
84638335 | 2496 | vm_stat_account(mm, vma->vm_flags, -nrpages); |
a8fb5618 | 2497 | vma = remove_vma(vma); |
146425a3 | 2498 | } while (vma); |
4f74d2c8 | 2499 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2500 | validate_mm(mm); |
2501 | } | |
2502 | ||
2503 | /* | |
2504 | * Get rid of page table information in the indicated region. | |
2505 | * | |
f10df686 | 2506 | * Called with the mm semaphore held. |
1da177e4 LT |
2507 | */ |
2508 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
2509 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
2510 | unsigned long start, unsigned long end) | |
1da177e4 | 2511 | { |
cc71aba3 | 2512 | struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; |
d16dfc55 | 2513 | struct mmu_gather tlb; |
1da177e4 LT |
2514 | |
2515 | lru_add_drain(); | |
2b047252 | 2516 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 2517 | update_hiwater_rss(mm); |
4f74d2c8 | 2518 | unmap_vmas(&tlb, vma, start, end); |
d16dfc55 | 2519 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
6ee8630e | 2520 | next ? next->vm_start : USER_PGTABLES_CEILING); |
d16dfc55 | 2521 | tlb_finish_mmu(&tlb, start, end); |
1da177e4 LT |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
2526 | * vma list as we go.. | |
2527 | */ | |
2528 | static void | |
2529 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
2530 | struct vm_area_struct *prev, unsigned long end) | |
2531 | { | |
2532 | struct vm_area_struct **insertion_point; | |
2533 | struct vm_area_struct *tail_vma = NULL; | |
2534 | ||
2535 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
297c5eee | 2536 | vma->vm_prev = NULL; |
1da177e4 | 2537 | do { |
d3737187 | 2538 | vma_rb_erase(vma, &mm->mm_rb); |
1da177e4 LT |
2539 | mm->map_count--; |
2540 | tail_vma = vma; | |
2541 | vma = vma->vm_next; | |
2542 | } while (vma && vma->vm_start < end); | |
2543 | *insertion_point = vma; | |
d3737187 | 2544 | if (vma) { |
297c5eee | 2545 | vma->vm_prev = prev; |
d3737187 ML |
2546 | vma_gap_update(vma); |
2547 | } else | |
1be7107f | 2548 | mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; |
1da177e4 | 2549 | tail_vma->vm_next = NULL; |
615d6e87 DB |
2550 | |
2551 | /* Kill the cache */ | |
2552 | vmacache_invalidate(mm); | |
1da177e4 LT |
2553 | } |
2554 | ||
2555 | /* | |
def5efe0 DR |
2556 | * __split_vma() bypasses sysctl_max_map_count checking. We use this where it |
2557 | * has already been checked or doesn't make sense to fail. | |
1da177e4 | 2558 | */ |
def5efe0 DR |
2559 | int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
2560 | unsigned long addr, int new_below) | |
1da177e4 | 2561 | { |
1da177e4 | 2562 | struct vm_area_struct *new; |
e3975891 | 2563 | int err; |
1da177e4 | 2564 | |
31383c68 DW |
2565 | if (vma->vm_ops && vma->vm_ops->split) { |
2566 | err = vma->vm_ops->split(vma, addr); | |
2567 | if (err) | |
2568 | return err; | |
2569 | } | |
1da177e4 | 2570 | |
e94b1766 | 2571 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 | 2572 | if (!new) |
e3975891 | 2573 | return -ENOMEM; |
1da177e4 LT |
2574 | |
2575 | /* most fields are the same, copy all, and then fixup */ | |
2576 | *new = *vma; | |
2577 | ||
5beb4930 RR |
2578 | INIT_LIST_HEAD(&new->anon_vma_chain); |
2579 | ||
1da177e4 LT |
2580 | if (new_below) |
2581 | new->vm_end = addr; | |
2582 | else { | |
2583 | new->vm_start = addr; | |
2584 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2585 | } | |
2586 | ||
ef0855d3 ON |
2587 | err = vma_dup_policy(vma, new); |
2588 | if (err) | |
5beb4930 | 2589 | goto out_free_vma; |
1da177e4 | 2590 | |
c4ea95d7 DF |
2591 | err = anon_vma_clone(new, vma); |
2592 | if (err) | |
5beb4930 RR |
2593 | goto out_free_mpol; |
2594 | ||
e9714acf | 2595 | if (new->vm_file) |
1da177e4 LT |
2596 | get_file(new->vm_file); |
2597 | ||
2598 | if (new->vm_ops && new->vm_ops->open) | |
2599 | new->vm_ops->open(new); | |
2600 | ||
2601 | if (new_below) | |
5beb4930 | 2602 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
1da177e4 LT |
2603 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
2604 | else | |
5beb4930 | 2605 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
1da177e4 | 2606 | |
5beb4930 RR |
2607 | /* Success. */ |
2608 | if (!err) | |
2609 | return 0; | |
2610 | ||
2611 | /* Clean everything up if vma_adjust failed. */ | |
58927533 RR |
2612 | if (new->vm_ops && new->vm_ops->close) |
2613 | new->vm_ops->close(new); | |
e9714acf | 2614 | if (new->vm_file) |
5beb4930 | 2615 | fput(new->vm_file); |
2aeadc30 | 2616 | unlink_anon_vmas(new); |
5beb4930 | 2617 | out_free_mpol: |
ef0855d3 | 2618 | mpol_put(vma_policy(new)); |
5beb4930 RR |
2619 | out_free_vma: |
2620 | kmem_cache_free(vm_area_cachep, new); | |
5beb4930 | 2621 | return err; |
1da177e4 LT |
2622 | } |
2623 | ||
659ace58 KM |
2624 | /* |
2625 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2626 | * either for the first part or the tail. | |
2627 | */ | |
2628 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, | |
2629 | unsigned long addr, int new_below) | |
2630 | { | |
2631 | if (mm->map_count >= sysctl_max_map_count) | |
2632 | return -ENOMEM; | |
2633 | ||
2634 | return __split_vma(mm, vma, addr, new_below); | |
2635 | } | |
2636 | ||
1da177e4 LT |
2637 | /* Munmap is split into 2 main parts -- this part which finds |
2638 | * what needs doing, and the areas themselves, which do the | |
2639 | * work. This now handles partial unmappings. | |
2640 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
2641 | */ | |
897ab3e0 MR |
2642 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, |
2643 | struct list_head *uf) | |
1da177e4 LT |
2644 | { |
2645 | unsigned long end; | |
146425a3 | 2646 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 | 2647 | |
de1741a1 | 2648 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) |
1da177e4 LT |
2649 | return -EINVAL; |
2650 | ||
cc71aba3 | 2651 | len = PAGE_ALIGN(len); |
2652 | if (len == 0) | |
1da177e4 LT |
2653 | return -EINVAL; |
2654 | ||
2655 | /* Find the first overlapping VMA */ | |
9be34c9d | 2656 | vma = find_vma(mm, start); |
146425a3 | 2657 | if (!vma) |
1da177e4 | 2658 | return 0; |
9be34c9d | 2659 | prev = vma->vm_prev; |
146425a3 | 2660 | /* we have start < vma->vm_end */ |
1da177e4 LT |
2661 | |
2662 | /* if it doesn't overlap, we have nothing.. */ | |
2663 | end = start + len; | |
146425a3 | 2664 | if (vma->vm_start >= end) |
1da177e4 LT |
2665 | return 0; |
2666 | ||
2667 | /* | |
2668 | * If we need to split any vma, do it now to save pain later. | |
2669 | * | |
2670 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2671 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2672 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2673 | */ | |
146425a3 | 2674 | if (start > vma->vm_start) { |
659ace58 KM |
2675 | int error; |
2676 | ||
2677 | /* | |
2678 | * Make sure that map_count on return from munmap() will | |
2679 | * not exceed its limit; but let map_count go just above | |
2680 | * its limit temporarily, to help free resources as expected. | |
2681 | */ | |
2682 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
2683 | return -ENOMEM; | |
2684 | ||
2685 | error = __split_vma(mm, vma, start, 0); | |
1da177e4 LT |
2686 | if (error) |
2687 | return error; | |
146425a3 | 2688 | prev = vma; |
1da177e4 LT |
2689 | } |
2690 | ||
2691 | /* Does it split the last one? */ | |
2692 | last = find_vma(mm, end); | |
2693 | if (last && end > last->vm_start) { | |
659ace58 | 2694 | int error = __split_vma(mm, last, end, 1); |
1da177e4 LT |
2695 | if (error) |
2696 | return error; | |
2697 | } | |
cc71aba3 | 2698 | vma = prev ? prev->vm_next : mm->mmap; |
1da177e4 | 2699 | |
2376dd7c AA |
2700 | if (unlikely(uf)) { |
2701 | /* | |
2702 | * If userfaultfd_unmap_prep returns an error the vmas | |
2703 | * will remain splitted, but userland will get a | |
2704 | * highly unexpected error anyway. This is no | |
2705 | * different than the case where the first of the two | |
2706 | * __split_vma fails, but we don't undo the first | |
2707 | * split, despite we could. This is unlikely enough | |
2708 | * failure that it's not worth optimizing it for. | |
2709 | */ | |
2710 | int error = userfaultfd_unmap_prep(vma, start, end, uf); | |
2711 | if (error) | |
2712 | return error; | |
2713 | } | |
2714 | ||
ba470de4 RR |
2715 | /* |
2716 | * unlock any mlock()ed ranges before detaching vmas | |
2717 | */ | |
2718 | if (mm->locked_vm) { | |
2719 | struct vm_area_struct *tmp = vma; | |
2720 | while (tmp && tmp->vm_start < end) { | |
2721 | if (tmp->vm_flags & VM_LOCKED) { | |
2722 | mm->locked_vm -= vma_pages(tmp); | |
2723 | munlock_vma_pages_all(tmp); | |
2724 | } | |
2725 | tmp = tmp->vm_next; | |
2726 | } | |
2727 | } | |
2728 | ||
1da177e4 LT |
2729 | /* |
2730 | * Remove the vma's, and unmap the actual pages | |
2731 | */ | |
146425a3 HD |
2732 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
2733 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 | 2734 | |
1de4fa14 DH |
2735 | arch_unmap(mm, vma, start, end); |
2736 | ||
1da177e4 | 2737 | /* Fix up all other VM information */ |
2c0b3814 | 2738 | remove_vma_list(mm, vma); |
1da177e4 LT |
2739 | |
2740 | return 0; | |
2741 | } | |
1da177e4 | 2742 | |
bfce281c | 2743 | int vm_munmap(unsigned long start, size_t len) |
1da177e4 LT |
2744 | { |
2745 | int ret; | |
bfce281c | 2746 | struct mm_struct *mm = current->mm; |
897ab3e0 | 2747 | LIST_HEAD(uf); |
1da177e4 | 2748 | |
ae798783 MH |
2749 | if (down_write_killable(&mm->mmap_sem)) |
2750 | return -EINTR; | |
2751 | ||
897ab3e0 | 2752 | ret = do_munmap(mm, start, len, &uf); |
1da177e4 | 2753 | up_write(&mm->mmap_sem); |
897ab3e0 | 2754 | userfaultfd_unmap_complete(mm, &uf); |
1da177e4 LT |
2755 | return ret; |
2756 | } | |
a46ef99d LT |
2757 | EXPORT_SYMBOL(vm_munmap); |
2758 | ||
2759 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2760 | { | |
2761 | profile_munmap(addr); | |
846b1a0f | 2762 | return vm_munmap(addr, len); |
a46ef99d | 2763 | } |
1da177e4 | 2764 | |
c8d78c18 KS |
2765 | |
2766 | /* | |
2767 | * Emulation of deprecated remap_file_pages() syscall. | |
2768 | */ | |
2769 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, | |
2770 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) | |
2771 | { | |
2772 | ||
2773 | struct mm_struct *mm = current->mm; | |
2774 | struct vm_area_struct *vma; | |
2775 | unsigned long populate = 0; | |
2776 | unsigned long ret = -EINVAL; | |
2777 | struct file *file; | |
2778 | ||
756a025f JP |
2779 | pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n", |
2780 | current->comm, current->pid); | |
c8d78c18 KS |
2781 | |
2782 | if (prot) | |
2783 | return ret; | |
2784 | start = start & PAGE_MASK; | |
2785 | size = size & PAGE_MASK; | |
2786 | ||
2787 | if (start + size <= start) | |
2788 | return ret; | |
2789 | ||
2790 | /* Does pgoff wrap? */ | |
2791 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) | |
2792 | return ret; | |
2793 | ||
dc0ef0df MH |
2794 | if (down_write_killable(&mm->mmap_sem)) |
2795 | return -EINTR; | |
2796 | ||
c8d78c18 KS |
2797 | vma = find_vma(mm, start); |
2798 | ||
2799 | if (!vma || !(vma->vm_flags & VM_SHARED)) | |
2800 | goto out; | |
2801 | ||
48f7df32 | 2802 | if (start < vma->vm_start) |
c8d78c18 KS |
2803 | goto out; |
2804 | ||
48f7df32 KS |
2805 | if (start + size > vma->vm_end) { |
2806 | struct vm_area_struct *next; | |
2807 | ||
2808 | for (next = vma->vm_next; next; next = next->vm_next) { | |
2809 | /* hole between vmas ? */ | |
2810 | if (next->vm_start != next->vm_prev->vm_end) | |
2811 | goto out; | |
2812 | ||
2813 | if (next->vm_file != vma->vm_file) | |
2814 | goto out; | |
2815 | ||
2816 | if (next->vm_flags != vma->vm_flags) | |
2817 | goto out; | |
2818 | ||
2819 | if (start + size <= next->vm_end) | |
2820 | break; | |
2821 | } | |
2822 | ||
2823 | if (!next) | |
2824 | goto out; | |
c8d78c18 KS |
2825 | } |
2826 | ||
2827 | prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; | |
2828 | prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; | |
2829 | prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; | |
2830 | ||
2831 | flags &= MAP_NONBLOCK; | |
2832 | flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; | |
2833 | if (vma->vm_flags & VM_LOCKED) { | |
48f7df32 | 2834 | struct vm_area_struct *tmp; |
c8d78c18 | 2835 | flags |= MAP_LOCKED; |
48f7df32 | 2836 | |
c8d78c18 | 2837 | /* drop PG_Mlocked flag for over-mapped range */ |
48f7df32 KS |
2838 | for (tmp = vma; tmp->vm_start >= start + size; |
2839 | tmp = tmp->vm_next) { | |
9a73f61b KS |
2840 | /* |
2841 | * Split pmd and munlock page on the border | |
2842 | * of the range. | |
2843 | */ | |
2844 | vma_adjust_trans_huge(tmp, start, start + size, 0); | |
2845 | ||
48f7df32 KS |
2846 | munlock_vma_pages_range(tmp, |
2847 | max(tmp->vm_start, start), | |
2848 | min(tmp->vm_end, start + size)); | |
2849 | } | |
c8d78c18 KS |
2850 | } |
2851 | ||
2852 | file = get_file(vma->vm_file); | |
2853 | ret = do_mmap_pgoff(vma->vm_file, start, size, | |
897ab3e0 | 2854 | prot, flags, pgoff, &populate, NULL); |
c8d78c18 KS |
2855 | fput(file); |
2856 | out: | |
2857 | up_write(&mm->mmap_sem); | |
2858 | if (populate) | |
2859 | mm_populate(ret, populate); | |
2860 | if (!IS_ERR_VALUE(ret)) | |
2861 | ret = 0; | |
2862 | return ret; | |
2863 | } | |
2864 | ||
1da177e4 LT |
2865 | static inline void verify_mm_writelocked(struct mm_struct *mm) |
2866 | { | |
a241ec65 | 2867 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
2868 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
2869 | WARN_ON(1); | |
2870 | up_read(&mm->mmap_sem); | |
2871 | } | |
2872 | #endif | |
2873 | } | |
2874 | ||
2875 | /* | |
2876 | * this is really a simplified "do_mmap". it only handles | |
2877 | * anonymous maps. eventually we may be able to do some | |
2878 | * brk-specific accounting here. | |
2879 | */ | |
897ab3e0 | 2880 | static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, struct list_head *uf) |
1da177e4 | 2881 | { |
cc71aba3 | 2882 | struct mm_struct *mm = current->mm; |
2883 | struct vm_area_struct *vma, *prev; | |
16e72e9b | 2884 | unsigned long len; |
cc71aba3 | 2885 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 2886 | pgoff_t pgoff = addr >> PAGE_SHIFT; |
3a459756 | 2887 | int error; |
1da177e4 | 2888 | |
ba093a6d KC |
2889 | len = PAGE_ALIGN(request); |
2890 | if (len < request) | |
2891 | return -ENOMEM; | |
1da177e4 | 2892 | if (!len) |
5d22fc25 | 2893 | return 0; |
1da177e4 | 2894 | |
16e72e9b DV |
2895 | /* Until we need other flags, refuse anything except VM_EXEC. */ |
2896 | if ((flags & (~VM_EXEC)) != 0) | |
2897 | return -EINVAL; | |
2898 | flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; | |
3a459756 | 2899 | |
2c6a1016 | 2900 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
de1741a1 | 2901 | if (offset_in_page(error)) |
3a459756 KK |
2902 | return error; |
2903 | ||
363ee17f DB |
2904 | error = mlock_future_check(mm, mm->def_flags, len); |
2905 | if (error) | |
2906 | return error; | |
1da177e4 LT |
2907 | |
2908 | /* | |
2909 | * mm->mmap_sem is required to protect against another thread | |
2910 | * changing the mappings in case we sleep. | |
2911 | */ | |
2912 | verify_mm_writelocked(mm); | |
2913 | ||
2914 | /* | |
2915 | * Clear old maps. this also does some error checking for us | |
2916 | */ | |
9fcd1457 RV |
2917 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
2918 | &rb_parent)) { | |
897ab3e0 | 2919 | if (do_munmap(mm, addr, len, uf)) |
1da177e4 | 2920 | return -ENOMEM; |
1da177e4 LT |
2921 | } |
2922 | ||
2923 | /* Check against address space limits *after* clearing old maps... */ | |
84638335 | 2924 | if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) |
1da177e4 LT |
2925 | return -ENOMEM; |
2926 | ||
2927 | if (mm->map_count > sysctl_max_map_count) | |
2928 | return -ENOMEM; | |
2929 | ||
191c5424 | 2930 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2931 | return -ENOMEM; |
2932 | ||
1da177e4 | 2933 | /* Can we just expand an old private anonymous mapping? */ |
ba470de4 | 2934 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
19a809af | 2935 | NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); |
ba470de4 | 2936 | if (vma) |
1da177e4 LT |
2937 | goto out; |
2938 | ||
2939 | /* | |
2940 | * create a vma struct for an anonymous mapping | |
2941 | */ | |
c5e3b83e | 2942 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2943 | if (!vma) { |
2944 | vm_unacct_memory(len >> PAGE_SHIFT); | |
2945 | return -ENOMEM; | |
2946 | } | |
1da177e4 | 2947 | |
5beb4930 | 2948 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
2949 | vma->vm_mm = mm; |
2950 | vma->vm_start = addr; | |
2951 | vma->vm_end = addr + len; | |
2952 | vma->vm_pgoff = pgoff; | |
2953 | vma->vm_flags = flags; | |
3ed75eb8 | 2954 | vma->vm_page_prot = vm_get_page_prot(flags); |
1da177e4 LT |
2955 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2956 | out: | |
3af9e859 | 2957 | perf_event_mmap(vma); |
1da177e4 | 2958 | mm->total_vm += len >> PAGE_SHIFT; |
84638335 | 2959 | mm->data_vm += len >> PAGE_SHIFT; |
128557ff ML |
2960 | if (flags & VM_LOCKED) |
2961 | mm->locked_vm += (len >> PAGE_SHIFT); | |
d9104d1c | 2962 | vma->vm_flags |= VM_SOFTDIRTY; |
5d22fc25 | 2963 | return 0; |
1da177e4 LT |
2964 | } |
2965 | ||
897ab3e0 | 2966 | static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf) |
16e72e9b | 2967 | { |
897ab3e0 | 2968 | return do_brk_flags(addr, len, 0, uf); |
16e72e9b DV |
2969 | } |
2970 | ||
2971 | int vm_brk_flags(unsigned long addr, unsigned long len, unsigned long flags) | |
e4eb1ff6 LT |
2972 | { |
2973 | struct mm_struct *mm = current->mm; | |
5d22fc25 | 2974 | int ret; |
128557ff | 2975 | bool populate; |
897ab3e0 | 2976 | LIST_HEAD(uf); |
e4eb1ff6 | 2977 | |
2d6c9282 MH |
2978 | if (down_write_killable(&mm->mmap_sem)) |
2979 | return -EINTR; | |
2980 | ||
897ab3e0 | 2981 | ret = do_brk_flags(addr, len, flags, &uf); |
128557ff | 2982 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
e4eb1ff6 | 2983 | up_write(&mm->mmap_sem); |
897ab3e0 | 2984 | userfaultfd_unmap_complete(mm, &uf); |
5d22fc25 | 2985 | if (populate && !ret) |
128557ff | 2986 | mm_populate(addr, len); |
e4eb1ff6 LT |
2987 | return ret; |
2988 | } | |
16e72e9b DV |
2989 | EXPORT_SYMBOL(vm_brk_flags); |
2990 | ||
2991 | int vm_brk(unsigned long addr, unsigned long len) | |
2992 | { | |
2993 | return vm_brk_flags(addr, len, 0); | |
2994 | } | |
e4eb1ff6 | 2995 | EXPORT_SYMBOL(vm_brk); |
1da177e4 LT |
2996 | |
2997 | /* Release all mmaps. */ | |
2998 | void exit_mmap(struct mm_struct *mm) | |
2999 | { | |
d16dfc55 | 3000 | struct mmu_gather tlb; |
ba470de4 | 3001 | struct vm_area_struct *vma; |
1da177e4 LT |
3002 | unsigned long nr_accounted = 0; |
3003 | ||
d6dd61c8 | 3004 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 3005 | mmu_notifier_release(mm); |
d6dd61c8 | 3006 | |
ba470de4 RR |
3007 | if (mm->locked_vm) { |
3008 | vma = mm->mmap; | |
3009 | while (vma) { | |
3010 | if (vma->vm_flags & VM_LOCKED) | |
3011 | munlock_vma_pages_all(vma); | |
3012 | vma = vma->vm_next; | |
3013 | } | |
3014 | } | |
9480c53e JF |
3015 | |
3016 | arch_exit_mmap(mm); | |
3017 | ||
ba470de4 | 3018 | vma = mm->mmap; |
9480c53e JF |
3019 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
3020 | return; | |
3021 | ||
1da177e4 | 3022 | lru_add_drain(); |
1da177e4 | 3023 | flush_cache_mm(mm); |
2b047252 | 3024 | tlb_gather_mmu(&tlb, mm, 0, -1); |
901608d9 | 3025 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
e0da382c | 3026 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
4f74d2c8 | 3027 | unmap_vmas(&tlb, vma, 0, -1); |
9ba69294 | 3028 | |
4837fe37 | 3029 | if (unlikely(mm_is_oom_victim(mm))) { |
21292580 AA |
3030 | /* |
3031 | * Wait for oom_reap_task() to stop working on this | |
3032 | * mm. Because MMF_OOM_SKIP is already set before | |
3033 | * calling down_read(), oom_reap_task() will not run | |
3034 | * on this "mm" post up_write(). | |
3035 | * | |
4837fe37 MH |
3036 | * mm_is_oom_victim() cannot be set from under us |
3037 | * either because victim->mm is already set to NULL | |
21292580 | 3038 | * under task_lock before calling mmput and oom_mm is |
4837fe37 | 3039 | * set not NULL by the OOM killer only if victim->mm |
21292580 AA |
3040 | * is found not NULL while holding the task_lock. |
3041 | */ | |
4837fe37 | 3042 | set_bit(MMF_OOM_SKIP, &mm->flags); |
21292580 AA |
3043 | down_write(&mm->mmap_sem); |
3044 | up_write(&mm->mmap_sem); | |
3045 | } | |
6ee8630e | 3046 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); |
853f5e26 | 3047 | tlb_finish_mmu(&tlb, 0, -1); |
1da177e4 | 3048 | |
1da177e4 | 3049 | /* |
8f4f8c16 HD |
3050 | * Walk the list again, actually closing and freeing it, |
3051 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 3052 | */ |
4f74d2c8 LT |
3053 | while (vma) { |
3054 | if (vma->vm_flags & VM_ACCOUNT) | |
3055 | nr_accounted += vma_pages(vma); | |
a8fb5618 | 3056 | vma = remove_vma(vma); |
4f74d2c8 LT |
3057 | } |
3058 | vm_unacct_memory(nr_accounted); | |
1da177e4 LT |
3059 | } |
3060 | ||
3061 | /* Insert vm structure into process list sorted by address | |
3062 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
c8c06efa | 3063 | * then i_mmap_rwsem is taken here. |
1da177e4 | 3064 | */ |
6597d783 | 3065 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 3066 | { |
6597d783 HD |
3067 | struct vm_area_struct *prev; |
3068 | struct rb_node **rb_link, *rb_parent; | |
1da177e4 | 3069 | |
c9d13f5f CG |
3070 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
3071 | &prev, &rb_link, &rb_parent)) | |
3072 | return -ENOMEM; | |
3073 | if ((vma->vm_flags & VM_ACCOUNT) && | |
3074 | security_vm_enough_memory_mm(mm, vma_pages(vma))) | |
3075 | return -ENOMEM; | |
3076 | ||
1da177e4 LT |
3077 | /* |
3078 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
3079 | * until its first write fault, when page's anon_vma and index | |
3080 | * are set. But now set the vm_pgoff it will almost certainly | |
3081 | * end up with (unless mremap moves it elsewhere before that | |
3082 | * first wfault), so /proc/pid/maps tells a consistent story. | |
3083 | * | |
3084 | * By setting it to reflect the virtual start address of the | |
3085 | * vma, merges and splits can happen in a seamless way, just | |
3086 | * using the existing file pgoff checks and manipulations. | |
3087 | * Similarly in do_mmap_pgoff and in do_brk. | |
3088 | */ | |
8a9cc3b5 | 3089 | if (vma_is_anonymous(vma)) { |
1da177e4 LT |
3090 | BUG_ON(vma->anon_vma); |
3091 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
3092 | } | |
2b144498 | 3093 | |
1da177e4 LT |
3094 | vma_link(mm, vma, prev, rb_link, rb_parent); |
3095 | return 0; | |
3096 | } | |
3097 | ||
3098 | /* | |
3099 | * Copy the vma structure to a new location in the same mm, | |
3100 | * prior to moving page table entries, to effect an mremap move. | |
3101 | */ | |
3102 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
3103 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
3104 | bool *need_rmap_locks) | |
1da177e4 LT |
3105 | { |
3106 | struct vm_area_struct *vma = *vmap; | |
3107 | unsigned long vma_start = vma->vm_start; | |
3108 | struct mm_struct *mm = vma->vm_mm; | |
3109 | struct vm_area_struct *new_vma, *prev; | |
3110 | struct rb_node **rb_link, *rb_parent; | |
948f017b | 3111 | bool faulted_in_anon_vma = true; |
1da177e4 LT |
3112 | |
3113 | /* | |
3114 | * If anonymous vma has not yet been faulted, update new pgoff | |
3115 | * to match new location, to increase its chance of merging. | |
3116 | */ | |
ce75799b | 3117 | if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { |
1da177e4 | 3118 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
3119 | faulted_in_anon_vma = false; |
3120 | } | |
1da177e4 | 3121 | |
6597d783 HD |
3122 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
3123 | return NULL; /* should never get here */ | |
1da177e4 | 3124 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
19a809af AA |
3125 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
3126 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
3127 | if (new_vma) { |
3128 | /* | |
3129 | * Source vma may have been merged into new_vma | |
3130 | */ | |
948f017b AA |
3131 | if (unlikely(vma_start >= new_vma->vm_start && |
3132 | vma_start < new_vma->vm_end)) { | |
3133 | /* | |
3134 | * The only way we can get a vma_merge with | |
3135 | * self during an mremap is if the vma hasn't | |
3136 | * been faulted in yet and we were allowed to | |
3137 | * reset the dst vma->vm_pgoff to the | |
3138 | * destination address of the mremap to allow | |
3139 | * the merge to happen. mremap must change the | |
3140 | * vm_pgoff linearity between src and dst vmas | |
3141 | * (in turn preventing a vma_merge) to be | |
3142 | * safe. It is only safe to keep the vm_pgoff | |
3143 | * linear if there are no pages mapped yet. | |
3144 | */ | |
81d1b09c | 3145 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
38a76013 | 3146 | *vmap = vma = new_vma; |
108d6642 | 3147 | } |
38a76013 | 3148 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 3149 | } else { |
e94b1766 | 3150 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
e3975891 CG |
3151 | if (!new_vma) |
3152 | goto out; | |
3153 | *new_vma = *vma; | |
3154 | new_vma->vm_start = addr; | |
3155 | new_vma->vm_end = addr + len; | |
3156 | new_vma->vm_pgoff = pgoff; | |
3157 | if (vma_dup_policy(vma, new_vma)) | |
3158 | goto out_free_vma; | |
3159 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); | |
3160 | if (anon_vma_clone(new_vma, vma)) | |
3161 | goto out_free_mempol; | |
3162 | if (new_vma->vm_file) | |
3163 | get_file(new_vma->vm_file); | |
3164 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
3165 | new_vma->vm_ops->open(new_vma); | |
3166 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
3167 | *need_rmap_locks = false; | |
1da177e4 LT |
3168 | } |
3169 | return new_vma; | |
5beb4930 | 3170 | |
e3975891 | 3171 | out_free_mempol: |
ef0855d3 | 3172 | mpol_put(vma_policy(new_vma)); |
e3975891 | 3173 | out_free_vma: |
5beb4930 | 3174 | kmem_cache_free(vm_area_cachep, new_vma); |
e3975891 | 3175 | out: |
5beb4930 | 3176 | return NULL; |
1da177e4 | 3177 | } |
119f657c | 3178 | |
3179 | /* | |
3180 | * Return true if the calling process may expand its vm space by the passed | |
3181 | * number of pages | |
3182 | */ | |
84638335 | 3183 | bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) |
119f657c | 3184 | { |
84638335 KK |
3185 | if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) |
3186 | return false; | |
119f657c | 3187 | |
d977d56c KK |
3188 | if (is_data_mapping(flags) && |
3189 | mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { | |
f4fcd558 KK |
3190 | /* Workaround for Valgrind */ |
3191 | if (rlimit(RLIMIT_DATA) == 0 && | |
3192 | mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) | |
3193 | return true; | |
57a7702b DW |
3194 | |
3195 | pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", | |
3196 | current->comm, current->pid, | |
3197 | (mm->data_vm + npages) << PAGE_SHIFT, | |
3198 | rlimit(RLIMIT_DATA), | |
3199 | ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); | |
3200 | ||
3201 | if (!ignore_rlimit_data) | |
d977d56c KK |
3202 | return false; |
3203 | } | |
119f657c | 3204 | |
84638335 KK |
3205 | return true; |
3206 | } | |
3207 | ||
3208 | void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) | |
3209 | { | |
3210 | mm->total_vm += npages; | |
3211 | ||
d977d56c | 3212 | if (is_exec_mapping(flags)) |
84638335 | 3213 | mm->exec_vm += npages; |
d977d56c | 3214 | else if (is_stack_mapping(flags)) |
84638335 | 3215 | mm->stack_vm += npages; |
d977d56c | 3216 | else if (is_data_mapping(flags)) |
84638335 | 3217 | mm->data_vm += npages; |
119f657c | 3218 | } |
fa5dc22f | 3219 | |
11bac800 | 3220 | static int special_mapping_fault(struct vm_fault *vmf); |
a62c34bd AL |
3221 | |
3222 | /* | |
3223 | * Having a close hook prevents vma merging regardless of flags. | |
3224 | */ | |
3225 | static void special_mapping_close(struct vm_area_struct *vma) | |
3226 | { | |
3227 | } | |
3228 | ||
3229 | static const char *special_mapping_name(struct vm_area_struct *vma) | |
3230 | { | |
3231 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; | |
3232 | } | |
3233 | ||
b059a453 DS |
3234 | static int special_mapping_mremap(struct vm_area_struct *new_vma) |
3235 | { | |
3236 | struct vm_special_mapping *sm = new_vma->vm_private_data; | |
3237 | ||
280e87e9 DS |
3238 | if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) |
3239 | return -EFAULT; | |
3240 | ||
b059a453 DS |
3241 | if (sm->mremap) |
3242 | return sm->mremap(sm, new_vma); | |
280e87e9 | 3243 | |
b059a453 DS |
3244 | return 0; |
3245 | } | |
3246 | ||
a62c34bd AL |
3247 | static const struct vm_operations_struct special_mapping_vmops = { |
3248 | .close = special_mapping_close, | |
3249 | .fault = special_mapping_fault, | |
b059a453 | 3250 | .mremap = special_mapping_mremap, |
a62c34bd AL |
3251 | .name = special_mapping_name, |
3252 | }; | |
3253 | ||
3254 | static const struct vm_operations_struct legacy_special_mapping_vmops = { | |
3255 | .close = special_mapping_close, | |
3256 | .fault = special_mapping_fault, | |
3257 | }; | |
fa5dc22f | 3258 | |
11bac800 | 3259 | static int special_mapping_fault(struct vm_fault *vmf) |
fa5dc22f | 3260 | { |
11bac800 | 3261 | struct vm_area_struct *vma = vmf->vma; |
b1d0e4f5 | 3262 | pgoff_t pgoff; |
fa5dc22f RM |
3263 | struct page **pages; |
3264 | ||
f872f540 | 3265 | if (vma->vm_ops == &legacy_special_mapping_vmops) { |
a62c34bd | 3266 | pages = vma->vm_private_data; |
f872f540 AL |
3267 | } else { |
3268 | struct vm_special_mapping *sm = vma->vm_private_data; | |
3269 | ||
3270 | if (sm->fault) | |
11bac800 | 3271 | return sm->fault(sm, vmf->vma, vmf); |
f872f540 AL |
3272 | |
3273 | pages = sm->pages; | |
3274 | } | |
a62c34bd | 3275 | |
8a9cc3b5 | 3276 | for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) |
b1d0e4f5 | 3277 | pgoff--; |
fa5dc22f RM |
3278 | |
3279 | if (*pages) { | |
3280 | struct page *page = *pages; | |
3281 | get_page(page); | |
b1d0e4f5 NP |
3282 | vmf->page = page; |
3283 | return 0; | |
fa5dc22f RM |
3284 | } |
3285 | ||
b1d0e4f5 | 3286 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
3287 | } |
3288 | ||
a62c34bd AL |
3289 | static struct vm_area_struct *__install_special_mapping( |
3290 | struct mm_struct *mm, | |
3291 | unsigned long addr, unsigned long len, | |
27f28b97 CG |
3292 | unsigned long vm_flags, void *priv, |
3293 | const struct vm_operations_struct *ops) | |
fa5dc22f | 3294 | { |
462e635e | 3295 | int ret; |
fa5dc22f RM |
3296 | struct vm_area_struct *vma; |
3297 | ||
3298 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
3299 | if (unlikely(vma == NULL)) | |
3935ed6a | 3300 | return ERR_PTR(-ENOMEM); |
fa5dc22f | 3301 | |
5beb4930 | 3302 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
fa5dc22f RM |
3303 | vma->vm_mm = mm; |
3304 | vma->vm_start = addr; | |
3305 | vma->vm_end = addr + len; | |
3306 | ||
d9104d1c | 3307 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; |
3ed75eb8 | 3308 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f | 3309 | |
a62c34bd AL |
3310 | vma->vm_ops = ops; |
3311 | vma->vm_private_data = priv; | |
fa5dc22f | 3312 | |
462e635e TO |
3313 | ret = insert_vm_struct(mm, vma); |
3314 | if (ret) | |
3315 | goto out; | |
fa5dc22f | 3316 | |
84638335 | 3317 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); |
fa5dc22f | 3318 | |
cdd6c482 | 3319 | perf_event_mmap(vma); |
089dd79d | 3320 | |
3935ed6a | 3321 | return vma; |
462e635e TO |
3322 | |
3323 | out: | |
3324 | kmem_cache_free(vm_area_cachep, vma); | |
3935ed6a SS |
3325 | return ERR_PTR(ret); |
3326 | } | |
3327 | ||
2eefd878 DS |
3328 | bool vma_is_special_mapping(const struct vm_area_struct *vma, |
3329 | const struct vm_special_mapping *sm) | |
3330 | { | |
3331 | return vma->vm_private_data == sm && | |
3332 | (vma->vm_ops == &special_mapping_vmops || | |
3333 | vma->vm_ops == &legacy_special_mapping_vmops); | |
3334 | } | |
3335 | ||
a62c34bd AL |
3336 | /* |
3337 | * Called with mm->mmap_sem held for writing. | |
3338 | * Insert a new vma covering the given region, with the given flags. | |
3339 | * Its pages are supplied by the given array of struct page *. | |
3340 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
3341 | * The region past the last page supplied will always produce SIGBUS. | |
3342 | * The array pointer and the pages it points to are assumed to stay alive | |
3343 | * for as long as this mapping might exist. | |
3344 | */ | |
3345 | struct vm_area_struct *_install_special_mapping( | |
3346 | struct mm_struct *mm, | |
3347 | unsigned long addr, unsigned long len, | |
3348 | unsigned long vm_flags, const struct vm_special_mapping *spec) | |
3349 | { | |
27f28b97 CG |
3350 | return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, |
3351 | &special_mapping_vmops); | |
a62c34bd AL |
3352 | } |
3353 | ||
3935ed6a SS |
3354 | int install_special_mapping(struct mm_struct *mm, |
3355 | unsigned long addr, unsigned long len, | |
3356 | unsigned long vm_flags, struct page **pages) | |
3357 | { | |
a62c34bd | 3358 | struct vm_area_struct *vma = __install_special_mapping( |
27f28b97 CG |
3359 | mm, addr, len, vm_flags, (void *)pages, |
3360 | &legacy_special_mapping_vmops); | |
3935ed6a | 3361 | |
14bd5b45 | 3362 | return PTR_ERR_OR_ZERO(vma); |
fa5dc22f | 3363 | } |
7906d00c AA |
3364 | |
3365 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3366 | ||
454ed842 | 3367 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3368 | { |
f808c13f | 3369 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { |
7906d00c AA |
3370 | /* |
3371 | * The LSB of head.next can't change from under us | |
3372 | * because we hold the mm_all_locks_mutex. | |
3373 | */ | |
572043c9 | 3374 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); |
7906d00c AA |
3375 | /* |
3376 | * We can safely modify head.next after taking the | |
5a505085 | 3377 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3378 | * the same anon_vma we won't take it again. |
3379 | * | |
3380 | * No need of atomic instructions here, head.next | |
3381 | * can't change from under us thanks to the | |
5a505085 | 3382 | * anon_vma->root->rwsem. |
7906d00c AA |
3383 | */ |
3384 | if (__test_and_set_bit(0, (unsigned long *) | |
f808c13f | 3385 | &anon_vma->root->rb_root.rb_root.rb_node)) |
7906d00c AA |
3386 | BUG(); |
3387 | } | |
3388 | } | |
3389 | ||
454ed842 | 3390 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3391 | { |
3392 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3393 | /* | |
3394 | * AS_MM_ALL_LOCKS can't change from under us because | |
3395 | * we hold the mm_all_locks_mutex. | |
3396 | * | |
3397 | * Operations on ->flags have to be atomic because | |
3398 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3399 | * mm_all_locks_mutex, there may be other cpus | |
3400 | * changing other bitflags in parallel to us. | |
3401 | */ | |
3402 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3403 | BUG(); | |
c8c06efa | 3404 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); |
7906d00c AA |
3405 | } |
3406 | } | |
3407 | ||
3408 | /* | |
3409 | * This operation locks against the VM for all pte/vma/mm related | |
3410 | * operations that could ever happen on a certain mm. This includes | |
3411 | * vmtruncate, try_to_unmap, and all page faults. | |
3412 | * | |
3413 | * The caller must take the mmap_sem in write mode before calling | |
3414 | * mm_take_all_locks(). The caller isn't allowed to release the | |
3415 | * mmap_sem until mm_drop_all_locks() returns. | |
3416 | * | |
3417 | * mmap_sem in write mode is required in order to block all operations | |
3418 | * that could modify pagetables and free pages without need of | |
27ba0644 | 3419 | * altering the vma layout. It's also needed in write mode to avoid new |
7906d00c AA |
3420 | * anon_vmas to be associated with existing vmas. |
3421 | * | |
3422 | * A single task can't take more than one mm_take_all_locks() in a row | |
3423 | * or it would deadlock. | |
3424 | * | |
bf181b9f | 3425 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3426 | * mapping->flags avoid to take the same lock twice, if more than one |
3427 | * vma in this mm is backed by the same anon_vma or address_space. | |
3428 | * | |
88f306b6 KS |
3429 | * We take locks in following order, accordingly to comment at beginning |
3430 | * of mm/rmap.c: | |
3431 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for | |
3432 | * hugetlb mapping); | |
3433 | * - all i_mmap_rwsem locks; | |
3434 | * - all anon_vma->rwseml | |
3435 | * | |
3436 | * We can take all locks within these types randomly because the VM code | |
3437 | * doesn't nest them and we protected from parallel mm_take_all_locks() by | |
3438 | * mm_all_locks_mutex. | |
7906d00c AA |
3439 | * |
3440 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3441 | * that may have to take thousand of locks. | |
3442 | * | |
3443 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3444 | */ | |
3445 | int mm_take_all_locks(struct mm_struct *mm) | |
3446 | { | |
3447 | struct vm_area_struct *vma; | |
5beb4930 | 3448 | struct anon_vma_chain *avc; |
7906d00c AA |
3449 | |
3450 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3451 | ||
3452 | mutex_lock(&mm_all_locks_mutex); | |
3453 | ||
3454 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3455 | if (signal_pending(current)) | |
3456 | goto out_unlock; | |
88f306b6 KS |
3457 | if (vma->vm_file && vma->vm_file->f_mapping && |
3458 | is_vm_hugetlb_page(vma)) | |
3459 | vm_lock_mapping(mm, vma->vm_file->f_mapping); | |
3460 | } | |
3461 | ||
3462 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3463 | if (signal_pending(current)) | |
3464 | goto out_unlock; | |
3465 | if (vma->vm_file && vma->vm_file->f_mapping && | |
3466 | !is_vm_hugetlb_page(vma)) | |
454ed842 | 3467 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3468 | } |
7cd5a02f PZ |
3469 | |
3470 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3471 | if (signal_pending(current)) | |
3472 | goto out_unlock; | |
3473 | if (vma->anon_vma) | |
5beb4930 RR |
3474 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3475 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3476 | } |
7cd5a02f | 3477 | |
584cff54 | 3478 | return 0; |
7906d00c AA |
3479 | |
3480 | out_unlock: | |
584cff54 KC |
3481 | mm_drop_all_locks(mm); |
3482 | return -EINTR; | |
7906d00c AA |
3483 | } |
3484 | ||
3485 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3486 | { | |
f808c13f | 3487 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { |
7906d00c AA |
3488 | /* |
3489 | * The LSB of head.next can't change to 0 from under | |
3490 | * us because we hold the mm_all_locks_mutex. | |
3491 | * | |
3492 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3493 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3494 | * never see our bitflag. |
3495 | * | |
3496 | * No need of atomic instructions here, head.next | |
3497 | * can't change from under us until we release the | |
5a505085 | 3498 | * anon_vma->root->rwsem. |
7906d00c AA |
3499 | */ |
3500 | if (!__test_and_clear_bit(0, (unsigned long *) | |
f808c13f | 3501 | &anon_vma->root->rb_root.rb_root.rb_node)) |
7906d00c | 3502 | BUG(); |
08b52706 | 3503 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3504 | } |
3505 | } | |
3506 | ||
3507 | static void vm_unlock_mapping(struct address_space *mapping) | |
3508 | { | |
3509 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3510 | /* | |
3511 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3512 | * because we hold the mm_all_locks_mutex. | |
3513 | */ | |
83cde9e8 | 3514 | i_mmap_unlock_write(mapping); |
7906d00c AA |
3515 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3516 | &mapping->flags)) | |
3517 | BUG(); | |
3518 | } | |
3519 | } | |
3520 | ||
3521 | /* | |
3522 | * The mmap_sem cannot be released by the caller until | |
3523 | * mm_drop_all_locks() returns. | |
3524 | */ | |
3525 | void mm_drop_all_locks(struct mm_struct *mm) | |
3526 | { | |
3527 | struct vm_area_struct *vma; | |
5beb4930 | 3528 | struct anon_vma_chain *avc; |
7906d00c AA |
3529 | |
3530 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3531 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); | |
3532 | ||
3533 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3534 | if (vma->anon_vma) | |
5beb4930 RR |
3535 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3536 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3537 | if (vma->vm_file && vma->vm_file->f_mapping) |
3538 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3539 | } | |
3540 | ||
3541 | mutex_unlock(&mm_all_locks_mutex); | |
3542 | } | |
8feae131 DH |
3543 | |
3544 | /* | |
3edf41d8 | 3545 | * initialise the percpu counter for VM |
8feae131 DH |
3546 | */ |
3547 | void __init mmap_init(void) | |
3548 | { | |
00a62ce9 KM |
3549 | int ret; |
3550 | ||
908c7f19 | 3551 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 3552 | VM_BUG_ON(ret); |
8feae131 | 3553 | } |
c9b1d098 AS |
3554 | |
3555 | /* | |
3556 | * Initialise sysctl_user_reserve_kbytes. | |
3557 | * | |
3558 | * This is intended to prevent a user from starting a single memory hogging | |
3559 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3560 | * mode. | |
3561 | * | |
3562 | * The default value is min(3% of free memory, 128MB) | |
3563 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3564 | */ | |
1640879a | 3565 | static int init_user_reserve(void) |
c9b1d098 AS |
3566 | { |
3567 | unsigned long free_kbytes; | |
3568 | ||
c41f012a | 3569 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
c9b1d098 AS |
3570 | |
3571 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3572 | return 0; | |
3573 | } | |
a64fb3cd | 3574 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
3575 | |
3576 | /* | |
3577 | * Initialise sysctl_admin_reserve_kbytes. | |
3578 | * | |
3579 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3580 | * to log in and kill a memory hogging process. | |
3581 | * | |
3582 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3583 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3584 | * only reserve 3% of free pages by default. | |
3585 | */ | |
1640879a | 3586 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3587 | { |
3588 | unsigned long free_kbytes; | |
3589 | ||
c41f012a | 3590 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
4eeab4f5 AS |
3591 | |
3592 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3593 | return 0; | |
3594 | } | |
a64fb3cd | 3595 | subsys_initcall(init_admin_reserve); |
1640879a AS |
3596 | |
3597 | /* | |
3598 | * Reinititalise user and admin reserves if memory is added or removed. | |
3599 | * | |
3600 | * The default user reserve max is 128MB, and the default max for the | |
3601 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3602 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3603 | * and tools like top. It may make sense to increase or even disable the | |
3604 | * reserve depending on the existence of swap or variations in the recovery | |
3605 | * tools. So, the admin may have changed them. | |
3606 | * | |
3607 | * If memory is added and the reserves have been eliminated or increased above | |
3608 | * the default max, then we'll trust the admin. | |
3609 | * | |
3610 | * If memory is removed and there isn't enough free memory, then we | |
3611 | * need to reset the reserves. | |
3612 | * | |
3613 | * Otherwise keep the reserve set by the admin. | |
3614 | */ | |
3615 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3616 | unsigned long action, void *data) | |
3617 | { | |
3618 | unsigned long tmp, free_kbytes; | |
3619 | ||
3620 | switch (action) { | |
3621 | case MEM_ONLINE: | |
3622 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3623 | tmp = sysctl_user_reserve_kbytes; | |
3624 | if (0 < tmp && tmp < (1UL << 17)) | |
3625 | init_user_reserve(); | |
3626 | ||
3627 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3628 | tmp = sysctl_admin_reserve_kbytes; | |
3629 | if (0 < tmp && tmp < (1UL << 13)) | |
3630 | init_admin_reserve(); | |
3631 | ||
3632 | break; | |
3633 | case MEM_OFFLINE: | |
c41f012a | 3634 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
1640879a AS |
3635 | |
3636 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3637 | init_user_reserve(); | |
3638 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3639 | sysctl_user_reserve_kbytes); | |
3640 | } | |
3641 | ||
3642 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3643 | init_admin_reserve(); | |
3644 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3645 | sysctl_admin_reserve_kbytes); | |
3646 | } | |
3647 | break; | |
3648 | default: | |
3649 | break; | |
3650 | } | |
3651 | return NOTIFY_OK; | |
3652 | } | |
3653 | ||
3654 | static struct notifier_block reserve_mem_nb = { | |
3655 | .notifier_call = reserve_mem_notifier, | |
3656 | }; | |
3657 | ||
3658 | static int __meminit init_reserve_notifier(void) | |
3659 | { | |
3660 | if (register_hotmemory_notifier(&reserve_mem_nb)) | |
b1de0d13 | 3661 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
1640879a AS |
3662 | |
3663 | return 0; | |
3664 | } | |
a64fb3cd | 3665 | subsys_initcall(init_reserve_notifier); |