Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * mm/mmap.c | |
4 | * | |
5 | * Written by obz. | |
6 | * | |
046c6884 | 7 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
8 | */ |
9 | ||
b1de0d13 MH |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
e8420a8e | 12 | #include <linux/kernel.h> |
1da177e4 | 13 | #include <linux/slab.h> |
4af3c9cc | 14 | #include <linux/backing-dev.h> |
1da177e4 | 15 | #include <linux/mm.h> |
17fca131 | 16 | #include <linux/mm_inline.h> |
1da177e4 LT |
17 | #include <linux/shm.h> |
18 | #include <linux/mman.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/swap.h> | |
21 | #include <linux/syscalls.h> | |
c59ede7b | 22 | #include <linux/capability.h> |
1da177e4 LT |
23 | #include <linux/init.h> |
24 | #include <linux/file.h> | |
25 | #include <linux/fs.h> | |
26 | #include <linux/personality.h> | |
27 | #include <linux/security.h> | |
28 | #include <linux/hugetlb.h> | |
c01d5b30 | 29 | #include <linux/shmem_fs.h> |
1da177e4 | 30 | #include <linux/profile.h> |
b95f1b31 | 31 | #include <linux/export.h> |
1da177e4 LT |
32 | #include <linux/mount.h> |
33 | #include <linux/mempolicy.h> | |
34 | #include <linux/rmap.h> | |
cddb8a5c | 35 | #include <linux/mmu_notifier.h> |
82f71ae4 | 36 | #include <linux/mmdebug.h> |
cdd6c482 | 37 | #include <linux/perf_event.h> |
120a795d | 38 | #include <linux/audit.h> |
b15d00b6 | 39 | #include <linux/khugepaged.h> |
2b144498 | 40 | #include <linux/uprobes.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> |
04f5866e | 48 | #include <linux/sched/mm.h> |
d7597f59 | 49 | #include <linux/ksm.h> |
1da177e4 | 50 | |
7c0f6ba6 | 51 | #include <linux/uaccess.h> |
1da177e4 LT |
52 | #include <asm/cacheflush.h> |
53 | #include <asm/tlb.h> | |
d6dd61c8 | 54 | #include <asm/mmu_context.h> |
1da177e4 | 55 | |
df529cab JK |
56 | #define CREATE_TRACE_POINTS |
57 | #include <trace/events/mmap.h> | |
58 | ||
42b77728 JB |
59 | #include "internal.h" |
60 | ||
3a459756 KK |
61 | #ifndef arch_mmap_check |
62 | #define arch_mmap_check(addr, len, flags) (0) | |
63 | #endif | |
64 | ||
d07e2259 DC |
65 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS |
66 | const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; | |
67 | const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; | |
68 | int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; | |
69 | #endif | |
70 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS | |
71 | const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; | |
72 | const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; | |
73 | int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; | |
74 | #endif | |
75 | ||
f4fcd558 | 76 | static bool ignore_rlimit_data; |
d977d56c | 77 | core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); |
d07e2259 | 78 | |
763ecb03 | 79 | static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, |
e0da382c | 80 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
763ecb03 | 81 | struct vm_area_struct *next, unsigned long start, |
68f48381 | 82 | unsigned long end, bool mm_wr_locked); |
e0da382c | 83 | |
64e45507 PF |
84 | static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) |
85 | { | |
86 | return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); | |
87 | } | |
88 | ||
89 | /* Update vma->vm_page_prot to reflect vma->vm_flags. */ | |
90 | void vma_set_page_prot(struct vm_area_struct *vma) | |
91 | { | |
92 | unsigned long vm_flags = vma->vm_flags; | |
6d2329f8 | 93 | pgprot_t vm_page_prot; |
64e45507 | 94 | |
6d2329f8 AA |
95 | vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); |
96 | if (vma_wants_writenotify(vma, vm_page_prot)) { | |
64e45507 | 97 | vm_flags &= ~VM_SHARED; |
6d2329f8 | 98 | vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); |
64e45507 | 99 | } |
c1e8d7c6 | 100 | /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ |
6d2329f8 | 101 | WRITE_ONCE(vma->vm_page_prot, vm_page_prot); |
64e45507 PF |
102 | } |
103 | ||
1da177e4 | 104 | /* |
c8c06efa | 105 | * Requires inode->i_mapping->i_mmap_rwsem |
1da177e4 LT |
106 | */ |
107 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
108 | struct file *file, struct address_space *mapping) | |
109 | { | |
1da177e4 | 110 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 111 | mapping_unmap_writable(mapping); |
1da177e4 LT |
112 | |
113 | flush_dcache_mmap_lock(mapping); | |
27ba0644 | 114 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
115 | flush_dcache_mmap_unlock(mapping); |
116 | } | |
117 | ||
118 | /* | |
6b2dbba8 | 119 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 120 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 121 | */ |
a8fb5618 | 122 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
123 | { |
124 | struct file *file = vma->vm_file; | |
125 | ||
1da177e4 LT |
126 | if (file) { |
127 | struct address_space *mapping = file->f_mapping; | |
83cde9e8 | 128 | i_mmap_lock_write(mapping); |
1da177e4 | 129 | __remove_shared_vm_struct(vma, file, mapping); |
83cde9e8 | 130 | i_mmap_unlock_write(mapping); |
1da177e4 | 131 | } |
a8fb5618 HD |
132 | } |
133 | ||
134 | /* | |
763ecb03 | 135 | * Close a vm structure and free it. |
a8fb5618 | 136 | */ |
0d2ebf9c | 137 | static void remove_vma(struct vm_area_struct *vma, bool unreachable) |
a8fb5618 | 138 | { |
a8fb5618 | 139 | might_sleep(); |
1da177e4 LT |
140 | if (vma->vm_ops && vma->vm_ops->close) |
141 | vma->vm_ops->close(vma); | |
e9714acf | 142 | if (vma->vm_file) |
a8fb5618 | 143 | fput(vma->vm_file); |
f0be3d32 | 144 | mpol_put(vma_policy(vma)); |
0d2ebf9c SB |
145 | if (unreachable) |
146 | __vm_area_free(vma); | |
147 | else | |
148 | vm_area_free(vma); | |
1da177e4 LT |
149 | } |
150 | ||
b62b633e LH |
151 | static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi, |
152 | unsigned long min) | |
153 | { | |
154 | return mas_prev(&vmi->mas, min); | |
155 | } | |
156 | ||
157 | static inline int vma_iter_clear_gfp(struct vma_iterator *vmi, | |
158 | unsigned long start, unsigned long end, gfp_t gfp) | |
159 | { | |
160 | vmi->mas.index = start; | |
161 | vmi->mas.last = end - 1; | |
162 | mas_store_gfp(&vmi->mas, NULL, gfp); | |
163 | if (unlikely(mas_is_err(&vmi->mas))) | |
164 | return -ENOMEM; | |
165 | ||
166 | return 0; | |
167 | } | |
168 | ||
2e7ce7d3 LH |
169 | /* |
170 | * check_brk_limits() - Use platform specific check of range & verify mlock | |
171 | * limits. | |
172 | * @addr: The address to check | |
173 | * @len: The size of increase. | |
174 | * | |
175 | * Return: 0 on success. | |
176 | */ | |
177 | static int check_brk_limits(unsigned long addr, unsigned long len) | |
178 | { | |
179 | unsigned long mapped_addr; | |
180 | ||
181 | mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); | |
182 | if (IS_ERR_VALUE(mapped_addr)) | |
183 | return mapped_addr; | |
184 | ||
3c54a298 LS |
185 | return mlock_future_check(current->mm, current->mm->def_flags, len) |
186 | ? 0 : -EAGAIN; | |
2e7ce7d3 | 187 | } |
92fed820 | 188 | static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma, |
763ecb03 | 189 | unsigned long addr, unsigned long request, unsigned long flags); |
6a6160a7 | 190 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 | 191 | { |
9bc8039e | 192 | unsigned long newbrk, oldbrk, origbrk; |
1da177e4 | 193 | struct mm_struct *mm = current->mm; |
2e7ce7d3 | 194 | struct vm_area_struct *brkvma, *next = NULL; |
a5b4592c | 195 | unsigned long min_brk; |
128557ff | 196 | bool populate; |
9bc8039e | 197 | bool downgraded = false; |
897ab3e0 | 198 | LIST_HEAD(uf); |
92fed820 | 199 | struct vma_iterator vmi; |
1da177e4 | 200 | |
d8ed45c5 | 201 | if (mmap_write_lock_killable(mm)) |
dc0ef0df | 202 | return -EINTR; |
1da177e4 | 203 | |
9bc8039e YS |
204 | origbrk = mm->brk; |
205 | ||
a5b4592c | 206 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
207 | /* |
208 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
209 | * randomize_va_space to 2, which will still cause mm->start_brk | |
210 | * to be arbitrarily shifted | |
211 | */ | |
4471a675 | 212 | if (current->brk_randomized) |
5520e894 JK |
213 | min_brk = mm->start_brk; |
214 | else | |
215 | min_brk = mm->end_data; | |
a5b4592c JK |
216 | #else |
217 | min_brk = mm->start_brk; | |
218 | #endif | |
219 | if (brk < min_brk) | |
1da177e4 | 220 | goto out; |
1e624196 RG |
221 | |
222 | /* | |
223 | * Check against rlimit here. If this check is done later after the test | |
224 | * of oldbrk with newbrk then it can escape the test and let the data | |
225 | * segment grow beyond its set limit the in case where the limit is | |
226 | * not page aligned -Ram Gupta | |
227 | */ | |
8764b338 CG |
228 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
229 | mm->end_data, mm->start_data)) | |
1e624196 RG |
230 | goto out; |
231 | ||
1da177e4 LT |
232 | newbrk = PAGE_ALIGN(brk); |
233 | oldbrk = PAGE_ALIGN(mm->brk); | |
9bc8039e YS |
234 | if (oldbrk == newbrk) { |
235 | mm->brk = brk; | |
236 | goto success; | |
237 | } | |
1da177e4 | 238 | |
9bc8039e YS |
239 | /* |
240 | * Always allow shrinking brk. | |
27b26701 | 241 | * do_vma_munmap() may downgrade mmap_lock to read. |
9bc8039e | 242 | */ |
1da177e4 | 243 | if (brk <= mm->brk) { |
9bc8039e YS |
244 | int ret; |
245 | ||
2e7ce7d3 | 246 | /* Search one past newbrk */ |
92fed820 LH |
247 | vma_iter_init(&vmi, mm, newbrk); |
248 | brkvma = vma_find(&vmi, oldbrk); | |
f5ad5083 | 249 | if (!brkvma || brkvma->vm_start >= oldbrk) |
2e7ce7d3 | 250 | goto out; /* mapping intersects with an existing non-brk vma. */ |
9bc8039e | 251 | /* |
2e7ce7d3 | 252 | * mm->brk must be protected by write mmap_lock. |
27b26701 LH |
253 | * do_vma_munmap() may downgrade the lock, so update it |
254 | * before calling do_vma_munmap(). | |
9bc8039e YS |
255 | */ |
256 | mm->brk = brk; | |
27b26701 | 257 | ret = do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true); |
2e7ce7d3 | 258 | if (ret == 1) { |
9bc8039e | 259 | downgraded = true; |
2e7ce7d3 LH |
260 | goto success; |
261 | } else if (!ret) | |
262 | goto success; | |
263 | ||
264 | mm->brk = origbrk; | |
265 | goto out; | |
1da177e4 LT |
266 | } |
267 | ||
2e7ce7d3 LH |
268 | if (check_brk_limits(oldbrk, newbrk - oldbrk)) |
269 | goto out; | |
270 | ||
271 | /* | |
272 | * Only check if the next VMA is within the stack_guard_gap of the | |
273 | * expansion area | |
274 | */ | |
92fed820 LH |
275 | vma_iter_init(&vmi, mm, oldbrk); |
276 | next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap); | |
1be7107f | 277 | if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) |
1da177e4 LT |
278 | goto out; |
279 | ||
92fed820 | 280 | brkvma = vma_prev_limit(&vmi, mm->start_brk); |
1da177e4 | 281 | /* Ok, looks good - let it rip. */ |
92fed820 | 282 | if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0) |
1da177e4 | 283 | goto out; |
2e7ce7d3 | 284 | |
1da177e4 | 285 | mm->brk = brk; |
9bc8039e YS |
286 | |
287 | success: | |
128557ff | 288 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
9bc8039e | 289 | if (downgraded) |
d8ed45c5 | 290 | mmap_read_unlock(mm); |
9bc8039e | 291 | else |
d8ed45c5 | 292 | mmap_write_unlock(mm); |
897ab3e0 | 293 | userfaultfd_unmap_complete(mm, &uf); |
128557ff ML |
294 | if (populate) |
295 | mm_populate(oldbrk, newbrk - oldbrk); | |
296 | return brk; | |
297 | ||
1da177e4 | 298 | out: |
d8ed45c5 | 299 | mmap_write_unlock(mm); |
b7204006 | 300 | return origbrk; |
1da177e4 LT |
301 | } |
302 | ||
d4af56c5 | 303 | #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) |
eafd4dc4 | 304 | static void validate_mm(struct mm_struct *mm) |
1da177e4 LT |
305 | { |
306 | int bug = 0; | |
307 | int i = 0; | |
763ecb03 | 308 | struct vm_area_struct *vma; |
b50e195f | 309 | VMA_ITERATOR(vmi, mm, 0); |
524e00b3 | 310 | |
b50e195f LH |
311 | mt_validate(&mm->mm_mt); |
312 | for_each_vma(vmi, vma) { | |
524e00b3 | 313 | #ifdef CONFIG_DEBUG_VM_RB |
12352d3c | 314 | struct anon_vma *anon_vma = vma->anon_vma; |
ed8ea815 | 315 | struct anon_vma_chain *avc; |
b50e195f LH |
316 | #endif |
317 | unsigned long vmi_start, vmi_end; | |
318 | bool warn = 0; | |
ff26f70f | 319 | |
b50e195f LH |
320 | vmi_start = vma_iter_addr(&vmi); |
321 | vmi_end = vma_iter_end(&vmi); | |
322 | if (VM_WARN_ON_ONCE_MM(vma->vm_end != vmi_end, mm)) | |
323 | warn = 1; | |
324 | ||
325 | if (VM_WARN_ON_ONCE_MM(vma->vm_start != vmi_start, mm)) | |
326 | warn = 1; | |
327 | ||
328 | if (warn) { | |
329 | pr_emerg("issue in %s\n", current->comm); | |
330 | dump_stack(); | |
331 | dump_vma(vma); | |
332 | pr_emerg("tree range: %px start %lx end %lx\n", vma, | |
333 | vmi_start, vmi_end - 1); | |
334 | vma_iter_dump_tree(&vmi); | |
335 | } | |
336 | ||
337 | #ifdef CONFIG_DEBUG_VM_RB | |
12352d3c KK |
338 | if (anon_vma) { |
339 | anon_vma_lock_read(anon_vma); | |
340 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
341 | anon_vma_interval_tree_verify(avc); | |
342 | anon_vma_unlock_read(anon_vma); | |
343 | } | |
524e00b3 | 344 | #endif |
1da177e4 LT |
345 | i++; |
346 | } | |
5a0768f6 | 347 | if (i != mm->map_count) { |
b50e195f | 348 | pr_emerg("map_count %d vma iterator %d\n", mm->map_count, i); |
5a0768f6 ML |
349 | bug = 1; |
350 | } | |
96dad67f | 351 | VM_BUG_ON_MM(bug, mm); |
1da177e4 | 352 | } |
524e00b3 LH |
353 | |
354 | #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */ | |
1da177e4 | 355 | #define validate_mm(mm) do { } while (0) |
524e00b3 | 356 | #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */ |
8f26e0b1 | 357 | |
bf181b9f ML |
358 | /* |
359 | * vma has some anon_vma assigned, and is already inserted on that | |
360 | * anon_vma's interval trees. | |
361 | * | |
362 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
363 | * vma must be removed from the anon_vma's interval trees using | |
364 | * anon_vma_interval_tree_pre_update_vma(). | |
365 | * | |
366 | * After the update, the vma will be reinserted using | |
367 | * anon_vma_interval_tree_post_update_vma(). | |
368 | * | |
c1e8d7c6 | 369 | * The entire update must be protected by exclusive mmap_lock and by |
bf181b9f ML |
370 | * the root anon_vma's mutex. |
371 | */ | |
372 | static inline void | |
373 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
374 | { | |
375 | struct anon_vma_chain *avc; | |
376 | ||
377 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
378 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
379 | } | |
380 | ||
381 | static inline void | |
382 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
383 | { | |
384 | struct anon_vma_chain *avc; | |
385 | ||
386 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
387 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
388 | } | |
389 | ||
e8420a8e CH |
390 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
391 | unsigned long addr, unsigned long end) | |
392 | { | |
2e3af1db | 393 | VMA_ITERATOR(vmi, mm, addr); |
e8420a8e | 394 | struct vm_area_struct *vma; |
2e3af1db | 395 | unsigned long nr_pages = 0; |
e8420a8e | 396 | |
2e3af1db MWO |
397 | for_each_vma_range(vmi, vma, end) { |
398 | unsigned long vm_start = max(addr, vma->vm_start); | |
399 | unsigned long vm_end = min(end, vma->vm_end); | |
e8420a8e | 400 | |
2e3af1db | 401 | nr_pages += PHYS_PFN(vm_end - vm_start); |
e8420a8e CH |
402 | } |
403 | ||
404 | return nr_pages; | |
405 | } | |
406 | ||
c154124f LH |
407 | static void __vma_link_file(struct vm_area_struct *vma, |
408 | struct address_space *mapping) | |
1da177e4 | 409 | { |
c154124f LH |
410 | if (vma->vm_flags & VM_SHARED) |
411 | mapping_allow_writable(mapping); | |
1da177e4 | 412 | |
c154124f LH |
413 | flush_dcache_mmap_lock(mapping); |
414 | vma_interval_tree_insert(vma, &mapping->i_mmap); | |
415 | flush_dcache_mmap_unlock(mapping); | |
1da177e4 LT |
416 | } |
417 | ||
763ecb03 | 418 | static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 419 | { |
79e4f2ca | 420 | VMA_ITERATOR(vmi, mm, 0); |
1da177e4 LT |
421 | struct address_space *mapping = NULL; |
422 | ||
79e4f2ca | 423 | if (vma_iter_prealloc(&vmi)) |
d4af56c5 LH |
424 | return -ENOMEM; |
425 | ||
64ac4940 | 426 | if (vma->vm_file) { |
1da177e4 | 427 | mapping = vma->vm_file->f_mapping; |
83cde9e8 | 428 | i_mmap_lock_write(mapping); |
64ac4940 | 429 | } |
1da177e4 | 430 | |
79e4f2ca | 431 | vma_iter_store(&vmi, vma); |
1da177e4 | 432 | |
c154124f LH |
433 | if (mapping) { |
434 | __vma_link_file(vma, mapping); | |
83cde9e8 | 435 | i_mmap_unlock_write(mapping); |
c154124f | 436 | } |
1da177e4 LT |
437 | |
438 | mm->map_count++; | |
439 | validate_mm(mm); | |
d4af56c5 | 440 | return 0; |
1da177e4 LT |
441 | } |
442 | ||
68cefec5 LH |
443 | /* |
444 | * init_multi_vma_prep() - Initializer for struct vma_prepare | |
445 | * @vp: The vma_prepare struct | |
446 | * @vma: The vma that will be altered once locked | |
447 | * @next: The next vma if it is to be adjusted | |
448 | * @remove: The first vma to be removed | |
449 | * @remove2: The second vma to be removed | |
450 | */ | |
451 | static inline void init_multi_vma_prep(struct vma_prepare *vp, | |
452 | struct vm_area_struct *vma, struct vm_area_struct *next, | |
453 | struct vm_area_struct *remove, struct vm_area_struct *remove2) | |
454 | { | |
455 | memset(vp, 0, sizeof(struct vma_prepare)); | |
456 | vp->vma = vma; | |
457 | vp->anon_vma = vma->anon_vma; | |
458 | vp->remove = remove; | |
459 | vp->remove2 = remove2; | |
460 | vp->adj_next = next; | |
461 | if (!vp->anon_vma && next) | |
462 | vp->anon_vma = next->anon_vma; | |
463 | ||
464 | vp->file = vma->vm_file; | |
465 | if (vp->file) | |
466 | vp->mapping = vma->vm_file->f_mapping; | |
467 | ||
468 | } | |
469 | ||
470 | /* | |
471 | * init_vma_prep() - Initializer wrapper for vma_prepare struct | |
472 | * @vp: The vma_prepare struct | |
473 | * @vma: The vma that will be altered once locked | |
474 | */ | |
475 | static inline void init_vma_prep(struct vma_prepare *vp, | |
476 | struct vm_area_struct *vma) | |
477 | { | |
478 | init_multi_vma_prep(vp, vma, NULL, NULL, NULL); | |
479 | } | |
480 | ||
481 | ||
440703e0 LH |
482 | /* |
483 | * vma_prepare() - Helper function for handling locking VMAs prior to altering | |
484 | * @vp: The initialized vma_prepare struct | |
485 | */ | |
486 | static inline void vma_prepare(struct vma_prepare *vp) | |
487 | { | |
10fca64a SB |
488 | vma_start_write(vp->vma); |
489 | if (vp->adj_next) | |
490 | vma_start_write(vp->adj_next); | |
491 | /* vp->insert is always a newly created VMA, no need for locking */ | |
492 | if (vp->remove) | |
493 | vma_start_write(vp->remove); | |
494 | if (vp->remove2) | |
495 | vma_start_write(vp->remove2); | |
496 | ||
440703e0 LH |
497 | if (vp->file) { |
498 | uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); | |
499 | ||
500 | if (vp->adj_next) | |
501 | uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, | |
502 | vp->adj_next->vm_end); | |
503 | ||
504 | i_mmap_lock_write(vp->mapping); | |
505 | if (vp->insert && vp->insert->vm_file) { | |
506 | /* | |
507 | * Put into interval tree now, so instantiated pages | |
508 | * are visible to arm/parisc __flush_dcache_page | |
509 | * throughout; but we cannot insert into address | |
510 | * space until vma start or end is updated. | |
511 | */ | |
512 | __vma_link_file(vp->insert, | |
513 | vp->insert->vm_file->f_mapping); | |
514 | } | |
515 | } | |
516 | ||
517 | if (vp->anon_vma) { | |
518 | anon_vma_lock_write(vp->anon_vma); | |
519 | anon_vma_interval_tree_pre_update_vma(vp->vma); | |
520 | if (vp->adj_next) | |
521 | anon_vma_interval_tree_pre_update_vma(vp->adj_next); | |
522 | } | |
523 | ||
524 | if (vp->file) { | |
525 | flush_dcache_mmap_lock(vp->mapping); | |
526 | vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); | |
527 | if (vp->adj_next) | |
528 | vma_interval_tree_remove(vp->adj_next, | |
529 | &vp->mapping->i_mmap); | |
530 | } | |
531 | ||
532 | } | |
533 | ||
534 | /* | |
535 | * vma_complete- Helper function for handling the unlocking after altering VMAs, | |
536 | * or for inserting a VMA. | |
537 | * | |
538 | * @vp: The vma_prepare struct | |
539 | * @vmi: The vma iterator | |
540 | * @mm: The mm_struct | |
541 | */ | |
542 | static inline void vma_complete(struct vma_prepare *vp, | |
543 | struct vma_iterator *vmi, struct mm_struct *mm) | |
544 | { | |
545 | if (vp->file) { | |
546 | if (vp->adj_next) | |
547 | vma_interval_tree_insert(vp->adj_next, | |
548 | &vp->mapping->i_mmap); | |
549 | vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); | |
550 | flush_dcache_mmap_unlock(vp->mapping); | |
551 | } | |
552 | ||
553 | if (vp->remove && vp->file) { | |
554 | __remove_shared_vm_struct(vp->remove, vp->file, vp->mapping); | |
555 | if (vp->remove2) | |
556 | __remove_shared_vm_struct(vp->remove2, vp->file, | |
557 | vp->mapping); | |
558 | } else if (vp->insert) { | |
559 | /* | |
560 | * split_vma has split insert from vma, and needs | |
561 | * us to insert it before dropping the locks | |
562 | * (it may either follow vma or precede it). | |
563 | */ | |
564 | vma_iter_store(vmi, vp->insert); | |
565 | mm->map_count++; | |
566 | } | |
567 | ||
568 | if (vp->anon_vma) { | |
569 | anon_vma_interval_tree_post_update_vma(vp->vma); | |
570 | if (vp->adj_next) | |
571 | anon_vma_interval_tree_post_update_vma(vp->adj_next); | |
572 | anon_vma_unlock_write(vp->anon_vma); | |
573 | } | |
574 | ||
575 | if (vp->file) { | |
576 | i_mmap_unlock_write(vp->mapping); | |
577 | uprobe_mmap(vp->vma); | |
578 | ||
579 | if (vp->adj_next) | |
580 | uprobe_mmap(vp->adj_next); | |
581 | } | |
582 | ||
583 | if (vp->remove) { | |
584 | again: | |
457f67be | 585 | vma_mark_detached(vp->remove, true); |
440703e0 LH |
586 | if (vp->file) { |
587 | uprobe_munmap(vp->remove, vp->remove->vm_start, | |
588 | vp->remove->vm_end); | |
589 | fput(vp->file); | |
590 | } | |
591 | if (vp->remove->anon_vma) | |
592 | anon_vma_merge(vp->vma, vp->remove); | |
593 | mm->map_count--; | |
594 | mpol_put(vma_policy(vp->remove)); | |
595 | if (!vp->remove2) | |
596 | WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); | |
597 | vm_area_free(vp->remove); | |
598 | ||
599 | /* | |
600 | * In mprotect's case 6 (see comments on vma_merge), | |
5ff783f1 | 601 | * we are removing both mid and next vmas |
440703e0 LH |
602 | */ |
603 | if (vp->remove2) { | |
604 | vp->remove = vp->remove2; | |
605 | vp->remove2 = NULL; | |
606 | goto again; | |
607 | } | |
608 | } | |
609 | if (vp->insert && vp->file) | |
610 | uprobe_mmap(vp->insert); | |
611 | } | |
612 | ||
04241ffe LH |
613 | /* |
614 | * dup_anon_vma() - Helper function to duplicate anon_vma | |
615 | * @dst: The destination VMA | |
616 | * @src: The source VMA | |
617 | * | |
618 | * Returns: 0 on success. | |
619 | */ | |
620 | static inline int dup_anon_vma(struct vm_area_struct *dst, | |
621 | struct vm_area_struct *src) | |
622 | { | |
623 | /* | |
624 | * Easily overlooked: when mprotect shifts the boundary, make sure the | |
625 | * expanding vma has anon_vma set if the shrinking vma had, to cover any | |
626 | * anon pages imported. | |
627 | */ | |
628 | if (src->anon_vma && !dst->anon_vma) { | |
629 | dst->anon_vma = src->anon_vma; | |
630 | return anon_vma_clone(dst, src); | |
631 | } | |
632 | ||
633 | return 0; | |
634 | } | |
635 | ||
9303d3e1 LH |
636 | /* |
637 | * vma_expand - Expand an existing VMA | |
638 | * | |
639 | * @vmi: The vma iterator | |
640 | * @vma: The vma to expand | |
641 | * @start: The start of the vma | |
642 | * @end: The exclusive end of the vma | |
643 | * @pgoff: The page offset of vma | |
644 | * @next: The current of next vma. | |
645 | * | |
646 | * Expand @vma to @start and @end. Can expand off the start and end. Will | |
647 | * expand over @next if it's different from @vma and @end == @next->vm_end. | |
648 | * Checking if the @vma can expand and merge with @next needs to be handled by | |
649 | * the caller. | |
650 | * | |
651 | * Returns: 0 on success | |
652 | */ | |
7c9813e8 LH |
653 | int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, |
654 | unsigned long start, unsigned long end, pgoff_t pgoff, | |
655 | struct vm_area_struct *next) | |
9303d3e1 | 656 | { |
68cefec5 | 657 | bool remove_next = false; |
9303d3e1 LH |
658 | struct vma_prepare vp; |
659 | ||
9303d3e1 | 660 | if (next && (vma != next) && (end == next->vm_end)) { |
04241ffe | 661 | int ret; |
9303d3e1 | 662 | |
04241ffe LH |
663 | remove_next = true; |
664 | ret = dup_anon_vma(vma, next); | |
665 | if (ret) | |
666 | return ret; | |
9303d3e1 LH |
667 | } |
668 | ||
68cefec5 | 669 | init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); |
9303d3e1 LH |
670 | /* Not merging but overwriting any part of next is not handled. */ |
671 | VM_WARN_ON(next && !vp.remove && | |
672 | next != vma && end > next->vm_start); | |
673 | /* Only handles expanding */ | |
674 | VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); | |
675 | ||
676 | if (vma_iter_prealloc(vmi)) | |
677 | goto nomem; | |
678 | ||
ccf1d78d | 679 | vma_prepare(&vp); |
9303d3e1 | 680 | vma_adjust_trans_huge(vma, start, end, 0); |
9303d3e1 LH |
681 | /* VMA iterator points to previous, so set to start if necessary */ |
682 | if (vma_iter_addr(vmi) != start) | |
683 | vma_iter_set(vmi, start); | |
684 | ||
9303d3e1 LH |
685 | vma->vm_start = start; |
686 | vma->vm_end = end; | |
687 | vma->vm_pgoff = pgoff; | |
688 | /* Note: mas must be pointing to the expanding VMA */ | |
689 | vma_iter_store(vmi, vma); | |
690 | ||
691 | vma_complete(&vp, vmi, vma->vm_mm); | |
692 | validate_mm(vma->vm_mm); | |
693 | return 0; | |
694 | ||
695 | nomem: | |
696 | return -ENOMEM; | |
697 | } | |
cf51e86d LH |
698 | |
699 | /* | |
700 | * vma_shrink() - Reduce an existing VMAs memory area | |
701 | * @vmi: The vma iterator | |
702 | * @vma: The VMA to modify | |
703 | * @start: The new start | |
704 | * @end: The new end | |
705 | * | |
706 | * Returns: 0 on success, -ENOMEM otherwise | |
707 | */ | |
708 | int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, | |
709 | unsigned long start, unsigned long end, pgoff_t pgoff) | |
710 | { | |
711 | struct vma_prepare vp; | |
712 | ||
713 | WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); | |
714 | ||
715 | if (vma_iter_prealloc(vmi)) | |
716 | return -ENOMEM; | |
717 | ||
718 | init_vma_prep(&vp, vma); | |
cf51e86d | 719 | vma_prepare(&vp); |
ccf1d78d | 720 | vma_adjust_trans_huge(vma, start, end, 0); |
cf51e86d LH |
721 | |
722 | if (vma->vm_start < start) | |
723 | vma_iter_clear(vmi, vma->vm_start, start); | |
724 | ||
725 | if (vma->vm_end > end) | |
726 | vma_iter_clear(vmi, end, vma->vm_end); | |
727 | ||
728 | vma->vm_start = start; | |
729 | vma->vm_end = end; | |
730 | vma->vm_pgoff = pgoff; | |
731 | vma_complete(&vp, vmi, vma->vm_mm); | |
732 | validate_mm(vma->vm_mm); | |
733 | return 0; | |
734 | } | |
735 | ||
1da177e4 LT |
736 | /* |
737 | * If the vma has a ->close operation then the driver probably needs to release | |
714965ca VB |
738 | * per-vma resources, so we don't attempt to merge those if the caller indicates |
739 | * the current vma may be removed as part of the merge. | |
1da177e4 | 740 | */ |
2dbf4010 VB |
741 | static inline bool is_mergeable_vma(struct vm_area_struct *vma, |
742 | struct file *file, unsigned long vm_flags, | |
743 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, | |
714965ca | 744 | struct anon_vma_name *anon_name, bool may_remove_vma) |
1da177e4 | 745 | { |
34228d47 CG |
746 | /* |
747 | * VM_SOFTDIRTY should not prevent from VMA merging, if we | |
748 | * match the flags but dirty bit -- the caller should mark | |
749 | * merged VMA as dirty. If dirty bit won't be excluded from | |
8bb4e7a2 | 750 | * comparison, we increase pressure on the memory system forcing |
34228d47 CG |
751 | * the kernel to generate new VMAs when old one could be |
752 | * extended instead. | |
753 | */ | |
754 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) | |
2dbf4010 | 755 | return false; |
1da177e4 | 756 | if (vma->vm_file != file) |
2dbf4010 | 757 | return false; |
714965ca | 758 | if (may_remove_vma && vma->vm_ops && vma->vm_ops->close) |
2dbf4010 | 759 | return false; |
19a809af | 760 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) |
2dbf4010 | 761 | return false; |
5c26f6ac | 762 | if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) |
2dbf4010 VB |
763 | return false; |
764 | return true; | |
1da177e4 LT |
765 | } |
766 | ||
2dbf4010 VB |
767 | static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, |
768 | struct anon_vma *anon_vma2, struct vm_area_struct *vma) | |
1da177e4 | 769 | { |
965f55de SL |
770 | /* |
771 | * The list_is_singular() test is to avoid merging VMA cloned from | |
772 | * parents. This can improve scalability caused by anon_vma lock. | |
773 | */ | |
774 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
775 | list_is_singular(&vma->anon_vma_chain))) | |
2dbf4010 | 776 | return true; |
965f55de | 777 | return anon_vma1 == anon_vma2; |
1da177e4 LT |
778 | } |
779 | ||
780 | /* | |
781 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
782 | * in front of (at a lower virtual address and file offset than) the vma. | |
783 | * | |
784 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
785 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
786 | * | |
787 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
45e55300 | 788 | * indices (16TB on ia32) because do_mmap() does not permit mmap's which |
1da177e4 | 789 | * wrap, nor mmaps which cover the final page at index -1UL. |
714965ca VB |
790 | * |
791 | * We assume the vma may be removed as part of the merge. | |
1da177e4 | 792 | */ |
2dbf4010 | 793 | static bool |
1da177e4 | 794 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, |
2dbf4010 VB |
795 | struct anon_vma *anon_vma, struct file *file, |
796 | pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, | |
797 | struct anon_vma_name *anon_name) | |
1da177e4 | 798 | { |
714965ca | 799 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) && |
965f55de | 800 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 801 | if (vma->vm_pgoff == vm_pgoff) |
2dbf4010 | 802 | return true; |
1da177e4 | 803 | } |
2dbf4010 | 804 | return false; |
1da177e4 LT |
805 | } |
806 | ||
807 | /* | |
808 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
809 | * beyond (at a higher virtual address and file offset than) the vma. | |
810 | * | |
811 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
812 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
714965ca VB |
813 | * |
814 | * We assume that vma is not removed as part of the merge. | |
1da177e4 | 815 | */ |
2dbf4010 | 816 | static bool |
1da177e4 | 817 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, |
2dbf4010 VB |
818 | struct anon_vma *anon_vma, struct file *file, |
819 | pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, | |
820 | struct anon_vma_name *anon_name) | |
1da177e4 | 821 | { |
714965ca | 822 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) && |
965f55de | 823 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 824 | pgoff_t vm_pglen; |
d6e93217 | 825 | vm_pglen = vma_pages(vma); |
1da177e4 | 826 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
2dbf4010 | 827 | return true; |
1da177e4 | 828 | } |
2dbf4010 | 829 | return false; |
1da177e4 LT |
830 | } |
831 | ||
832 | /* | |
9a10064f CC |
833 | * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), |
834 | * figure out whether that can be merged with its predecessor or its | |
835 | * successor. Or both (it neatly fills a hole). | |
1da177e4 LT |
836 | * |
837 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
838 | * certain not to be mapped by the time vma_merge is called; but when | |
839 | * called for mprotect, it is certain to be already mapped (either at | |
840 | * an offset within prev, or at the start of next), and the flags of | |
841 | * this area are about to be changed to vm_flags - and the no-change | |
842 | * case has already been eliminated. | |
843 | * | |
fcfccd91 | 844 | * The following mprotect cases have to be considered, where **** is |
1da177e4 | 845 | * the area passed down from mprotect_fixup, never extending beyond one |
fcfccd91 LS |
846 | * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts |
847 | * at the same address as **** and is of the same or larger span, and | |
848 | * NNNN the next vma after ****: | |
1da177e4 | 849 | * |
fcfccd91 LS |
850 | * **** **** **** |
851 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC | |
5d42ab29 | 852 | * cannot merge might become might become |
fcfccd91 | 853 | * PPNNNNNNNNNN PPPPPPPPPPCC |
5d42ab29 WY |
854 | * mmap, brk or case 4 below case 5 below |
855 | * mremap move: | |
fcfccd91 LS |
856 | * **** **** |
857 | * PPPP NNNN PPPPCCCCNNNN | |
5d42ab29 WY |
858 | * might become might become |
859 | * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or | |
fcfccd91 LS |
860 | * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or |
861 | * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8 | |
1da177e4 | 862 | * |
fcfccd91 LS |
863 | * It is important for case 8 that the vma CCCC overlapping the |
864 | * region **** is never going to extended over NNNN. Instead NNNN must | |
865 | * be extended in region **** and CCCC must be removed. This way in | |
0503ea8f | 866 | * all cases where vma_merge succeeds, the moment vma_merge drops the |
e86f15ee AA |
867 | * rmap_locks, the properties of the merged vma will be already |
868 | * correct for the whole merged range. Some of those properties like | |
869 | * vm_page_prot/vm_flags may be accessed by rmap_walks and they must | |
870 | * be correct for the whole merged range immediately after the | |
fcfccd91 LS |
871 | * rmap_locks are released. Otherwise if NNNN would be removed and |
872 | * CCCC would be extended over the NNNN range, remove_migration_ptes | |
e86f15ee | 873 | * or other rmap walkers (if working on addresses beyond the "end" |
fcfccd91 LS |
874 | * parameter) may establish ptes with the wrong permissions of CCCC |
875 | * instead of the right permissions of NNNN. | |
0503ea8f LH |
876 | * |
877 | * In the code below: | |
878 | * PPPP is represented by *prev | |
fcfccd91 LS |
879 | * CCCC is represented by *curr or not represented at all (NULL) |
880 | * NNNN is represented by *next or not represented at all (NULL) | |
881 | * **** is not represented - it will be merged and the vma containing the | |
9e8a39d2 | 882 | * area is returned, or the function will return NULL |
1da177e4 | 883 | */ |
9760ebff | 884 | struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, |
1da177e4 LT |
885 | struct vm_area_struct *prev, unsigned long addr, |
886 | unsigned long end, unsigned long vm_flags, | |
cc71aba3 | 887 | struct anon_vma *anon_vma, struct file *file, |
19a809af | 888 | pgoff_t pgoff, struct mempolicy *policy, |
9a10064f | 889 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, |
5c26f6ac | 890 | struct anon_vma_name *anon_name) |
1da177e4 | 891 | { |
b0729ae0 | 892 | struct vm_area_struct *curr, *next, *res; |
0503ea8f | 893 | struct vm_area_struct *vma, *adjust, *remove, *remove2; |
0173db4f LS |
894 | struct vma_prepare vp; |
895 | pgoff_t vma_pgoff; | |
896 | int err = 0; | |
eef19944 JM |
897 | bool merge_prev = false; |
898 | bool merge_next = false; | |
0503ea8f | 899 | bool vma_expanded = false; |
0503ea8f | 900 | unsigned long vma_start = addr; |
0503ea8f | 901 | unsigned long vma_end = end; |
0173db4f | 902 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; |
1e76454f | 903 | long adj_start = 0; |
1da177e4 | 904 | |
0503ea8f | 905 | validate_mm(mm); |
1da177e4 LT |
906 | /* |
907 | * We later require that vma->vm_flags == vm_flags, | |
908 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
909 | */ | |
910 | if (vm_flags & VM_SPECIAL) | |
911 | return NULL; | |
912 | ||
00cd00a6 LS |
913 | /* Does the input range span an existing VMA? (cases 5 - 8) */ |
914 | curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end); | |
1da177e4 | 915 | |
00cd00a6 LS |
916 | if (!curr || /* cases 1 - 4 */ |
917 | end == curr->vm_end) /* cases 6 - 8, adjacent VMA */ | |
918 | next = vma_lookup(mm, end); | |
919 | else | |
920 | next = NULL; /* case 5 */ | |
e86f15ee | 921 | |
0503ea8f | 922 | if (prev) { |
0503ea8f LH |
923 | vma_start = prev->vm_start; |
924 | vma_pgoff = prev->vm_pgoff; | |
0173db4f | 925 | |
0503ea8f | 926 | /* Can we merge the predecessor? */ |
0173db4f | 927 | if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy) |
0503ea8f | 928 | && can_vma_merge_after(prev, vm_flags, anon_vma, file, |
0173db4f | 929 | pgoff, vm_userfaultfd_ctx, anon_name)) { |
0503ea8f | 930 | merge_prev = true; |
18b098af | 931 | vma_prev(vmi); |
0503ea8f | 932 | } |
1da177e4 | 933 | } |
b0729ae0 | 934 | |
eef19944 | 935 | /* Can we merge the successor? */ |
00cd00a6 | 936 | if (next && mpol_equal(policy, vma_policy(next)) && |
0173db4f | 937 | can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen, |
00cd00a6 | 938 | vm_userfaultfd_ctx, anon_name)) { |
eef19944 JM |
939 | merge_next = true; |
940 | } | |
0503ea8f | 941 | |
29417d29 LS |
942 | /* Verify some invariant that must be enforced by the caller. */ |
943 | VM_WARN_ON(prev && addr <= prev->vm_start); | |
944 | VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end)); | |
945 | VM_WARN_ON(addr >= end); | |
946 | ||
0173db4f LS |
947 | if (!merge_prev && !merge_next) |
948 | return NULL; /* Not mergeable. */ | |
949 | ||
950 | res = vma = prev; | |
0503ea8f | 951 | remove = remove2 = adjust = NULL; |
0173db4f | 952 | |
eef19944 JM |
953 | /* Can we merge both the predecessor and the successor? */ |
954 | if (merge_prev && merge_next && | |
0503ea8f | 955 | is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { |
5ff783f1 | 956 | remove = next; /* case 1 */ |
0503ea8f | 957 | vma_end = next->vm_end; |
5ff783f1 | 958 | err = dup_anon_vma(prev, next); |
fcfccd91 LS |
959 | if (curr) { /* case 6 */ |
960 | remove = curr; | |
0503ea8f | 961 | remove2 = next; |
5ff783f1 | 962 | if (!next->anon_vma) |
fcfccd91 | 963 | err = dup_anon_vma(prev, curr); |
0503ea8f | 964 | } |
0173db4f | 965 | } else if (merge_prev) { /* case 2 */ |
fcfccd91 LS |
966 | if (curr) { |
967 | err = dup_anon_vma(prev, curr); | |
968 | if (end == curr->vm_end) { /* case 7 */ | |
969 | remove = curr; | |
0503ea8f | 970 | } else { /* case 5 */ |
fcfccd91 LS |
971 | adjust = curr; |
972 | adj_start = (end - curr->vm_start); | |
0503ea8f LH |
973 | } |
974 | } | |
0173db4f | 975 | } else { /* merge_next */ |
eef19944 | 976 | res = next; |
0503ea8f LH |
977 | if (prev && addr < prev->vm_end) { /* case 4 */ |
978 | vma_end = addr; | |
183b7a60 | 979 | adjust = next; |
1e76454f | 980 | adj_start = -(prev->vm_end - addr); |
183b7a60 | 981 | err = dup_anon_vma(next, prev); |
0503ea8f | 982 | } else { |
b0729ae0 LS |
983 | /* |
984 | * Note that cases 3 and 8 are the ONLY ones where prev | |
985 | * is permitted to be (but is not necessarily) NULL. | |
986 | */ | |
0503ea8f LH |
987 | vma = next; /* case 3 */ |
988 | vma_start = addr; | |
989 | vma_end = next->vm_end; | |
7e775787 | 990 | vma_pgoff = next->vm_pgoff - pglen; |
fcfccd91 LS |
991 | if (curr) { /* case 8 */ |
992 | vma_pgoff = curr->vm_pgoff; | |
993 | remove = curr; | |
994 | err = dup_anon_vma(next, curr); | |
0503ea8f LH |
995 | } |
996 | } | |
1da177e4 LT |
997 | } |
998 | ||
0173db4f | 999 | /* Error in anon_vma clone. */ |
eef19944 JM |
1000 | if (err) |
1001 | return NULL; | |
0503ea8f LH |
1002 | |
1003 | if (vma_iter_prealloc(vmi)) | |
1004 | return NULL; | |
1005 | ||
0503ea8f LH |
1006 | init_multi_vma_prep(&vp, vma, adjust, remove, remove2); |
1007 | VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && | |
1008 | vp.anon_vma != adjust->anon_vma); | |
1009 | ||
1010 | vma_prepare(&vp); | |
ccf1d78d | 1011 | vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); |
0503ea8f LH |
1012 | if (vma_start < vma->vm_start || vma_end > vma->vm_end) |
1013 | vma_expanded = true; | |
1014 | ||
1015 | vma->vm_start = vma_start; | |
1016 | vma->vm_end = vma_end; | |
1017 | vma->vm_pgoff = vma_pgoff; | |
1018 | ||
1019 | if (vma_expanded) | |
1020 | vma_iter_store(vmi, vma); | |
1021 | ||
1e76454f VB |
1022 | if (adj_start) { |
1023 | adjust->vm_start += adj_start; | |
1024 | adjust->vm_pgoff += adj_start >> PAGE_SHIFT; | |
1025 | if (adj_start < 0) { | |
0503ea8f LH |
1026 | WARN_ON(vma_expanded); |
1027 | vma_iter_store(vmi, next); | |
1028 | } | |
1029 | } | |
1030 | ||
1031 | vma_complete(&vp, vmi, mm); | |
1032 | vma_iter_free(vmi); | |
1033 | validate_mm(mm); | |
eef19944 | 1034 | khugepaged_enter_vma(res, vm_flags); |
1da177e4 | 1035 | |
9760ebff | 1036 | return res; |
f2ebfe43 LH |
1037 | } |
1038 | ||
d0e9fe17 | 1039 | /* |
b4f315b4 | 1040 | * Rough compatibility check to quickly see if it's even worth looking |
d0e9fe17 LT |
1041 | * at sharing an anon_vma. |
1042 | * | |
1043 | * They need to have the same vm_file, and the flags can only differ | |
1044 | * in things that mprotect may change. | |
1045 | * | |
1046 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1047 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1048 | * there is a vm_ops->close() function, because that indicates that the | |
1049 | * driver is doing some kind of reference counting. But that doesn't | |
1050 | * really matter for the anon_vma sharing case. | |
1051 | */ | |
1052 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1053 | { | |
1054 | return a->vm_end == b->vm_start && | |
1055 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1056 | a->vm_file == b->vm_file && | |
6cb4d9a2 | 1057 | !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && |
d0e9fe17 LT |
1058 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1063 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1064 | * the same as 'old', the other will be the new one that is trying | |
1065 | * to share the anon_vma. | |
1066 | * | |
5b449489 | 1067 | * NOTE! This runs with mmap_lock held for reading, so it is possible that |
d0e9fe17 LT |
1068 | * the anon_vma of 'old' is concurrently in the process of being set up |
1069 | * by another page fault trying to merge _that_. But that's ok: if it | |
1070 | * is being set up, that automatically means that it will be a singleton | |
1071 | * acceptable for merging, so we can do all of this optimistically. But | |
4db0c3c2 | 1072 | * we do that READ_ONCE() to make sure that we never re-load the pointer. |
d0e9fe17 LT |
1073 | * |
1074 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1075 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1076 | * is to return an anon_vma that is "complex" due to having gone through | |
1077 | * a fork). | |
1078 | * | |
1079 | * We also make sure that the two vma's are compatible (adjacent, | |
1080 | * and with the same memory policies). That's all stable, even with just | |
5b449489 | 1081 | * a read lock on the mmap_lock. |
d0e9fe17 LT |
1082 | */ |
1083 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1084 | { | |
1085 | if (anon_vma_compatible(a, b)) { | |
4db0c3c2 | 1086 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); |
d0e9fe17 LT |
1087 | |
1088 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1089 | return anon_vma; | |
1090 | } | |
1091 | return NULL; | |
1092 | } | |
1093 | ||
1da177e4 LT |
1094 | /* |
1095 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1096 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1097 | * to allocate a new anon_vma. It checks because a repetitive | |
1098 | * sequence of mprotects and faults may otherwise lead to distinct | |
1099 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1100 | * mprotect. | |
1101 | */ | |
1102 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1103 | { | |
763ecb03 | 1104 | MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); |
a67c8caa | 1105 | struct anon_vma *anon_vma = NULL; |
763ecb03 | 1106 | struct vm_area_struct *prev, *next; |
a67c8caa ML |
1107 | |
1108 | /* Try next first. */ | |
763ecb03 LH |
1109 | next = mas_walk(&mas); |
1110 | if (next) { | |
1111 | anon_vma = reusable_anon_vma(next, vma, next); | |
a67c8caa ML |
1112 | if (anon_vma) |
1113 | return anon_vma; | |
1114 | } | |
1115 | ||
763ecb03 LH |
1116 | prev = mas_prev(&mas, 0); |
1117 | VM_BUG_ON_VMA(prev != vma, vma); | |
1118 | prev = mas_prev(&mas, 0); | |
a67c8caa | 1119 | /* Try prev next. */ |
763ecb03 LH |
1120 | if (prev) |
1121 | anon_vma = reusable_anon_vma(prev, prev, vma); | |
a67c8caa | 1122 | |
1da177e4 | 1123 | /* |
a67c8caa ML |
1124 | * We might reach here with anon_vma == NULL if we can't find |
1125 | * any reusable anon_vma. | |
1da177e4 LT |
1126 | * There's no absolute need to look only at touching neighbours: |
1127 | * we could search further afield for "compatible" anon_vmas. | |
1128 | * But it would probably just be a waste of time searching, | |
1129 | * or lead to too many vmas hanging off the same anon_vma. | |
1130 | * We're trying to allow mprotect remerging later on, | |
1131 | * not trying to minimize memory used for anon_vmas. | |
1132 | */ | |
a67c8caa | 1133 | return anon_vma; |
1da177e4 LT |
1134 | } |
1135 | ||
40401530 AV |
1136 | /* |
1137 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1138 | * low as possible but still greater than mmap_min_addr | |
1139 | */ | |
1140 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1141 | { | |
1142 | hint &= PAGE_MASK; | |
1143 | if (((void *)hint != NULL) && | |
1144 | (hint < mmap_min_addr)) | |
1145 | return PAGE_ALIGN(mmap_min_addr); | |
1146 | return hint; | |
1147 | } | |
1148 | ||
3c54a298 LS |
1149 | bool mlock_future_check(struct mm_struct *mm, unsigned long flags, |
1150 | unsigned long bytes) | |
363ee17f | 1151 | { |
3c54a298 LS |
1152 | unsigned long locked_pages, limit_pages; |
1153 | ||
1154 | if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) | |
1155 | return true; | |
1156 | ||
1157 | locked_pages = bytes >> PAGE_SHIFT; | |
1158 | locked_pages += mm->locked_vm; | |
1159 | ||
1160 | limit_pages = rlimit(RLIMIT_MEMLOCK); | |
1161 | limit_pages >>= PAGE_SHIFT; | |
1162 | ||
1163 | return locked_pages <= limit_pages; | |
363ee17f DB |
1164 | } |
1165 | ||
be83bbf8 LT |
1166 | static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) |
1167 | { | |
1168 | if (S_ISREG(inode->i_mode)) | |
423913ad | 1169 | return MAX_LFS_FILESIZE; |
be83bbf8 LT |
1170 | |
1171 | if (S_ISBLK(inode->i_mode)) | |
1172 | return MAX_LFS_FILESIZE; | |
1173 | ||
76f34950 IK |
1174 | if (S_ISSOCK(inode->i_mode)) |
1175 | return MAX_LFS_FILESIZE; | |
1176 | ||
be83bbf8 LT |
1177 | /* Special "we do even unsigned file positions" case */ |
1178 | if (file->f_mode & FMODE_UNSIGNED_OFFSET) | |
1179 | return 0; | |
1180 | ||
1181 | /* Yes, random drivers might want more. But I'm tired of buggy drivers */ | |
1182 | return ULONG_MAX; | |
1183 | } | |
1184 | ||
1185 | static inline bool file_mmap_ok(struct file *file, struct inode *inode, | |
1186 | unsigned long pgoff, unsigned long len) | |
1187 | { | |
1188 | u64 maxsize = file_mmap_size_max(file, inode); | |
1189 | ||
1190 | if (maxsize && len > maxsize) | |
1191 | return false; | |
1192 | maxsize -= len; | |
1193 | if (pgoff > maxsize >> PAGE_SHIFT) | |
1194 | return false; | |
1195 | return true; | |
1196 | } | |
1197 | ||
1da177e4 | 1198 | /* |
3e4e28c5 | 1199 | * The caller must write-lock current->mm->mmap_lock. |
1da177e4 | 1200 | */ |
1fcfd8db | 1201 | unsigned long do_mmap(struct file *file, unsigned long addr, |
1da177e4 | 1202 | unsigned long len, unsigned long prot, |
45e55300 PC |
1203 | unsigned long flags, unsigned long pgoff, |
1204 | unsigned long *populate, struct list_head *uf) | |
1da177e4 | 1205 | { |
cc71aba3 | 1206 | struct mm_struct *mm = current->mm; |
45e55300 | 1207 | vm_flags_t vm_flags; |
62b5f7d0 | 1208 | int pkey = 0; |
1da177e4 | 1209 | |
524e00b3 | 1210 | validate_mm(mm); |
41badc15 | 1211 | *populate = 0; |
bebeb3d6 | 1212 | |
e37609bb PK |
1213 | if (!len) |
1214 | return -EINVAL; | |
1215 | ||
1da177e4 LT |
1216 | /* |
1217 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1218 | * | |
1219 | * (the exception is when the underlying filesystem is noexec | |
1220 | * mounted, in which case we dont add PROT_EXEC.) | |
1221 | */ | |
1222 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
90f8572b | 1223 | if (!(file && path_noexec(&file->f_path))) |
1da177e4 LT |
1224 | prot |= PROT_EXEC; |
1225 | ||
a4ff8e86 MH |
1226 | /* force arch specific MAP_FIXED handling in get_unmapped_area */ |
1227 | if (flags & MAP_FIXED_NOREPLACE) | |
1228 | flags |= MAP_FIXED; | |
1229 | ||
7cd94146 EP |
1230 | if (!(flags & MAP_FIXED)) |
1231 | addr = round_hint_to_min(addr); | |
1232 | ||
1da177e4 LT |
1233 | /* Careful about overflows.. */ |
1234 | len = PAGE_ALIGN(len); | |
9206de95 | 1235 | if (!len) |
1da177e4 LT |
1236 | return -ENOMEM; |
1237 | ||
1238 | /* offset overflow? */ | |
1239 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
cc71aba3 | 1240 | return -EOVERFLOW; |
1da177e4 LT |
1241 | |
1242 | /* Too many mappings? */ | |
1243 | if (mm->map_count > sysctl_max_map_count) | |
1244 | return -ENOMEM; | |
1245 | ||
1246 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1247 | * that it represents a valid section of the address space. | |
1248 | */ | |
1249 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
ff68dac6 | 1250 | if (IS_ERR_VALUE(addr)) |
1da177e4 LT |
1251 | return addr; |
1252 | ||
a4ff8e86 | 1253 | if (flags & MAP_FIXED_NOREPLACE) { |
35e43c5f | 1254 | if (find_vma_intersection(mm, addr, addr + len)) |
a4ff8e86 MH |
1255 | return -EEXIST; |
1256 | } | |
1257 | ||
62b5f7d0 DH |
1258 | if (prot == PROT_EXEC) { |
1259 | pkey = execute_only_pkey(mm); | |
1260 | if (pkey < 0) | |
1261 | pkey = 0; | |
1262 | } | |
1263 | ||
1da177e4 LT |
1264 | /* Do simple checking here so the lower-level routines won't have |
1265 | * to. we assume access permissions have been handled by the open | |
1266 | * of the memory object, so we don't do any here. | |
1267 | */ | |
45e55300 | 1268 | vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | |
1da177e4 LT |
1269 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1270 | ||
cdf7b341 | 1271 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1272 | if (!can_do_mlock()) |
1273 | return -EPERM; | |
ba470de4 | 1274 | |
3c54a298 | 1275 | if (!mlock_future_check(mm, vm_flags, len)) |
363ee17f | 1276 | return -EAGAIN; |
1da177e4 | 1277 | |
1da177e4 | 1278 | if (file) { |
077bf22b | 1279 | struct inode *inode = file_inode(file); |
1c972597 DW |
1280 | unsigned long flags_mask; |
1281 | ||
be83bbf8 LT |
1282 | if (!file_mmap_ok(file, inode, pgoff, len)) |
1283 | return -EOVERFLOW; | |
1284 | ||
1c972597 | 1285 | flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; |
077bf22b | 1286 | |
1da177e4 LT |
1287 | switch (flags & MAP_TYPE) { |
1288 | case MAP_SHARED: | |
1c972597 DW |
1289 | /* |
1290 | * Force use of MAP_SHARED_VALIDATE with non-legacy | |
1291 | * flags. E.g. MAP_SYNC is dangerous to use with | |
1292 | * MAP_SHARED as you don't know which consistency model | |
1293 | * you will get. We silently ignore unsupported flags | |
1294 | * with MAP_SHARED to preserve backward compatibility. | |
1295 | */ | |
1296 | flags &= LEGACY_MAP_MASK; | |
e4a9bc58 | 1297 | fallthrough; |
1c972597 DW |
1298 | case MAP_SHARED_VALIDATE: |
1299 | if (flags & ~flags_mask) | |
1300 | return -EOPNOTSUPP; | |
dc617f29 DW |
1301 | if (prot & PROT_WRITE) { |
1302 | if (!(file->f_mode & FMODE_WRITE)) | |
1303 | return -EACCES; | |
1304 | if (IS_SWAPFILE(file->f_mapping->host)) | |
1305 | return -ETXTBSY; | |
1306 | } | |
1da177e4 LT |
1307 | |
1308 | /* | |
1309 | * Make sure we don't allow writing to an append-only | |
1310 | * file.. | |
1311 | */ | |
1312 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1313 | return -EACCES; | |
1314 | ||
1da177e4 LT |
1315 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1316 | if (!(file->f_mode & FMODE_WRITE)) | |
1317 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
e4a9bc58 | 1318 | fallthrough; |
1da177e4 LT |
1319 | case MAP_PRIVATE: |
1320 | if (!(file->f_mode & FMODE_READ)) | |
1321 | return -EACCES; | |
90f8572b | 1322 | if (path_noexec(&file->f_path)) { |
80c5606c LT |
1323 | if (vm_flags & VM_EXEC) |
1324 | return -EPERM; | |
1325 | vm_flags &= ~VM_MAYEXEC; | |
1326 | } | |
80c5606c | 1327 | |
72c2d531 | 1328 | if (!file->f_op->mmap) |
80c5606c | 1329 | return -ENODEV; |
b2c56e4f ON |
1330 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1331 | return -EINVAL; | |
1da177e4 LT |
1332 | break; |
1333 | ||
1334 | default: | |
1335 | return -EINVAL; | |
1336 | } | |
1337 | } else { | |
1338 | switch (flags & MAP_TYPE) { | |
1339 | case MAP_SHARED: | |
b2c56e4f ON |
1340 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1341 | return -EINVAL; | |
ce363942 TH |
1342 | /* |
1343 | * Ignore pgoff. | |
1344 | */ | |
1345 | pgoff = 0; | |
1da177e4 LT |
1346 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1347 | break; | |
1348 | case MAP_PRIVATE: | |
1349 | /* | |
1350 | * Set pgoff according to addr for anon_vma. | |
1351 | */ | |
1352 | pgoff = addr >> PAGE_SHIFT; | |
1353 | break; | |
1354 | default: | |
1355 | return -EINVAL; | |
1356 | } | |
1357 | } | |
1358 | ||
c22c0d63 ML |
1359 | /* |
1360 | * Set 'VM_NORESERVE' if we should not account for the | |
1361 | * memory use of this mapping. | |
1362 | */ | |
1363 | if (flags & MAP_NORESERVE) { | |
1364 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1365 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1366 | vm_flags |= VM_NORESERVE; | |
1367 | ||
1368 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1369 | if (file && is_file_hugepages(file)) | |
1370 | vm_flags |= VM_NORESERVE; | |
1371 | } | |
1372 | ||
897ab3e0 | 1373 | addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); |
09a9f1d2 ML |
1374 | if (!IS_ERR_VALUE(addr) && |
1375 | ((vm_flags & VM_LOCKED) || | |
1376 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1377 | *populate = len; |
bebeb3d6 | 1378 | return addr; |
0165ab44 | 1379 | } |
6be5ceb0 | 1380 | |
a90f590a DB |
1381 | unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, |
1382 | unsigned long prot, unsigned long flags, | |
1383 | unsigned long fd, unsigned long pgoff) | |
66f0dc48 HD |
1384 | { |
1385 | struct file *file = NULL; | |
1e3ee14b | 1386 | unsigned long retval; |
66f0dc48 HD |
1387 | |
1388 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1389 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1390 | file = fget(fd); |
1391 | if (!file) | |
1e3ee14b | 1392 | return -EBADF; |
7bba8f0e | 1393 | if (is_file_hugepages(file)) { |
af73e4d9 | 1394 | len = ALIGN(len, huge_page_size(hstate_file(file))); |
7bba8f0e ZL |
1395 | } else if (unlikely(flags & MAP_HUGETLB)) { |
1396 | retval = -EINVAL; | |
493af578 | 1397 | goto out_fput; |
7bba8f0e | 1398 | } |
66f0dc48 | 1399 | } else if (flags & MAP_HUGETLB) { |
c103a4dc | 1400 | struct hstate *hs; |
af73e4d9 | 1401 | |
20ac2893 | 1402 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); |
091d0d55 LZ |
1403 | if (!hs) |
1404 | return -EINVAL; | |
1405 | ||
1406 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1407 | /* |
1408 | * VM_NORESERVE is used because the reservations will be | |
1409 | * taken when vm_ops->mmap() is called | |
66f0dc48 | 1410 | */ |
af73e4d9 | 1411 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f | 1412 | VM_NORESERVE, |
83c1fd76 | 1413 | HUGETLB_ANONHUGE_INODE, |
42d7395f | 1414 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); |
66f0dc48 HD |
1415 | if (IS_ERR(file)) |
1416 | return PTR_ERR(file); | |
1417 | } | |
1418 | ||
9fbeb5ab | 1419 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
493af578 | 1420 | out_fput: |
66f0dc48 HD |
1421 | if (file) |
1422 | fput(file); | |
66f0dc48 HD |
1423 | return retval; |
1424 | } | |
1425 | ||
a90f590a DB |
1426 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1427 | unsigned long, prot, unsigned long, flags, | |
1428 | unsigned long, fd, unsigned long, pgoff) | |
1429 | { | |
1430 | return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); | |
1431 | } | |
1432 | ||
a4679373 CH |
1433 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1434 | struct mmap_arg_struct { | |
1435 | unsigned long addr; | |
1436 | unsigned long len; | |
1437 | unsigned long prot; | |
1438 | unsigned long flags; | |
1439 | unsigned long fd; | |
1440 | unsigned long offset; | |
1441 | }; | |
1442 | ||
1443 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1444 | { | |
1445 | struct mmap_arg_struct a; | |
1446 | ||
1447 | if (copy_from_user(&a, arg, sizeof(a))) | |
1448 | return -EFAULT; | |
de1741a1 | 1449 | if (offset_in_page(a.offset)) |
a4679373 CH |
1450 | return -EINVAL; |
1451 | ||
a90f590a DB |
1452 | return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, |
1453 | a.offset >> PAGE_SHIFT); | |
a4679373 CH |
1454 | } |
1455 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1456 | ||
4e950f6f | 1457 | /* |
8bb4e7a2 | 1458 | * Some shared mappings will want the pages marked read-only |
4e950f6f AD |
1459 | * to track write events. If so, we'll downgrade vm_page_prot |
1460 | * to the private version (using protection_map[] without the | |
1461 | * VM_SHARED bit). | |
1462 | */ | |
6d2329f8 | 1463 | int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) |
4e950f6f | 1464 | { |
ca16d140 | 1465 | vm_flags_t vm_flags = vma->vm_flags; |
8a04446a | 1466 | const struct vm_operations_struct *vm_ops = vma->vm_ops; |
4e950f6f AD |
1467 | |
1468 | /* If it was private or non-writable, the write bit is already clear */ | |
1469 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1470 | return 0; | |
1471 | ||
1472 | /* The backer wishes to know when pages are first written to? */ | |
8a04446a | 1473 | if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) |
4e950f6f AD |
1474 | return 1; |
1475 | ||
64e45507 PF |
1476 | /* The open routine did something to the protections that pgprot_modify |
1477 | * won't preserve? */ | |
6d2329f8 AA |
1478 | if (pgprot_val(vm_page_prot) != |
1479 | pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) | |
4e950f6f AD |
1480 | return 0; |
1481 | ||
f96f7a40 DH |
1482 | /* |
1483 | * Do we need to track softdirty? hugetlb does not support softdirty | |
1484 | * tracking yet. | |
1485 | */ | |
1486 | if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) | |
64e45507 PF |
1487 | return 1; |
1488 | ||
51d3d5eb DH |
1489 | /* Do we need write faults for uffd-wp tracking? */ |
1490 | if (userfaultfd_wp(vma)) | |
1491 | return 1; | |
1492 | ||
4e950f6f | 1493 | /* Specialty mapping? */ |
4b6e1e37 | 1494 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1495 | return 0; |
1496 | ||
1497 | /* Can the mapping track the dirty pages? */ | |
1498 | return vma->vm_file && vma->vm_file->f_mapping && | |
f56753ac | 1499 | mapping_can_writeback(vma->vm_file->f_mapping); |
4e950f6f AD |
1500 | } |
1501 | ||
fc8744ad LT |
1502 | /* |
1503 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1504 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1505 | */ |
ca16d140 | 1506 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1507 | { |
5a6fe125 MG |
1508 | /* |
1509 | * hugetlb has its own accounting separate from the core VM | |
1510 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1511 | */ | |
1512 | if (file && is_file_hugepages(file)) | |
1513 | return 0; | |
1514 | ||
fc8744ad LT |
1515 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1516 | } | |
1517 | ||
3499a131 LH |
1518 | /** |
1519 | * unmapped_area() - Find an area between the low_limit and the high_limit with | |
1520 | * the correct alignment and offset, all from @info. Note: current->mm is used | |
1521 | * for the search. | |
1522 | * | |
82b24936 VY |
1523 | * @info: The unmapped area information including the range [low_limit - |
1524 | * high_limit), the alignment offset and mask. | |
3499a131 LH |
1525 | * |
1526 | * Return: A memory address or -ENOMEM. | |
1527 | */ | |
baceaf1c | 1528 | static unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
db4fbfb9 | 1529 | { |
6b008640 LT |
1530 | unsigned long length, gap; |
1531 | unsigned long low_limit, high_limit; | |
58c5d0d6 | 1532 | struct vm_area_struct *tmp; |
db4fbfb9 | 1533 | |
3499a131 | 1534 | MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); |
db4fbfb9 ML |
1535 | |
1536 | /* Adjust search length to account for worst case alignment overhead */ | |
1537 | length = info->length + info->align_mask; | |
1538 | if (length < info->length) | |
1539 | return -ENOMEM; | |
1540 | ||
58c5d0d6 | 1541 | low_limit = info->low_limit; |
6b008640 LT |
1542 | if (low_limit < mmap_min_addr) |
1543 | low_limit = mmap_min_addr; | |
1544 | high_limit = info->high_limit; | |
58c5d0d6 | 1545 | retry: |
6b008640 | 1546 | if (mas_empty_area(&mas, low_limit, high_limit - 1, length)) |
db4fbfb9 ML |
1547 | return -ENOMEM; |
1548 | ||
3499a131 LH |
1549 | gap = mas.index; |
1550 | gap += (info->align_offset - gap) & info->align_mask; | |
58c5d0d6 LH |
1551 | tmp = mas_next(&mas, ULONG_MAX); |
1552 | if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ | |
1553 | if (vm_start_gap(tmp) < gap + length - 1) { | |
1554 | low_limit = tmp->vm_end; | |
1555 | mas_reset(&mas); | |
1556 | goto retry; | |
1557 | } | |
1558 | } else { | |
1559 | tmp = mas_prev(&mas, 0); | |
1560 | if (tmp && vm_end_gap(tmp) > gap) { | |
1561 | low_limit = vm_end_gap(tmp); | |
1562 | mas_reset(&mas); | |
1563 | goto retry; | |
1564 | } | |
1565 | } | |
1566 | ||
3499a131 | 1567 | return gap; |
db4fbfb9 ML |
1568 | } |
1569 | ||
3499a131 LH |
1570 | /** |
1571 | * unmapped_area_topdown() - Find an area between the low_limit and the | |
82b24936 | 1572 | * high_limit with the correct alignment and offset at the highest available |
3499a131 LH |
1573 | * address, all from @info. Note: current->mm is used for the search. |
1574 | * | |
82b24936 VY |
1575 | * @info: The unmapped area information including the range [low_limit - |
1576 | * high_limit), the alignment offset and mask. | |
3499a131 LH |
1577 | * |
1578 | * Return: A memory address or -ENOMEM. | |
1579 | */ | |
baceaf1c | 1580 | static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) |
db4fbfb9 | 1581 | { |
6b008640 LT |
1582 | unsigned long length, gap, gap_end; |
1583 | unsigned long low_limit, high_limit; | |
58c5d0d6 | 1584 | struct vm_area_struct *tmp; |
db4fbfb9 | 1585 | |
3499a131 | 1586 | MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); |
db4fbfb9 ML |
1587 | /* Adjust search length to account for worst case alignment overhead */ |
1588 | length = info->length + info->align_mask; | |
1589 | if (length < info->length) | |
1590 | return -ENOMEM; | |
1591 | ||
6b008640 LT |
1592 | low_limit = info->low_limit; |
1593 | if (low_limit < mmap_min_addr) | |
1594 | low_limit = mmap_min_addr; | |
58c5d0d6 LH |
1595 | high_limit = info->high_limit; |
1596 | retry: | |
6b008640 | 1597 | if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length)) |
db4fbfb9 | 1598 | return -ENOMEM; |
db4fbfb9 | 1599 | |
3499a131 LH |
1600 | gap = mas.last + 1 - info->length; |
1601 | gap -= (gap - info->align_offset) & info->align_mask; | |
58c5d0d6 LH |
1602 | gap_end = mas.last; |
1603 | tmp = mas_next(&mas, ULONG_MAX); | |
1604 | if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ | |
1605 | if (vm_start_gap(tmp) <= gap_end) { | |
1606 | high_limit = vm_start_gap(tmp); | |
1607 | mas_reset(&mas); | |
1608 | goto retry; | |
1609 | } | |
1610 | } else { | |
1611 | tmp = mas_prev(&mas, 0); | |
1612 | if (tmp && vm_end_gap(tmp) > gap) { | |
1613 | high_limit = tmp->vm_start; | |
1614 | mas_reset(&mas); | |
1615 | goto retry; | |
1616 | } | |
1617 | } | |
1618 | ||
3499a131 | 1619 | return gap; |
db4fbfb9 ML |
1620 | } |
1621 | ||
baceaf1c JK |
1622 | /* |
1623 | * Search for an unmapped address range. | |
1624 | * | |
1625 | * We are looking for a range that: | |
1626 | * - does not intersect with any VMA; | |
1627 | * - is contained within the [low_limit, high_limit) interval; | |
1628 | * - is at least the desired size. | |
1629 | * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) | |
1630 | */ | |
1631 | unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) | |
1632 | { | |
df529cab JK |
1633 | unsigned long addr; |
1634 | ||
baceaf1c | 1635 | if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) |
df529cab | 1636 | addr = unmapped_area_topdown(info); |
baceaf1c | 1637 | else |
df529cab JK |
1638 | addr = unmapped_area(info); |
1639 | ||
1640 | trace_vm_unmapped_area(addr, info); | |
1641 | return addr; | |
baceaf1c | 1642 | } |
f6795053 | 1643 | |
1da177e4 LT |
1644 | /* Get an address range which is currently unmapped. |
1645 | * For shmat() with addr=0. | |
1646 | * | |
1647 | * Ugly calling convention alert: | |
1648 | * Return value with the low bits set means error value, | |
1649 | * ie | |
1650 | * if (ret & ~PAGE_MASK) | |
1651 | * error = ret; | |
1652 | * | |
1653 | * This function "knows" that -ENOMEM has the bits set. | |
1654 | */ | |
1da177e4 | 1655 | unsigned long |
4b439e25 CL |
1656 | generic_get_unmapped_area(struct file *filp, unsigned long addr, |
1657 | unsigned long len, unsigned long pgoff, | |
1658 | unsigned long flags) | |
1da177e4 LT |
1659 | { |
1660 | struct mm_struct *mm = current->mm; | |
1be7107f | 1661 | struct vm_area_struct *vma, *prev; |
db4fbfb9 | 1662 | struct vm_unmapped_area_info info; |
2cb4de08 | 1663 | const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); |
1da177e4 | 1664 | |
f6795053 | 1665 | if (len > mmap_end - mmap_min_addr) |
1da177e4 LT |
1666 | return -ENOMEM; |
1667 | ||
06abdfb4 BH |
1668 | if (flags & MAP_FIXED) |
1669 | return addr; | |
1670 | ||
1da177e4 LT |
1671 | if (addr) { |
1672 | addr = PAGE_ALIGN(addr); | |
1be7107f | 1673 | vma = find_vma_prev(mm, addr, &prev); |
f6795053 | 1674 | if (mmap_end - len >= addr && addr >= mmap_min_addr && |
1be7107f HD |
1675 | (!vma || addr + len <= vm_start_gap(vma)) && |
1676 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
1677 | return addr; |
1678 | } | |
1da177e4 | 1679 | |
db4fbfb9 ML |
1680 | info.flags = 0; |
1681 | info.length = len; | |
4e99b021 | 1682 | info.low_limit = mm->mmap_base; |
f6795053 | 1683 | info.high_limit = mmap_end; |
db4fbfb9 | 1684 | info.align_mask = 0; |
09ef5283 | 1685 | info.align_offset = 0; |
db4fbfb9 | 1686 | return vm_unmapped_area(&info); |
1da177e4 | 1687 | } |
4b439e25 CL |
1688 | |
1689 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1690 | unsigned long | |
1691 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1692 | unsigned long len, unsigned long pgoff, | |
1693 | unsigned long flags) | |
1694 | { | |
1695 | return generic_get_unmapped_area(filp, addr, len, pgoff, flags); | |
1696 | } | |
cc71aba3 | 1697 | #endif |
1da177e4 | 1698 | |
1da177e4 LT |
1699 | /* |
1700 | * This mmap-allocator allocates new areas top-down from below the | |
1701 | * stack's low limit (the base): | |
1702 | */ | |
1da177e4 | 1703 | unsigned long |
4b439e25 CL |
1704 | generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, |
1705 | unsigned long len, unsigned long pgoff, | |
1706 | unsigned long flags) | |
1da177e4 | 1707 | { |
1be7107f | 1708 | struct vm_area_struct *vma, *prev; |
1da177e4 | 1709 | struct mm_struct *mm = current->mm; |
db4fbfb9 | 1710 | struct vm_unmapped_area_info info; |
2cb4de08 | 1711 | const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); |
1da177e4 LT |
1712 | |
1713 | /* requested length too big for entire address space */ | |
f6795053 | 1714 | if (len > mmap_end - mmap_min_addr) |
1da177e4 LT |
1715 | return -ENOMEM; |
1716 | ||
06abdfb4 BH |
1717 | if (flags & MAP_FIXED) |
1718 | return addr; | |
1719 | ||
1da177e4 LT |
1720 | /* requesting a specific address */ |
1721 | if (addr) { | |
1722 | addr = PAGE_ALIGN(addr); | |
1be7107f | 1723 | vma = find_vma_prev(mm, addr, &prev); |
f6795053 | 1724 | if (mmap_end - len >= addr && addr >= mmap_min_addr && |
1be7107f HD |
1725 | (!vma || addr + len <= vm_start_gap(vma)) && |
1726 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
1727 | return addr; |
1728 | } | |
1729 | ||
db4fbfb9 ML |
1730 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
1731 | info.length = len; | |
6b008640 | 1732 | info.low_limit = PAGE_SIZE; |
f6795053 | 1733 | info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); |
db4fbfb9 | 1734 | info.align_mask = 0; |
09ef5283 | 1735 | info.align_offset = 0; |
db4fbfb9 | 1736 | addr = vm_unmapped_area(&info); |
b716ad95 | 1737 | |
1da177e4 LT |
1738 | /* |
1739 | * A failed mmap() very likely causes application failure, | |
1740 | * so fall back to the bottom-up function here. This scenario | |
1741 | * can happen with large stack limits and large mmap() | |
1742 | * allocations. | |
1743 | */ | |
de1741a1 | 1744 | if (offset_in_page(addr)) { |
db4fbfb9 ML |
1745 | VM_BUG_ON(addr != -ENOMEM); |
1746 | info.flags = 0; | |
1747 | info.low_limit = TASK_UNMAPPED_BASE; | |
f6795053 | 1748 | info.high_limit = mmap_end; |
db4fbfb9 ML |
1749 | addr = vm_unmapped_area(&info); |
1750 | } | |
1da177e4 LT |
1751 | |
1752 | return addr; | |
1753 | } | |
4b439e25 CL |
1754 | |
1755 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1756 | unsigned long | |
1757 | arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, | |
1758 | unsigned long len, unsigned long pgoff, | |
1759 | unsigned long flags) | |
1760 | { | |
1761 | return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); | |
1762 | } | |
1da177e4 LT |
1763 | #endif |
1764 | ||
1da177e4 LT |
1765 | unsigned long |
1766 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1767 | unsigned long pgoff, unsigned long flags) | |
1768 | { | |
06abdfb4 BH |
1769 | unsigned long (*get_area)(struct file *, unsigned long, |
1770 | unsigned long, unsigned long, unsigned long); | |
1771 | ||
9206de95 AV |
1772 | unsigned long error = arch_mmap_check(addr, len, flags); |
1773 | if (error) | |
1774 | return error; | |
1775 | ||
1776 | /* Careful about overflows.. */ | |
1777 | if (len > TASK_SIZE) | |
1778 | return -ENOMEM; | |
1779 | ||
06abdfb4 | 1780 | get_area = current->mm->get_unmapped_area; |
c01d5b30 HD |
1781 | if (file) { |
1782 | if (file->f_op->get_unmapped_area) | |
1783 | get_area = file->f_op->get_unmapped_area; | |
1784 | } else if (flags & MAP_SHARED) { | |
1785 | /* | |
1786 | * mmap_region() will call shmem_zero_setup() to create a file, | |
1787 | * so use shmem's get_unmapped_area in case it can be huge. | |
45e55300 | 1788 | * do_mmap() will clear pgoff, so match alignment. |
c01d5b30 HD |
1789 | */ |
1790 | pgoff = 0; | |
1791 | get_area = shmem_get_unmapped_area; | |
1792 | } | |
1793 | ||
06abdfb4 BH |
1794 | addr = get_area(file, addr, len, pgoff, flags); |
1795 | if (IS_ERR_VALUE(addr)) | |
1796 | return addr; | |
1da177e4 | 1797 | |
07ab67c8 LT |
1798 | if (addr > TASK_SIZE - len) |
1799 | return -ENOMEM; | |
de1741a1 | 1800 | if (offset_in_page(addr)) |
07ab67c8 | 1801 | return -EINVAL; |
06abdfb4 | 1802 | |
9ac4ed4b AV |
1803 | error = security_mmap_addr(addr); |
1804 | return error ? error : addr; | |
1da177e4 LT |
1805 | } |
1806 | ||
1807 | EXPORT_SYMBOL(get_unmapped_area); | |
1808 | ||
abdba2dd LH |
1809 | /** |
1810 | * find_vma_intersection() - Look up the first VMA which intersects the interval | |
1811 | * @mm: The process address space. | |
1812 | * @start_addr: The inclusive start user address. | |
1813 | * @end_addr: The exclusive end user address. | |
1814 | * | |
1815 | * Returns: The first VMA within the provided range, %NULL otherwise. Assumes | |
1816 | * start_addr < end_addr. | |
1817 | */ | |
1818 | struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, | |
1819 | unsigned long start_addr, | |
1820 | unsigned long end_addr) | |
1821 | { | |
abdba2dd LH |
1822 | unsigned long index = start_addr; |
1823 | ||
1824 | mmap_assert_locked(mm); | |
7964cf8c | 1825 | return mt_find(&mm->mm_mt, &index, end_addr - 1); |
abdba2dd LH |
1826 | } |
1827 | EXPORT_SYMBOL(find_vma_intersection); | |
1828 | ||
be8432e7 LH |
1829 | /** |
1830 | * find_vma() - Find the VMA for a given address, or the next VMA. | |
1831 | * @mm: The mm_struct to check | |
1832 | * @addr: The address | |
1833 | * | |
1834 | * Returns: The VMA associated with addr, or the next VMA. | |
1835 | * May return %NULL in the case of no VMA at addr or above. | |
1836 | */ | |
48aae425 | 1837 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 1838 | { |
be8432e7 | 1839 | unsigned long index = addr; |
1da177e4 | 1840 | |
5b78ed24 | 1841 | mmap_assert_locked(mm); |
7964cf8c | 1842 | return mt_find(&mm->mm_mt, &index, ULONG_MAX); |
1da177e4 | 1843 | } |
1da177e4 LT |
1844 | EXPORT_SYMBOL(find_vma); |
1845 | ||
7fdbd37d LH |
1846 | /** |
1847 | * find_vma_prev() - Find the VMA for a given address, or the next vma and | |
1848 | * set %pprev to the previous VMA, if any. | |
1849 | * @mm: The mm_struct to check | |
1850 | * @addr: The address | |
1851 | * @pprev: The pointer to set to the previous VMA | |
1852 | * | |
1853 | * Note that RCU lock is missing here since the external mmap_lock() is used | |
1854 | * instead. | |
1855 | * | |
1856 | * Returns: The VMA associated with @addr, or the next vma. | |
1857 | * May return %NULL in the case of no vma at addr or above. | |
6bd4837d | 1858 | */ |
1da177e4 LT |
1859 | struct vm_area_struct * |
1860 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
1861 | struct vm_area_struct **pprev) | |
1862 | { | |
6bd4837d | 1863 | struct vm_area_struct *vma; |
7fdbd37d | 1864 | MA_STATE(mas, &mm->mm_mt, addr, addr); |
1da177e4 | 1865 | |
7fdbd37d LH |
1866 | vma = mas_walk(&mas); |
1867 | *pprev = mas_prev(&mas, 0); | |
1868 | if (!vma) | |
1869 | vma = mas_next(&mas, ULONG_MAX); | |
6bd4837d | 1870 | return vma; |
1da177e4 LT |
1871 | } |
1872 | ||
1873 | /* | |
1874 | * Verify that the stack growth is acceptable and | |
1875 | * update accounting. This is shared with both the | |
1876 | * grow-up and grow-down cases. | |
1877 | */ | |
1be7107f HD |
1878 | static int acct_stack_growth(struct vm_area_struct *vma, |
1879 | unsigned long size, unsigned long grow) | |
1da177e4 LT |
1880 | { |
1881 | struct mm_struct *mm = vma->vm_mm; | |
1be7107f | 1882 | unsigned long new_start; |
1da177e4 LT |
1883 | |
1884 | /* address space limit tests */ | |
84638335 | 1885 | if (!may_expand_vm(mm, vma->vm_flags, grow)) |
1da177e4 LT |
1886 | return -ENOMEM; |
1887 | ||
1888 | /* Stack limit test */ | |
24c79d8e | 1889 | if (size > rlimit(RLIMIT_STACK)) |
1da177e4 LT |
1890 | return -ENOMEM; |
1891 | ||
1892 | /* mlock limit tests */ | |
3c54a298 | 1893 | if (!mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) |
c5d8a364 | 1894 | return -ENOMEM; |
1da177e4 | 1895 | |
0d59a01b AL |
1896 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
1897 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
1898 | vma->vm_end - size; | |
1899 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
1900 | return -EFAULT; | |
1901 | ||
1da177e4 LT |
1902 | /* |
1903 | * Overcommit.. This must be the final test, as it will | |
1904 | * update security statistics. | |
1905 | */ | |
05fa199d | 1906 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
1907 | return -ENOMEM; |
1908 | ||
1da177e4 LT |
1909 | return 0; |
1910 | } | |
1911 | ||
46dea3d0 | 1912 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 1913 | /* |
46dea3d0 HD |
1914 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
1915 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 1916 | */ |
46dea3d0 | 1917 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 1918 | { |
09357814 | 1919 | struct mm_struct *mm = vma->vm_mm; |
1be7107f HD |
1920 | struct vm_area_struct *next; |
1921 | unsigned long gap_addr; | |
12352d3c | 1922 | int error = 0; |
d4af56c5 | 1923 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
1da177e4 LT |
1924 | |
1925 | if (!(vma->vm_flags & VM_GROWSUP)) | |
1926 | return -EFAULT; | |
1927 | ||
bd726c90 | 1928 | /* Guard against exceeding limits of the address space. */ |
1be7107f | 1929 | address &= PAGE_MASK; |
37511fb5 | 1930 | if (address >= (TASK_SIZE & PAGE_MASK)) |
12352d3c | 1931 | return -ENOMEM; |
bd726c90 | 1932 | address += PAGE_SIZE; |
12352d3c | 1933 | |
1be7107f HD |
1934 | /* Enforce stack_guard_gap */ |
1935 | gap_addr = address + stack_guard_gap; | |
bd726c90 HD |
1936 | |
1937 | /* Guard against overflow */ | |
1938 | if (gap_addr < address || gap_addr > TASK_SIZE) | |
1939 | gap_addr = TASK_SIZE; | |
1940 | ||
763ecb03 LH |
1941 | next = find_vma_intersection(mm, vma->vm_end, gap_addr); |
1942 | if (next && vma_is_accessible(next)) { | |
1be7107f HD |
1943 | if (!(next->vm_flags & VM_GROWSUP)) |
1944 | return -ENOMEM; | |
1945 | /* Check that both stack segments have the same anon_vma? */ | |
1946 | } | |
1947 | ||
c5d5546e | 1948 | if (mas_preallocate(&mas, GFP_KERNEL)) |
d4af56c5 LH |
1949 | return -ENOMEM; |
1950 | ||
12352d3c | 1951 | /* We must make sure the anon_vma is allocated. */ |
d4af56c5 LH |
1952 | if (unlikely(anon_vma_prepare(vma))) { |
1953 | mas_destroy(&mas); | |
1da177e4 | 1954 | return -ENOMEM; |
d4af56c5 | 1955 | } |
1da177e4 LT |
1956 | |
1957 | /* | |
1958 | * vma->vm_start/vm_end cannot change under us because the caller | |
c1e8d7c6 | 1959 | * is required to hold the mmap_lock in read mode. We need the |
1da177e4 LT |
1960 | * anon_vma lock to serialize against concurrent expand_stacks. |
1961 | */ | |
12352d3c | 1962 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
1963 | |
1964 | /* Somebody else might have raced and expanded it already */ | |
1965 | if (address > vma->vm_end) { | |
1966 | unsigned long size, grow; | |
1967 | ||
1968 | size = address - vma->vm_start; | |
1969 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
1970 | ||
42c36f63 HD |
1971 | error = -ENOMEM; |
1972 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
1973 | error = acct_stack_growth(vma, size, grow); | |
1974 | if (!error) { | |
4128997b | 1975 | /* |
524e00b3 LH |
1976 | * We only hold a shared mmap_lock lock here, so |
1977 | * we need to protect against concurrent vma | |
1978 | * expansions. anon_vma_lock_write() doesn't | |
1979 | * help here, as we don't guarantee that all | |
1980 | * growable vmas in a mm share the same root | |
1981 | * anon vma. So, we reuse mm->page_table_lock | |
1982 | * to guard against concurrent vma expansions. | |
4128997b | 1983 | */ |
09357814 | 1984 | spin_lock(&mm->page_table_lock); |
87e8827b | 1985 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 1986 | mm->locked_vm += grow; |
84638335 | 1987 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 1988 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 1989 | vma->vm_end = address; |
d4af56c5 | 1990 | /* Overwrite old entry in mtree. */ |
fbcc3104 LH |
1991 | mas_set_range(&mas, vma->vm_start, address - 1); |
1992 | mas_store_prealloc(&mas, vma); | |
bf181b9f | 1993 | anon_vma_interval_tree_post_update_vma(vma); |
09357814 | 1994 | spin_unlock(&mm->page_table_lock); |
4128997b | 1995 | |
42c36f63 HD |
1996 | perf_event_mmap(vma); |
1997 | } | |
3af9e859 | 1998 | } |
1da177e4 | 1999 | } |
12352d3c | 2000 | anon_vma_unlock_write(vma->anon_vma); |
c791576c | 2001 | khugepaged_enter_vma(vma, vma->vm_flags); |
d4af56c5 | 2002 | mas_destroy(&mas); |
1da177e4 LT |
2003 | return error; |
2004 | } | |
46dea3d0 HD |
2005 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2006 | ||
1da177e4 LT |
2007 | /* |
2008 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
2009 | */ | |
524e00b3 | 2010 | int expand_downwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 2011 | { |
09357814 | 2012 | struct mm_struct *mm = vma->vm_mm; |
763ecb03 | 2013 | MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); |
1be7107f | 2014 | struct vm_area_struct *prev; |
0a1d5299 | 2015 | int error = 0; |
1da177e4 | 2016 | |
8869477a | 2017 | address &= PAGE_MASK; |
0a1d5299 JH |
2018 | if (address < mmap_min_addr) |
2019 | return -EPERM; | |
8869477a | 2020 | |
1be7107f | 2021 | /* Enforce stack_guard_gap */ |
763ecb03 | 2022 | prev = mas_prev(&mas, 0); |
32e4e6d5 ON |
2023 | /* Check that both stack segments have the same anon_vma? */ |
2024 | if (prev && !(prev->vm_flags & VM_GROWSDOWN) && | |
3122e80e | 2025 | vma_is_accessible(prev)) { |
32e4e6d5 | 2026 | if (address - prev->vm_end < stack_guard_gap) |
1be7107f | 2027 | return -ENOMEM; |
1be7107f HD |
2028 | } |
2029 | ||
c5d5546e | 2030 | if (mas_preallocate(&mas, GFP_KERNEL)) |
d4af56c5 LH |
2031 | return -ENOMEM; |
2032 | ||
12352d3c | 2033 | /* We must make sure the anon_vma is allocated. */ |
d4af56c5 LH |
2034 | if (unlikely(anon_vma_prepare(vma))) { |
2035 | mas_destroy(&mas); | |
12352d3c | 2036 | return -ENOMEM; |
d4af56c5 | 2037 | } |
1da177e4 LT |
2038 | |
2039 | /* | |
2040 | * vma->vm_start/vm_end cannot change under us because the caller | |
c1e8d7c6 | 2041 | * is required to hold the mmap_lock in read mode. We need the |
1da177e4 LT |
2042 | * anon_vma lock to serialize against concurrent expand_stacks. |
2043 | */ | |
12352d3c | 2044 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2045 | |
2046 | /* Somebody else might have raced and expanded it already */ | |
2047 | if (address < vma->vm_start) { | |
2048 | unsigned long size, grow; | |
2049 | ||
2050 | size = vma->vm_end - address; | |
2051 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2052 | ||
a626ca6a LT |
2053 | error = -ENOMEM; |
2054 | if (grow <= vma->vm_pgoff) { | |
2055 | error = acct_stack_growth(vma, size, grow); | |
2056 | if (!error) { | |
4128997b | 2057 | /* |
524e00b3 LH |
2058 | * We only hold a shared mmap_lock lock here, so |
2059 | * we need to protect against concurrent vma | |
2060 | * expansions. anon_vma_lock_write() doesn't | |
2061 | * help here, as we don't guarantee that all | |
2062 | * growable vmas in a mm share the same root | |
2063 | * anon vma. So, we reuse mm->page_table_lock | |
2064 | * to guard against concurrent vma expansions. | |
4128997b | 2065 | */ |
09357814 | 2066 | spin_lock(&mm->page_table_lock); |
87e8827b | 2067 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2068 | mm->locked_vm += grow; |
84638335 | 2069 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2070 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2071 | vma->vm_start = address; |
2072 | vma->vm_pgoff -= grow; | |
d4af56c5 | 2073 | /* Overwrite old entry in mtree. */ |
fbcc3104 LH |
2074 | mas_set_range(&mas, address, vma->vm_end - 1); |
2075 | mas_store_prealloc(&mas, vma); | |
bf181b9f | 2076 | anon_vma_interval_tree_post_update_vma(vma); |
09357814 | 2077 | spin_unlock(&mm->page_table_lock); |
4128997b | 2078 | |
a626ca6a LT |
2079 | perf_event_mmap(vma); |
2080 | } | |
1da177e4 LT |
2081 | } |
2082 | } | |
12352d3c | 2083 | anon_vma_unlock_write(vma->anon_vma); |
c791576c | 2084 | khugepaged_enter_vma(vma, vma->vm_flags); |
d4af56c5 | 2085 | mas_destroy(&mas); |
1da177e4 LT |
2086 | return error; |
2087 | } | |
2088 | ||
1be7107f HD |
2089 | /* enforced gap between the expanding stack and other mappings. */ |
2090 | unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; | |
2091 | ||
2092 | static int __init cmdline_parse_stack_guard_gap(char *p) | |
2093 | { | |
2094 | unsigned long val; | |
2095 | char *endptr; | |
2096 | ||
2097 | val = simple_strtoul(p, &endptr, 10); | |
2098 | if (!*endptr) | |
2099 | stack_guard_gap = val << PAGE_SHIFT; | |
2100 | ||
e6d09493 | 2101 | return 1; |
1be7107f HD |
2102 | } |
2103 | __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); | |
2104 | ||
b6a2fea3 OW |
2105 | #ifdef CONFIG_STACK_GROWSUP |
2106 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2107 | { | |
2108 | return expand_upwards(vma, address); | |
2109 | } | |
2110 | ||
2111 | struct vm_area_struct * | |
2112 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2113 | { | |
2114 | struct vm_area_struct *vma, *prev; | |
2115 | ||
2116 | addr &= PAGE_MASK; | |
2117 | vma = find_vma_prev(mm, addr, &prev); | |
2118 | if (vma && (vma->vm_start <= addr)) | |
2119 | return vma; | |
4d45e75a | 2120 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2121 | return NULL; |
cea10a19 | 2122 | if (prev->vm_flags & VM_LOCKED) |
fc05f566 | 2123 | populate_vma_page_range(prev, addr, prev->vm_end, NULL); |
b6a2fea3 OW |
2124 | return prev; |
2125 | } | |
2126 | #else | |
2127 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2128 | { | |
2129 | return expand_downwards(vma, address); | |
2130 | } | |
2131 | ||
1da177e4 | 2132 | struct vm_area_struct * |
cc71aba3 | 2133 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2134 | { |
cc71aba3 | 2135 | struct vm_area_struct *vma; |
1da177e4 LT |
2136 | unsigned long start; |
2137 | ||
2138 | addr &= PAGE_MASK; | |
cc71aba3 | 2139 | vma = find_vma(mm, addr); |
1da177e4 LT |
2140 | if (!vma) |
2141 | return NULL; | |
2142 | if (vma->vm_start <= addr) | |
2143 | return vma; | |
2144 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2145 | return NULL; | |
2146 | start = vma->vm_start; | |
2147 | if (expand_stack(vma, addr)) | |
2148 | return NULL; | |
cea10a19 | 2149 | if (vma->vm_flags & VM_LOCKED) |
fc05f566 | 2150 | populate_vma_page_range(vma, addr, start, NULL); |
1da177e4 LT |
2151 | return vma; |
2152 | } | |
2153 | #endif | |
2154 | ||
e1d6d01a JB |
2155 | EXPORT_SYMBOL_GPL(find_extend_vma); |
2156 | ||
1da177e4 | 2157 | /* |
763ecb03 LH |
2158 | * Ok - we have the memory areas we should free on a maple tree so release them, |
2159 | * and do the vma updates. | |
2c0b3814 HD |
2160 | * |
2161 | * Called with the mm semaphore held. | |
1da177e4 | 2162 | */ |
763ecb03 | 2163 | static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) |
1da177e4 | 2164 | { |
4f74d2c8 | 2165 | unsigned long nr_accounted = 0; |
763ecb03 | 2166 | struct vm_area_struct *vma; |
4f74d2c8 | 2167 | |
365e9c87 HD |
2168 | /* Update high watermark before we lower total_vm */ |
2169 | update_hiwater_vm(mm); | |
763ecb03 | 2170 | mas_for_each(mas, vma, ULONG_MAX) { |
2c0b3814 HD |
2171 | long nrpages = vma_pages(vma); |
2172 | ||
4f74d2c8 LT |
2173 | if (vma->vm_flags & VM_ACCOUNT) |
2174 | nr_accounted += nrpages; | |
84638335 | 2175 | vm_stat_account(mm, vma->vm_flags, -nrpages); |
0d2ebf9c | 2176 | remove_vma(vma, false); |
763ecb03 | 2177 | } |
4f74d2c8 | 2178 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2179 | validate_mm(mm); |
2180 | } | |
2181 | ||
2182 | /* | |
2183 | * Get rid of page table information in the indicated region. | |
2184 | * | |
f10df686 | 2185 | * Called with the mm semaphore held. |
1da177e4 | 2186 | */ |
763ecb03 | 2187 | static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, |
e0da382c | 2188 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
763ecb03 | 2189 | struct vm_area_struct *next, |
68f48381 | 2190 | unsigned long start, unsigned long end, bool mm_wr_locked) |
1da177e4 | 2191 | { |
d16dfc55 | 2192 | struct mmu_gather tlb; |
1da177e4 LT |
2193 | |
2194 | lru_add_drain(); | |
a72afd87 | 2195 | tlb_gather_mmu(&tlb, mm); |
365e9c87 | 2196 | update_hiwater_rss(mm); |
68f48381 | 2197 | unmap_vmas(&tlb, mt, vma, start, end, mm_wr_locked); |
763ecb03 | 2198 | free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
98e51a22 SB |
2199 | next ? next->vm_start : USER_PGTABLES_CEILING, |
2200 | mm_wr_locked); | |
ae8eba8b | 2201 | tlb_finish_mmu(&tlb); |
1da177e4 LT |
2202 | } |
2203 | ||
1da177e4 | 2204 | /* |
def5efe0 DR |
2205 | * __split_vma() bypasses sysctl_max_map_count checking. We use this where it |
2206 | * has already been checked or doesn't make sense to fail. | |
0fd5a9e2 | 2207 | * VMA Iterator will point to the end VMA. |
1da177e4 | 2208 | */ |
9760ebff | 2209 | int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, |
def5efe0 | 2210 | unsigned long addr, int new_below) |
1da177e4 | 2211 | { |
b2b3b886 | 2212 | struct vma_prepare vp; |
1da177e4 | 2213 | struct vm_area_struct *new; |
e3975891 | 2214 | int err; |
9760ebff | 2215 | |
b50e195f | 2216 | validate_mm(vma->vm_mm); |
1da177e4 | 2217 | |
b2b3b886 LH |
2218 | WARN_ON(vma->vm_start >= addr); |
2219 | WARN_ON(vma->vm_end <= addr); | |
2220 | ||
dd3b614f DS |
2221 | if (vma->vm_ops && vma->vm_ops->may_split) { |
2222 | err = vma->vm_ops->may_split(vma, addr); | |
31383c68 DW |
2223 | if (err) |
2224 | return err; | |
2225 | } | |
1da177e4 | 2226 | |
3928d4f5 | 2227 | new = vm_area_dup(vma); |
1da177e4 | 2228 | if (!new) |
e3975891 | 2229 | return -ENOMEM; |
1da177e4 | 2230 | |
b2b3b886 LH |
2231 | err = -ENOMEM; |
2232 | if (vma_iter_prealloc(vmi)) | |
2233 | goto out_free_vma; | |
2234 | ||
2235 | if (new_below) { | |
1da177e4 | 2236 | new->vm_end = addr; |
b2b3b886 | 2237 | } else { |
1da177e4 LT |
2238 | new->vm_start = addr; |
2239 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2240 | } | |
2241 | ||
ef0855d3 ON |
2242 | err = vma_dup_policy(vma, new); |
2243 | if (err) | |
b2b3b886 | 2244 | goto out_free_vmi; |
1da177e4 | 2245 | |
c4ea95d7 DF |
2246 | err = anon_vma_clone(new, vma); |
2247 | if (err) | |
5beb4930 RR |
2248 | goto out_free_mpol; |
2249 | ||
e9714acf | 2250 | if (new->vm_file) |
1da177e4 LT |
2251 | get_file(new->vm_file); |
2252 | ||
2253 | if (new->vm_ops && new->vm_ops->open) | |
2254 | new->vm_ops->open(new); | |
2255 | ||
b2b3b886 LH |
2256 | init_vma_prep(&vp, vma); |
2257 | vp.insert = new; | |
2258 | vma_prepare(&vp); | |
ccf1d78d | 2259 | vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); |
1da177e4 | 2260 | |
b2b3b886 LH |
2261 | if (new_below) { |
2262 | vma->vm_start = addr; | |
2263 | vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; | |
2264 | } else { | |
2265 | vma->vm_end = addr; | |
9760ebff | 2266 | } |
5beb4930 | 2267 | |
b2b3b886 LH |
2268 | /* vma_complete stores the new vma */ |
2269 | vma_complete(&vp, vmi, vma->vm_mm); | |
2270 | ||
2271 | /* Success. */ | |
2272 | if (new_below) | |
2273 | vma_next(vmi); | |
b50e195f | 2274 | validate_mm(vma->vm_mm); |
b2b3b886 LH |
2275 | return 0; |
2276 | ||
2277 | out_free_mpol: | |
ef0855d3 | 2278 | mpol_put(vma_policy(new)); |
b2b3b886 LH |
2279 | out_free_vmi: |
2280 | vma_iter_free(vmi); | |
2281 | out_free_vma: | |
3928d4f5 | 2282 | vm_area_free(new); |
b50e195f | 2283 | validate_mm(vma->vm_mm); |
5beb4930 | 2284 | return err; |
1da177e4 LT |
2285 | } |
2286 | ||
659ace58 KM |
2287 | /* |
2288 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2289 | * either for the first part or the tail. | |
2290 | */ | |
9760ebff | 2291 | int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, |
659ace58 KM |
2292 | unsigned long addr, int new_below) |
2293 | { | |
9760ebff | 2294 | if (vma->vm_mm->map_count >= sysctl_max_map_count) |
659ace58 KM |
2295 | return -ENOMEM; |
2296 | ||
9760ebff | 2297 | return __split_vma(vmi, vma, addr, new_below); |
f2ebfe43 LH |
2298 | } |
2299 | ||
763ecb03 LH |
2300 | static inline int munmap_sidetree(struct vm_area_struct *vma, |
2301 | struct ma_state *mas_detach) | |
1da177e4 | 2302 | { |
73046fd0 | 2303 | vma_start_write(vma); |
763ecb03 LH |
2304 | mas_set_range(mas_detach, vma->vm_start, vma->vm_end - 1); |
2305 | if (mas_store_gfp(mas_detach, vma, GFP_KERNEL)) | |
2306 | return -ENOMEM; | |
1da177e4 | 2307 | |
457f67be | 2308 | vma_mark_detached(vma, true); |
763ecb03 LH |
2309 | if (vma->vm_flags & VM_LOCKED) |
2310 | vma->vm_mm->locked_vm -= vma_pages(vma); | |
1da177e4 | 2311 | |
763ecb03 | 2312 | return 0; |
11f9a21a | 2313 | } |
5a28fc94 | 2314 | |
11f9a21a | 2315 | /* |
183654ce LH |
2316 | * do_vmi_align_munmap() - munmap the aligned region from @start to @end. |
2317 | * @vmi: The vma iterator | |
11f9a21a LH |
2318 | * @vma: The starting vm_area_struct |
2319 | * @mm: The mm_struct | |
2320 | * @start: The aligned start address to munmap. | |
2321 | * @end: The aligned end address to munmap. | |
2322 | * @uf: The userfaultfd list_head | |
8651a137 | 2323 | * @downgrade: Set to true to attempt a write downgrade of the mmap_lock |
11f9a21a LH |
2324 | * |
2325 | * If @downgrade is true, check return code for potential release of the lock. | |
2326 | */ | |
2327 | static int | |
183654ce | 2328 | do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, |
11f9a21a LH |
2329 | struct mm_struct *mm, unsigned long start, |
2330 | unsigned long end, struct list_head *uf, bool downgrade) | |
2331 | { | |
763ecb03 LH |
2332 | struct vm_area_struct *prev, *next = NULL; |
2333 | struct maple_tree mt_detach; | |
2334 | int count = 0; | |
11f9a21a | 2335 | int error = -ENOMEM; |
763ecb03 | 2336 | MA_STATE(mas_detach, &mt_detach, 0, 0); |
3dd44325 | 2337 | mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK); |
763ecb03 | 2338 | mt_set_external_lock(&mt_detach, &mm->mmap_lock); |
d4af56c5 | 2339 | |
1da177e4 LT |
2340 | /* |
2341 | * If we need to split any vma, do it now to save pain later. | |
2342 | * | |
2343 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2344 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2345 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2346 | */ | |
763ecb03 LH |
2347 | |
2348 | /* Does it split the first one? */ | |
146425a3 | 2349 | if (start > vma->vm_start) { |
659ace58 KM |
2350 | |
2351 | /* | |
2352 | * Make sure that map_count on return from munmap() will | |
2353 | * not exceed its limit; but let map_count go just above | |
2354 | * its limit temporarily, to help free resources as expected. | |
2355 | */ | |
2356 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
d4af56c5 | 2357 | goto map_count_exceeded; |
659ace58 | 2358 | |
9760ebff | 2359 | error = __split_vma(vmi, vma, start, 0); |
1da177e4 | 2360 | if (error) |
763ecb03 | 2361 | goto start_split_failed; |
11f9a21a | 2362 | |
0fd5a9e2 | 2363 | vma = vma_iter_load(vmi); |
1da177e4 LT |
2364 | } |
2365 | ||
183654ce | 2366 | prev = vma_prev(vmi); |
763ecb03 | 2367 | if (unlikely((!prev))) |
183654ce | 2368 | vma_iter_set(vmi, start); |
763ecb03 LH |
2369 | |
2370 | /* | |
2371 | * Detach a range of VMAs from the mm. Using next as a temp variable as | |
2372 | * it is always overwritten. | |
2373 | */ | |
183654ce | 2374 | for_each_vma_range(*vmi, next, end) { |
763ecb03 LH |
2375 | /* Does it split the end? */ |
2376 | if (next->vm_end > end) { | |
6b73cff2 | 2377 | error = __split_vma(vmi, next, end, 0); |
763ecb03 LH |
2378 | if (error) |
2379 | goto end_split_failed; | |
763ecb03 LH |
2380 | } |
2381 | error = munmap_sidetree(next, &mas_detach); | |
1da177e4 | 2382 | if (error) |
763ecb03 | 2383 | goto munmap_sidetree_failed; |
11f9a21a | 2384 | |
763ecb03 LH |
2385 | count++; |
2386 | #ifdef CONFIG_DEBUG_VM_MAPLE_TREE | |
2387 | BUG_ON(next->vm_start < start); | |
2388 | BUG_ON(next->vm_start > end); | |
2389 | #endif | |
1da177e4 | 2390 | } |
1da177e4 | 2391 | |
15c0c60b LH |
2392 | if (vma_iter_end(vmi) > end) |
2393 | next = vma_iter_load(vmi); | |
2394 | ||
2395 | if (!next) | |
2396 | next = vma_next(vmi); | |
2397 | ||
2376dd7c AA |
2398 | if (unlikely(uf)) { |
2399 | /* | |
2400 | * If userfaultfd_unmap_prep returns an error the vmas | |
f0953a1b | 2401 | * will remain split, but userland will get a |
2376dd7c AA |
2402 | * highly unexpected error anyway. This is no |
2403 | * different than the case where the first of the two | |
2404 | * __split_vma fails, but we don't undo the first | |
2405 | * split, despite we could. This is unlikely enough | |
2406 | * failure that it's not worth optimizing it for. | |
2407 | */ | |
69dbe6da | 2408 | error = userfaultfd_unmap_prep(mm, start, end, uf); |
11f9a21a | 2409 | |
2376dd7c | 2410 | if (error) |
d4af56c5 | 2411 | goto userfaultfd_error; |
2376dd7c AA |
2412 | } |
2413 | ||
763ecb03 LH |
2414 | #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) |
2415 | /* Make sure no VMAs are about to be lost. */ | |
2416 | { | |
2417 | MA_STATE(test, &mt_detach, start, end - 1); | |
2418 | struct vm_area_struct *vma_mas, *vma_test; | |
2419 | int test_count = 0; | |
2420 | ||
183654ce | 2421 | vma_iter_set(vmi, start); |
763ecb03 LH |
2422 | rcu_read_lock(); |
2423 | vma_test = mas_find(&test, end - 1); | |
183654ce | 2424 | for_each_vma_range(*vmi, vma_mas, end) { |
763ecb03 LH |
2425 | BUG_ON(vma_mas != vma_test); |
2426 | test_count++; | |
2427 | vma_test = mas_next(&test, end - 1); | |
2428 | } | |
2429 | rcu_read_unlock(); | |
2430 | BUG_ON(count != test_count); | |
763ecb03 LH |
2431 | } |
2432 | #endif | |
0378c0a0 | 2433 | /* Point of no return */ |
183654ce LH |
2434 | vma_iter_set(vmi, start); |
2435 | if (vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL)) | |
0378c0a0 LH |
2436 | return -ENOMEM; |
2437 | ||
763ecb03 | 2438 | mm->map_count -= count; |
11f9a21a LH |
2439 | /* |
2440 | * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or | |
2441 | * VM_GROWSUP VMA. Such VMAs can change their size under | |
2442 | * down_read(mmap_lock) and collide with the VMA we are about to unmap. | |
2443 | */ | |
2444 | if (downgrade) { | |
763ecb03 | 2445 | if (next && (next->vm_flags & VM_GROWSDOWN)) |
11f9a21a LH |
2446 | downgrade = false; |
2447 | else if (prev && (prev->vm_flags & VM_GROWSUP)) | |
2448 | downgrade = false; | |
2449 | else | |
2450 | mmap_write_downgrade(mm); | |
2451 | } | |
dd2283f2 | 2452 | |
68f48381 SB |
2453 | /* |
2454 | * We can free page tables without write-locking mmap_lock because VMAs | |
2455 | * were isolated before we downgraded mmap_lock. | |
2456 | */ | |
2457 | unmap_region(mm, &mt_detach, vma, prev, next, start, end, !downgrade); | |
763ecb03 LH |
2458 | /* Statistics and freeing VMAs */ |
2459 | mas_set(&mas_detach, start); | |
2460 | remove_mt(mm, &mas_detach); | |
2461 | __mt_destroy(&mt_detach); | |
1da177e4 | 2462 | |
524e00b3 LH |
2463 | |
2464 | validate_mm(mm); | |
dd2283f2 | 2465 | return downgrade ? 1 : 0; |
d4af56c5 | 2466 | |
d4af56c5 | 2467 | userfaultfd_error: |
763ecb03 LH |
2468 | munmap_sidetree_failed: |
2469 | end_split_failed: | |
2470 | __mt_destroy(&mt_detach); | |
2471 | start_split_failed: | |
2472 | map_count_exceeded: | |
d4af56c5 | 2473 | return error; |
1da177e4 | 2474 | } |
1da177e4 | 2475 | |
11f9a21a | 2476 | /* |
183654ce LH |
2477 | * do_vmi_munmap() - munmap a given range. |
2478 | * @vmi: The vma iterator | |
11f9a21a LH |
2479 | * @mm: The mm_struct |
2480 | * @start: The start address to munmap | |
2481 | * @len: The length of the range to munmap | |
2482 | * @uf: The userfaultfd list_head | |
2483 | * @downgrade: set to true if the user wants to attempt to write_downgrade the | |
8651a137 | 2484 | * mmap_lock |
11f9a21a LH |
2485 | * |
2486 | * This function takes a @mas that is either pointing to the previous VMA or set | |
2487 | * to MA_START and sets it up to remove the mapping(s). The @len will be | |
2488 | * aligned and any arch_unmap work will be preformed. | |
2489 | * | |
2490 | * Returns: -EINVAL on failure, 1 on success and unlock, 0 otherwise. | |
2491 | */ | |
183654ce | 2492 | int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, |
11f9a21a LH |
2493 | unsigned long start, size_t len, struct list_head *uf, |
2494 | bool downgrade) | |
2495 | { | |
2496 | unsigned long end; | |
2497 | struct vm_area_struct *vma; | |
2498 | ||
2499 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) | |
2500 | return -EINVAL; | |
2501 | ||
2502 | end = start + PAGE_ALIGN(len); | |
2503 | if (end == start) | |
2504 | return -EINVAL; | |
2505 | ||
2506 | /* arch_unmap() might do unmaps itself. */ | |
2507 | arch_unmap(mm, start, end); | |
2508 | ||
2509 | /* Find the first overlapping VMA */ | |
183654ce | 2510 | vma = vma_find(vmi, end); |
11f9a21a LH |
2511 | if (!vma) |
2512 | return 0; | |
2513 | ||
183654ce | 2514 | return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); |
11f9a21a LH |
2515 | } |
2516 | ||
2517 | /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. | |
2518 | * @mm: The mm_struct | |
2519 | * @start: The start address to munmap | |
2520 | * @len: The length to be munmapped. | |
2521 | * @uf: The userfaultfd list_head | |
2522 | */ | |
dd2283f2 YS |
2523 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, |
2524 | struct list_head *uf) | |
2525 | { | |
183654ce | 2526 | VMA_ITERATOR(vmi, mm, start); |
11f9a21a | 2527 | |
183654ce | 2528 | return do_vmi_munmap(&vmi, mm, start, len, uf, false); |
dd2283f2 YS |
2529 | } |
2530 | ||
e99668a5 LH |
2531 | unsigned long mmap_region(struct file *file, unsigned long addr, |
2532 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, | |
2533 | struct list_head *uf) | |
2534 | { | |
2535 | struct mm_struct *mm = current->mm; | |
2536 | struct vm_area_struct *vma = NULL; | |
2537 | struct vm_area_struct *next, *prev, *merge; | |
2538 | pgoff_t pglen = len >> PAGE_SHIFT; | |
2539 | unsigned long charged = 0; | |
2540 | unsigned long end = addr + len; | |
2541 | unsigned long merge_start = addr, merge_end = end; | |
2542 | pgoff_t vm_pgoff; | |
2543 | int error; | |
183654ce | 2544 | VMA_ITERATOR(vmi, mm, addr); |
e99668a5 LH |
2545 | |
2546 | /* Check against address space limit. */ | |
2547 | if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { | |
2548 | unsigned long nr_pages; | |
2549 | ||
2550 | /* | |
2551 | * MAP_FIXED may remove pages of mappings that intersects with | |
2552 | * requested mapping. Account for the pages it would unmap. | |
2553 | */ | |
2554 | nr_pages = count_vma_pages_range(mm, addr, end); | |
2555 | ||
2556 | if (!may_expand_vm(mm, vm_flags, | |
2557 | (len >> PAGE_SHIFT) - nr_pages)) | |
2558 | return -ENOMEM; | |
2559 | } | |
2560 | ||
2561 | /* Unmap any existing mapping in the area */ | |
183654ce | 2562 | if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) |
e99668a5 LH |
2563 | return -ENOMEM; |
2564 | ||
2565 | /* | |
2566 | * Private writable mapping: check memory availability | |
2567 | */ | |
2568 | if (accountable_mapping(file, vm_flags)) { | |
2569 | charged = len >> PAGE_SHIFT; | |
2570 | if (security_vm_enough_memory_mm(mm, charged)) | |
2571 | return -ENOMEM; | |
2572 | vm_flags |= VM_ACCOUNT; | |
2573 | } | |
2574 | ||
183654ce LH |
2575 | next = vma_next(&vmi); |
2576 | prev = vma_prev(&vmi); | |
e99668a5 LH |
2577 | if (vm_flags & VM_SPECIAL) |
2578 | goto cannot_expand; | |
2579 | ||
2580 | /* Attempt to expand an old mapping */ | |
2581 | /* Check next */ | |
2582 | if (next && next->vm_start == end && !vma_policy(next) && | |
2583 | can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, | |
2584 | NULL_VM_UFFD_CTX, NULL)) { | |
2585 | merge_end = next->vm_end; | |
2586 | vma = next; | |
2587 | vm_pgoff = next->vm_pgoff - pglen; | |
2588 | } | |
2589 | ||
2590 | /* Check prev */ | |
2591 | if (prev && prev->vm_end == addr && !vma_policy(prev) && | |
2592 | (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, | |
2593 | pgoff, vma->vm_userfaultfd_ctx, NULL) : | |
2594 | can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, | |
2595 | NULL_VM_UFFD_CTX, NULL))) { | |
2596 | merge_start = prev->vm_start; | |
2597 | vma = prev; | |
2598 | vm_pgoff = prev->vm_pgoff; | |
2599 | } | |
2600 | ||
2601 | ||
2602 | /* Actually expand, if possible */ | |
2603 | if (vma && | |
3c441ab7 | 2604 | !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { |
e99668a5 LH |
2605 | khugepaged_enter_vma(vma, vm_flags); |
2606 | goto expanded; | |
2607 | } | |
2608 | ||
e99668a5 | 2609 | cannot_expand: |
5c1c03de LH |
2610 | if (prev) |
2611 | vma_iter_next_range(&vmi); | |
2612 | ||
e99668a5 LH |
2613 | /* |
2614 | * Determine the object being mapped and call the appropriate | |
2615 | * specific mapper. the address has already been validated, but | |
2616 | * not unmapped, but the maps are removed from the list. | |
2617 | */ | |
2618 | vma = vm_area_alloc(mm); | |
2619 | if (!vma) { | |
2620 | error = -ENOMEM; | |
2621 | goto unacct_error; | |
2622 | } | |
2623 | ||
0fd5a9e2 | 2624 | vma_iter_set(&vmi, addr); |
e99668a5 LH |
2625 | vma->vm_start = addr; |
2626 | vma->vm_end = end; | |
1c71222e | 2627 | vm_flags_init(vma, vm_flags); |
e99668a5 LH |
2628 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
2629 | vma->vm_pgoff = pgoff; | |
2630 | ||
2631 | if (file) { | |
2632 | if (vm_flags & VM_SHARED) { | |
2633 | error = mapping_map_writable(file->f_mapping); | |
2634 | if (error) | |
2635 | goto free_vma; | |
2636 | } | |
2637 | ||
2638 | vma->vm_file = get_file(file); | |
2639 | error = call_mmap(file, vma); | |
2640 | if (error) | |
2641 | goto unmap_and_free_vma; | |
2642 | ||
a57b7051 LH |
2643 | /* |
2644 | * Expansion is handled above, merging is handled below. | |
2645 | * Drivers should not alter the address of the VMA. | |
e99668a5 | 2646 | */ |
cc8d1b09 LH |
2647 | error = -EINVAL; |
2648 | if (WARN_ON((addr != vma->vm_start))) | |
a57b7051 | 2649 | goto close_and_free_vma; |
e99668a5 | 2650 | |
cc8d1b09 | 2651 | vma_iter_set(&vmi, addr); |
e99668a5 LH |
2652 | /* |
2653 | * If vm_flags changed after call_mmap(), we should try merge | |
2654 | * vma again as we may succeed this time. | |
2655 | */ | |
2656 | if (unlikely(vm_flags != vma->vm_flags && prev)) { | |
9760ebff LH |
2657 | merge = vma_merge(&vmi, mm, prev, vma->vm_start, |
2658 | vma->vm_end, vma->vm_flags, NULL, | |
2659 | vma->vm_file, vma->vm_pgoff, NULL, | |
2660 | NULL_VM_UFFD_CTX, NULL); | |
e99668a5 LH |
2661 | if (merge) { |
2662 | /* | |
2663 | * ->mmap() can change vma->vm_file and fput | |
2664 | * the original file. So fput the vma->vm_file | |
2665 | * here or we would add an extra fput for file | |
2666 | * and cause general protection fault | |
2667 | * ultimately. | |
2668 | */ | |
2669 | fput(vma->vm_file); | |
2670 | vm_area_free(vma); | |
2671 | vma = merge; | |
2672 | /* Update vm_flags to pick up the change. */ | |
e99668a5 LH |
2673 | vm_flags = vma->vm_flags; |
2674 | goto unmap_writable; | |
2675 | } | |
2676 | } | |
2677 | ||
2678 | vm_flags = vma->vm_flags; | |
2679 | } else if (vm_flags & VM_SHARED) { | |
2680 | error = shmem_zero_setup(vma); | |
2681 | if (error) | |
2682 | goto free_vma; | |
2683 | } else { | |
2684 | vma_set_anonymous(vma); | |
2685 | } | |
2686 | ||
b507808e JG |
2687 | if (map_deny_write_exec(vma, vma->vm_flags)) { |
2688 | error = -EACCES; | |
6bbf1090 | 2689 | goto close_and_free_vma; |
b507808e JG |
2690 | } |
2691 | ||
e99668a5 | 2692 | /* Allow architectures to sanity-check the vm_flags */ |
cc8d1b09 LH |
2693 | error = -EINVAL; |
2694 | if (!arch_validate_flags(vma->vm_flags)) | |
2695 | goto close_and_free_vma; | |
e99668a5 | 2696 | |
cc8d1b09 LH |
2697 | error = -ENOMEM; |
2698 | if (vma_iter_prealloc(&vmi)) | |
2699 | goto close_and_free_vma; | |
e99668a5 LH |
2700 | |
2701 | if (vma->vm_file) | |
2702 | i_mmap_lock_write(vma->vm_file->f_mapping); | |
2703 | ||
183654ce | 2704 | vma_iter_store(&vmi, vma); |
e99668a5 LH |
2705 | mm->map_count++; |
2706 | if (vma->vm_file) { | |
2707 | if (vma->vm_flags & VM_SHARED) | |
2708 | mapping_allow_writable(vma->vm_file->f_mapping); | |
2709 | ||
2710 | flush_dcache_mmap_lock(vma->vm_file->f_mapping); | |
2711 | vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); | |
2712 | flush_dcache_mmap_unlock(vma->vm_file->f_mapping); | |
2713 | i_mmap_unlock_write(vma->vm_file->f_mapping); | |
2714 | } | |
2715 | ||
2716 | /* | |
2717 | * vma_merge() calls khugepaged_enter_vma() either, the below | |
2718 | * call covers the non-merge case. | |
2719 | */ | |
2720 | khugepaged_enter_vma(vma, vma->vm_flags); | |
2721 | ||
2722 | /* Once vma denies write, undo our temporary denial count */ | |
2723 | unmap_writable: | |
2724 | if (file && vm_flags & VM_SHARED) | |
2725 | mapping_unmap_writable(file->f_mapping); | |
2726 | file = vma->vm_file; | |
d7597f59 | 2727 | ksm_add_vma(vma); |
e99668a5 LH |
2728 | expanded: |
2729 | perf_event_mmap(vma); | |
2730 | ||
2731 | vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); | |
2732 | if (vm_flags & VM_LOCKED) { | |
2733 | if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || | |
2734 | is_vm_hugetlb_page(vma) || | |
2735 | vma == get_gate_vma(current->mm)) | |
e430a95a | 2736 | vm_flags_clear(vma, VM_LOCKED_MASK); |
e99668a5 LH |
2737 | else |
2738 | mm->locked_vm += (len >> PAGE_SHIFT); | |
2739 | } | |
2740 | ||
2741 | if (file) | |
2742 | uprobe_mmap(vma); | |
2743 | ||
2744 | /* | |
2745 | * New (or expanded) vma always get soft dirty status. | |
2746 | * Otherwise user-space soft-dirty page tracker won't | |
2747 | * be able to distinguish situation when vma area unmapped, | |
2748 | * then new mapped in-place (which must be aimed as | |
2749 | * a completely new data area). | |
2750 | */ | |
1c71222e | 2751 | vm_flags_set(vma, VM_SOFTDIRTY); |
e99668a5 LH |
2752 | |
2753 | vma_set_page_prot(vma); | |
2754 | ||
2755 | validate_mm(mm); | |
2756 | return addr; | |
2757 | ||
deb0f656 | 2758 | close_and_free_vma: |
cc8d1b09 | 2759 | if (file && vma->vm_ops && vma->vm_ops->close) |
deb0f656 | 2760 | vma->vm_ops->close(vma); |
cc8d1b09 LH |
2761 | |
2762 | if (file || vma->vm_file) { | |
e99668a5 | 2763 | unmap_and_free_vma: |
cc8d1b09 LH |
2764 | fput(vma->vm_file); |
2765 | vma->vm_file = NULL; | |
e99668a5 | 2766 | |
cc8d1b09 LH |
2767 | /* Undo any partial mapping done by a device driver. */ |
2768 | unmap_region(mm, &mm->mm_mt, vma, prev, next, vma->vm_start, | |
68f48381 | 2769 | vma->vm_end, true); |
cc8d1b09 | 2770 | } |
cc674ab3 | 2771 | if (file && (vm_flags & VM_SHARED)) |
e99668a5 LH |
2772 | mapping_unmap_writable(file->f_mapping); |
2773 | free_vma: | |
2774 | vm_area_free(vma); | |
2775 | unacct_error: | |
2776 | if (charged) | |
2777 | vm_unacct_memory(charged); | |
2778 | validate_mm(mm); | |
2779 | return error; | |
2780 | } | |
2781 | ||
dd2283f2 | 2782 | static int __vm_munmap(unsigned long start, size_t len, bool downgrade) |
1da177e4 LT |
2783 | { |
2784 | int ret; | |
bfce281c | 2785 | struct mm_struct *mm = current->mm; |
897ab3e0 | 2786 | LIST_HEAD(uf); |
183654ce | 2787 | VMA_ITERATOR(vmi, mm, start); |
1da177e4 | 2788 | |
d8ed45c5 | 2789 | if (mmap_write_lock_killable(mm)) |
ae798783 MH |
2790 | return -EINTR; |
2791 | ||
183654ce | 2792 | ret = do_vmi_munmap(&vmi, mm, start, len, &uf, downgrade); |
dd2283f2 | 2793 | /* |
c1e8d7c6 | 2794 | * Returning 1 indicates mmap_lock is downgraded. |
dd2283f2 YS |
2795 | * But 1 is not legal return value of vm_munmap() and munmap(), reset |
2796 | * it to 0 before return. | |
2797 | */ | |
2798 | if (ret == 1) { | |
d8ed45c5 | 2799 | mmap_read_unlock(mm); |
dd2283f2 YS |
2800 | ret = 0; |
2801 | } else | |
d8ed45c5 | 2802 | mmap_write_unlock(mm); |
dd2283f2 | 2803 | |
897ab3e0 | 2804 | userfaultfd_unmap_complete(mm, &uf); |
1da177e4 LT |
2805 | return ret; |
2806 | } | |
dd2283f2 YS |
2807 | |
2808 | int vm_munmap(unsigned long start, size_t len) | |
2809 | { | |
2810 | return __vm_munmap(start, len, false); | |
2811 | } | |
a46ef99d LT |
2812 | EXPORT_SYMBOL(vm_munmap); |
2813 | ||
2814 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2815 | { | |
ce18d171 | 2816 | addr = untagged_addr(addr); |
dd2283f2 | 2817 | return __vm_munmap(addr, len, true); |
a46ef99d | 2818 | } |
1da177e4 | 2819 | |
c8d78c18 KS |
2820 | |
2821 | /* | |
2822 | * Emulation of deprecated remap_file_pages() syscall. | |
2823 | */ | |
2824 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, | |
2825 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) | |
2826 | { | |
2827 | ||
2828 | struct mm_struct *mm = current->mm; | |
2829 | struct vm_area_struct *vma; | |
2830 | unsigned long populate = 0; | |
2831 | unsigned long ret = -EINVAL; | |
2832 | struct file *file; | |
2833 | ||
ee65728e | 2834 | pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", |
756a025f | 2835 | current->comm, current->pid); |
c8d78c18 KS |
2836 | |
2837 | if (prot) | |
2838 | return ret; | |
2839 | start = start & PAGE_MASK; | |
2840 | size = size & PAGE_MASK; | |
2841 | ||
2842 | if (start + size <= start) | |
2843 | return ret; | |
2844 | ||
2845 | /* Does pgoff wrap? */ | |
2846 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) | |
2847 | return ret; | |
2848 | ||
d8ed45c5 | 2849 | if (mmap_write_lock_killable(mm)) |
dc0ef0df MH |
2850 | return -EINTR; |
2851 | ||
9b593cb2 | 2852 | vma = vma_lookup(mm, start); |
c8d78c18 KS |
2853 | |
2854 | if (!vma || !(vma->vm_flags & VM_SHARED)) | |
2855 | goto out; | |
2856 | ||
48f7df32 | 2857 | if (start + size > vma->vm_end) { |
763ecb03 LH |
2858 | VMA_ITERATOR(vmi, mm, vma->vm_end); |
2859 | struct vm_area_struct *next, *prev = vma; | |
48f7df32 | 2860 | |
763ecb03 | 2861 | for_each_vma_range(vmi, next, start + size) { |
48f7df32 | 2862 | /* hole between vmas ? */ |
763ecb03 | 2863 | if (next->vm_start != prev->vm_end) |
48f7df32 KS |
2864 | goto out; |
2865 | ||
2866 | if (next->vm_file != vma->vm_file) | |
2867 | goto out; | |
2868 | ||
2869 | if (next->vm_flags != vma->vm_flags) | |
2870 | goto out; | |
2871 | ||
1db43d3f LH |
2872 | if (start + size <= next->vm_end) |
2873 | break; | |
2874 | ||
763ecb03 | 2875 | prev = next; |
48f7df32 KS |
2876 | } |
2877 | ||
2878 | if (!next) | |
2879 | goto out; | |
c8d78c18 KS |
2880 | } |
2881 | ||
2882 | prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; | |
2883 | prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; | |
2884 | prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; | |
2885 | ||
2886 | flags &= MAP_NONBLOCK; | |
2887 | flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; | |
fce000b1 | 2888 | if (vma->vm_flags & VM_LOCKED) |
c8d78c18 | 2889 | flags |= MAP_LOCKED; |
48f7df32 | 2890 | |
c8d78c18 | 2891 | file = get_file(vma->vm_file); |
45e55300 | 2892 | ret = do_mmap(vma->vm_file, start, size, |
897ab3e0 | 2893 | prot, flags, pgoff, &populate, NULL); |
c8d78c18 KS |
2894 | fput(file); |
2895 | out: | |
d8ed45c5 | 2896 | mmap_write_unlock(mm); |
c8d78c18 KS |
2897 | if (populate) |
2898 | mm_populate(ret, populate); | |
2899 | if (!IS_ERR_VALUE(ret)) | |
2900 | ret = 0; | |
2901 | return ret; | |
2902 | } | |
2903 | ||
1da177e4 | 2904 | /* |
27b26701 LH |
2905 | * do_vma_munmap() - Unmap a full or partial vma. |
2906 | * @vmi: The vma iterator pointing at the vma | |
2907 | * @vma: The first vma to be munmapped | |
2908 | * @start: the start of the address to unmap | |
2909 | * @end: The end of the address to unmap | |
2e7ce7d3 | 2910 | * @uf: The userfaultfd list_head |
27b26701 | 2911 | * @downgrade: Attempt to downgrade or not |
2e7ce7d3 | 2912 | * |
27b26701 LH |
2913 | * Returns: 0 on success and not downgraded, 1 on success and downgraded. |
2914 | * unmaps a VMA mapping when the vma iterator is already in position. | |
2915 | * Does not handle alignment. | |
1da177e4 | 2916 | */ |
27b26701 LH |
2917 | int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, |
2918 | unsigned long start, unsigned long end, | |
2919 | struct list_head *uf, bool downgrade) | |
1da177e4 | 2920 | { |
2e7ce7d3 LH |
2921 | struct mm_struct *mm = vma->vm_mm; |
2922 | int ret; | |
3a459756 | 2923 | |
27b26701 LH |
2924 | arch_unmap(mm, start, end); |
2925 | ret = do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); | |
b50e195f | 2926 | validate_mm(mm); |
2e7ce7d3 LH |
2927 | return ret; |
2928 | } | |
1da177e4 | 2929 | |
2e7ce7d3 LH |
2930 | /* |
2931 | * do_brk_flags() - Increase the brk vma if the flags match. | |
92fed820 | 2932 | * @vmi: The vma iterator |
2e7ce7d3 LH |
2933 | * @addr: The start address |
2934 | * @len: The length of the increase | |
2935 | * @vma: The vma, | |
2936 | * @flags: The VMA Flags | |
2937 | * | |
2938 | * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags | |
2939 | * do not match then create a new anonymous VMA. Eventually we may be able to | |
2940 | * do some brk-specific accounting here. | |
2941 | */ | |
92fed820 | 2942 | static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, |
763ecb03 | 2943 | unsigned long addr, unsigned long len, unsigned long flags) |
2e7ce7d3 LH |
2944 | { |
2945 | struct mm_struct *mm = current->mm; | |
287051b1 | 2946 | struct vma_prepare vp; |
1da177e4 | 2947 | |
b50e195f | 2948 | validate_mm(mm); |
2e7ce7d3 LH |
2949 | /* |
2950 | * Check against address space limits by the changed size | |
2951 | * Note: This happens *after* clearing old mappings in some code paths. | |
2952 | */ | |
2953 | flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; | |
84638335 | 2954 | if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) |
1da177e4 LT |
2955 | return -ENOMEM; |
2956 | ||
2957 | if (mm->map_count > sysctl_max_map_count) | |
2958 | return -ENOMEM; | |
2959 | ||
191c5424 | 2960 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2961 | return -ENOMEM; |
2962 | ||
1da177e4 | 2963 | /* |
2e7ce7d3 LH |
2964 | * Expand the existing vma if possible; Note that singular lists do not |
2965 | * occur after forking, so the expand will only happen on new VMAs. | |
1da177e4 | 2966 | */ |
6c28ca64 LH |
2967 | if (vma && vma->vm_end == addr && !vma_policy(vma) && |
2968 | can_vma_merge_after(vma, flags, NULL, NULL, | |
2969 | addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { | |
92fed820 | 2970 | if (vma_iter_prealloc(vmi)) |
675eaca1 | 2971 | goto unacct_fail; |
28c5609f | 2972 | |
287051b1 LH |
2973 | init_vma_prep(&vp, vma); |
2974 | vma_prepare(&vp); | |
ccf1d78d | 2975 | vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); |
2e7ce7d3 | 2976 | vma->vm_end = addr + len; |
1c71222e | 2977 | vm_flags_set(vma, VM_SOFTDIRTY); |
92fed820 | 2978 | vma_iter_store(vmi, vma); |
2e7ce7d3 | 2979 | |
287051b1 | 2980 | vma_complete(&vp, vmi, mm); |
2e7ce7d3 LH |
2981 | khugepaged_enter_vma(vma, flags); |
2982 | goto out; | |
1da177e4 | 2983 | } |
2e7ce7d3 LH |
2984 | |
2985 | /* create a vma struct for an anonymous mapping */ | |
2986 | vma = vm_area_alloc(mm); | |
2987 | if (!vma) | |
675eaca1 | 2988 | goto unacct_fail; |
1da177e4 | 2989 | |
bfd40eaf | 2990 | vma_set_anonymous(vma); |
1da177e4 LT |
2991 | vma->vm_start = addr; |
2992 | vma->vm_end = addr + len; | |
2e7ce7d3 | 2993 | vma->vm_pgoff = addr >> PAGE_SHIFT; |
1c71222e | 2994 | vm_flags_init(vma, flags); |
3ed75eb8 | 2995 | vma->vm_page_prot = vm_get_page_prot(flags); |
92fed820 | 2996 | if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) |
2e7ce7d3 | 2997 | goto mas_store_fail; |
d4af56c5 | 2998 | |
2e7ce7d3 | 2999 | mm->map_count++; |
d7597f59 | 3000 | ksm_add_vma(vma); |
1da177e4 | 3001 | out: |
3af9e859 | 3002 | perf_event_mmap(vma); |
1da177e4 | 3003 | mm->total_vm += len >> PAGE_SHIFT; |
84638335 | 3004 | mm->data_vm += len >> PAGE_SHIFT; |
128557ff ML |
3005 | if (flags & VM_LOCKED) |
3006 | mm->locked_vm += (len >> PAGE_SHIFT); | |
1c71222e | 3007 | vm_flags_set(vma, VM_SOFTDIRTY); |
763ecb03 | 3008 | validate_mm(mm); |
5d22fc25 | 3009 | return 0; |
d4af56c5 | 3010 | |
2e7ce7d3 | 3011 | mas_store_fail: |
d4af56c5 | 3012 | vm_area_free(vma); |
675eaca1 | 3013 | unacct_fail: |
2e7ce7d3 LH |
3014 | vm_unacct_memory(len >> PAGE_SHIFT); |
3015 | return -ENOMEM; | |
1da177e4 LT |
3016 | } |
3017 | ||
bb177a73 | 3018 | int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) |
e4eb1ff6 LT |
3019 | { |
3020 | struct mm_struct *mm = current->mm; | |
2e7ce7d3 | 3021 | struct vm_area_struct *vma = NULL; |
bb177a73 | 3022 | unsigned long len; |
5d22fc25 | 3023 | int ret; |
128557ff | 3024 | bool populate; |
897ab3e0 | 3025 | LIST_HEAD(uf); |
92fed820 | 3026 | VMA_ITERATOR(vmi, mm, addr); |
e4eb1ff6 | 3027 | |
bb177a73 MH |
3028 | len = PAGE_ALIGN(request); |
3029 | if (len < request) | |
3030 | return -ENOMEM; | |
3031 | if (!len) | |
3032 | return 0; | |
3033 | ||
d8ed45c5 | 3034 | if (mmap_write_lock_killable(mm)) |
2d6c9282 MH |
3035 | return -EINTR; |
3036 | ||
2e7ce7d3 LH |
3037 | /* Until we need other flags, refuse anything except VM_EXEC. */ |
3038 | if ((flags & (~VM_EXEC)) != 0) | |
3039 | return -EINVAL; | |
3040 | ||
3041 | ret = check_brk_limits(addr, len); | |
3042 | if (ret) | |
3043 | goto limits_failed; | |
3044 | ||
183654ce | 3045 | ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); |
2e7ce7d3 LH |
3046 | if (ret) |
3047 | goto munmap_failed; | |
3048 | ||
92fed820 LH |
3049 | vma = vma_prev(&vmi); |
3050 | ret = do_brk_flags(&vmi, vma, addr, len, flags); | |
128557ff | 3051 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
d8ed45c5 | 3052 | mmap_write_unlock(mm); |
897ab3e0 | 3053 | userfaultfd_unmap_complete(mm, &uf); |
5d22fc25 | 3054 | if (populate && !ret) |
128557ff | 3055 | mm_populate(addr, len); |
e4eb1ff6 | 3056 | return ret; |
2e7ce7d3 LH |
3057 | |
3058 | munmap_failed: | |
3059 | limits_failed: | |
3060 | mmap_write_unlock(mm); | |
3061 | return ret; | |
e4eb1ff6 | 3062 | } |
16e72e9b DV |
3063 | EXPORT_SYMBOL(vm_brk_flags); |
3064 | ||
3065 | int vm_brk(unsigned long addr, unsigned long len) | |
3066 | { | |
3067 | return vm_brk_flags(addr, len, 0); | |
3068 | } | |
e4eb1ff6 | 3069 | EXPORT_SYMBOL(vm_brk); |
1da177e4 LT |
3070 | |
3071 | /* Release all mmaps. */ | |
3072 | void exit_mmap(struct mm_struct *mm) | |
3073 | { | |
d16dfc55 | 3074 | struct mmu_gather tlb; |
ba470de4 | 3075 | struct vm_area_struct *vma; |
1da177e4 | 3076 | unsigned long nr_accounted = 0; |
763ecb03 LH |
3077 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
3078 | int count = 0; | |
1da177e4 | 3079 | |
d6dd61c8 | 3080 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 3081 | mmu_notifier_release(mm); |
d6dd61c8 | 3082 | |
bf3980c8 | 3083 | mmap_read_lock(mm); |
9480c53e JF |
3084 | arch_exit_mmap(mm); |
3085 | ||
763ecb03 | 3086 | vma = mas_find(&mas, ULONG_MAX); |
64591e86 SB |
3087 | if (!vma) { |
3088 | /* Can happen if dup_mmap() received an OOM */ | |
bf3980c8 | 3089 | mmap_read_unlock(mm); |
9480c53e | 3090 | return; |
64591e86 | 3091 | } |
9480c53e | 3092 | |
1da177e4 | 3093 | lru_add_drain(); |
1da177e4 | 3094 | flush_cache_mm(mm); |
d8b45053 | 3095 | tlb_gather_mmu_fullmm(&tlb, mm); |
901608d9 | 3096 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
763ecb03 | 3097 | /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ |
68f48381 | 3098 | unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX, false); |
bf3980c8 SB |
3099 | mmap_read_unlock(mm); |
3100 | ||
3101 | /* | |
3102 | * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper | |
b3541d91 | 3103 | * because the memory has been already freed. |
bf3980c8 SB |
3104 | */ |
3105 | set_bit(MMF_OOM_SKIP, &mm->flags); | |
3106 | mmap_write_lock(mm); | |
3dd44325 | 3107 | mt_clear_in_rcu(&mm->mm_mt); |
763ecb03 | 3108 | free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS, |
98e51a22 | 3109 | USER_PGTABLES_CEILING, true); |
ae8eba8b | 3110 | tlb_finish_mmu(&tlb); |
1da177e4 | 3111 | |
763ecb03 LH |
3112 | /* |
3113 | * Walk the list again, actually closing and freeing it, with preemption | |
3114 | * enabled, without holding any MM locks besides the unreachable | |
3115 | * mmap_write_lock. | |
3116 | */ | |
3117 | do { | |
4f74d2c8 LT |
3118 | if (vma->vm_flags & VM_ACCOUNT) |
3119 | nr_accounted += vma_pages(vma); | |
0d2ebf9c | 3120 | remove_vma(vma, true); |
763ecb03 | 3121 | count++; |
0a3b3c25 | 3122 | cond_resched(); |
763ecb03 LH |
3123 | } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); |
3124 | ||
3125 | BUG_ON(count != mm->map_count); | |
d4af56c5 LH |
3126 | |
3127 | trace_exit_mmap(mm); | |
3128 | __mt_destroy(&mm->mm_mt); | |
64591e86 | 3129 | mmap_write_unlock(mm); |
4f74d2c8 | 3130 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
3131 | } |
3132 | ||
3133 | /* Insert vm structure into process list sorted by address | |
3134 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
c8c06efa | 3135 | * then i_mmap_rwsem is taken here. |
1da177e4 | 3136 | */ |
6597d783 | 3137 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 3138 | { |
d4af56c5 | 3139 | unsigned long charged = vma_pages(vma); |
1da177e4 | 3140 | |
d4af56c5 | 3141 | |
d0601a50 | 3142 | if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) |
c9d13f5f | 3143 | return -ENOMEM; |
d4af56c5 | 3144 | |
c9d13f5f | 3145 | if ((vma->vm_flags & VM_ACCOUNT) && |
d4af56c5 | 3146 | security_vm_enough_memory_mm(mm, charged)) |
c9d13f5f CG |
3147 | return -ENOMEM; |
3148 | ||
1da177e4 LT |
3149 | /* |
3150 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
3151 | * until its first write fault, when page's anon_vma and index | |
3152 | * are set. But now set the vm_pgoff it will almost certainly | |
3153 | * end up with (unless mremap moves it elsewhere before that | |
3154 | * first wfault), so /proc/pid/maps tells a consistent story. | |
3155 | * | |
3156 | * By setting it to reflect the virtual start address of the | |
3157 | * vma, merges and splits can happen in a seamless way, just | |
3158 | * using the existing file pgoff checks and manipulations. | |
8332326e | 3159 | * Similarly in do_mmap and in do_brk_flags. |
1da177e4 | 3160 | */ |
8a9cc3b5 | 3161 | if (vma_is_anonymous(vma)) { |
1da177e4 LT |
3162 | BUG_ON(vma->anon_vma); |
3163 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
3164 | } | |
2b144498 | 3165 | |
763ecb03 | 3166 | if (vma_link(mm, vma)) { |
d4af56c5 LH |
3167 | vm_unacct_memory(charged); |
3168 | return -ENOMEM; | |
3169 | } | |
3170 | ||
1da177e4 LT |
3171 | return 0; |
3172 | } | |
3173 | ||
3174 | /* | |
3175 | * Copy the vma structure to a new location in the same mm, | |
3176 | * prior to moving page table entries, to effect an mremap move. | |
3177 | */ | |
3178 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
3179 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
3180 | bool *need_rmap_locks) | |
1da177e4 LT |
3181 | { |
3182 | struct vm_area_struct *vma = *vmap; | |
3183 | unsigned long vma_start = vma->vm_start; | |
3184 | struct mm_struct *mm = vma->vm_mm; | |
3185 | struct vm_area_struct *new_vma, *prev; | |
948f017b | 3186 | bool faulted_in_anon_vma = true; |
076f16bf | 3187 | VMA_ITERATOR(vmi, mm, addr); |
1da177e4 | 3188 | |
b50e195f | 3189 | validate_mm(mm); |
1da177e4 LT |
3190 | /* |
3191 | * If anonymous vma has not yet been faulted, update new pgoff | |
3192 | * to match new location, to increase its chance of merging. | |
3193 | */ | |
ce75799b | 3194 | if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { |
1da177e4 | 3195 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
3196 | faulted_in_anon_vma = false; |
3197 | } | |
1da177e4 | 3198 | |
763ecb03 LH |
3199 | new_vma = find_vma_prev(mm, addr, &prev); |
3200 | if (new_vma && new_vma->vm_start < addr + len) | |
6597d783 | 3201 | return NULL; /* should never get here */ |
524e00b3 | 3202 | |
9760ebff | 3203 | new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, |
19a809af | 3204 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
5c26f6ac | 3205 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); |
1da177e4 LT |
3206 | if (new_vma) { |
3207 | /* | |
3208 | * Source vma may have been merged into new_vma | |
3209 | */ | |
948f017b AA |
3210 | if (unlikely(vma_start >= new_vma->vm_start && |
3211 | vma_start < new_vma->vm_end)) { | |
3212 | /* | |
3213 | * The only way we can get a vma_merge with | |
3214 | * self during an mremap is if the vma hasn't | |
3215 | * been faulted in yet and we were allowed to | |
3216 | * reset the dst vma->vm_pgoff to the | |
3217 | * destination address of the mremap to allow | |
3218 | * the merge to happen. mremap must change the | |
3219 | * vm_pgoff linearity between src and dst vmas | |
3220 | * (in turn preventing a vma_merge) to be | |
3221 | * safe. It is only safe to keep the vm_pgoff | |
3222 | * linear if there are no pages mapped yet. | |
3223 | */ | |
81d1b09c | 3224 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
38a76013 | 3225 | *vmap = vma = new_vma; |
108d6642 | 3226 | } |
38a76013 | 3227 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 3228 | } else { |
3928d4f5 | 3229 | new_vma = vm_area_dup(vma); |
e3975891 CG |
3230 | if (!new_vma) |
3231 | goto out; | |
e3975891 CG |
3232 | new_vma->vm_start = addr; |
3233 | new_vma->vm_end = addr + len; | |
3234 | new_vma->vm_pgoff = pgoff; | |
3235 | if (vma_dup_policy(vma, new_vma)) | |
3236 | goto out_free_vma; | |
e3975891 CG |
3237 | if (anon_vma_clone(new_vma, vma)) |
3238 | goto out_free_mempol; | |
3239 | if (new_vma->vm_file) | |
3240 | get_file(new_vma->vm_file); | |
3241 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
3242 | new_vma->vm_ops->open(new_vma); | |
d6ac235d | 3243 | vma_start_write(new_vma); |
763ecb03 | 3244 | if (vma_link(mm, new_vma)) |
524e00b3 | 3245 | goto out_vma_link; |
e3975891 | 3246 | *need_rmap_locks = false; |
1da177e4 | 3247 | } |
b50e195f | 3248 | validate_mm(mm); |
1da177e4 | 3249 | return new_vma; |
5beb4930 | 3250 | |
524e00b3 LH |
3251 | out_vma_link: |
3252 | if (new_vma->vm_ops && new_vma->vm_ops->close) | |
3253 | new_vma->vm_ops->close(new_vma); | |
92b73996 LH |
3254 | |
3255 | if (new_vma->vm_file) | |
3256 | fput(new_vma->vm_file); | |
3257 | ||
3258 | unlink_anon_vmas(new_vma); | |
e3975891 | 3259 | out_free_mempol: |
ef0855d3 | 3260 | mpol_put(vma_policy(new_vma)); |
e3975891 | 3261 | out_free_vma: |
3928d4f5 | 3262 | vm_area_free(new_vma); |
e3975891 | 3263 | out: |
b50e195f | 3264 | validate_mm(mm); |
5beb4930 | 3265 | return NULL; |
1da177e4 | 3266 | } |
119f657c | 3267 | |
3268 | /* | |
3269 | * Return true if the calling process may expand its vm space by the passed | |
3270 | * number of pages | |
3271 | */ | |
84638335 | 3272 | bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) |
119f657c | 3273 | { |
84638335 KK |
3274 | if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) |
3275 | return false; | |
119f657c | 3276 | |
d977d56c KK |
3277 | if (is_data_mapping(flags) && |
3278 | mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { | |
f4fcd558 KK |
3279 | /* Workaround for Valgrind */ |
3280 | if (rlimit(RLIMIT_DATA) == 0 && | |
3281 | mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) | |
3282 | return true; | |
57a7702b DW |
3283 | |
3284 | pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", | |
3285 | current->comm, current->pid, | |
3286 | (mm->data_vm + npages) << PAGE_SHIFT, | |
3287 | rlimit(RLIMIT_DATA), | |
3288 | ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); | |
3289 | ||
3290 | if (!ignore_rlimit_data) | |
d977d56c KK |
3291 | return false; |
3292 | } | |
119f657c | 3293 | |
84638335 KK |
3294 | return true; |
3295 | } | |
3296 | ||
3297 | void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) | |
3298 | { | |
7866076b | 3299 | WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); |
84638335 | 3300 | |
d977d56c | 3301 | if (is_exec_mapping(flags)) |
84638335 | 3302 | mm->exec_vm += npages; |
d977d56c | 3303 | else if (is_stack_mapping(flags)) |
84638335 | 3304 | mm->stack_vm += npages; |
d977d56c | 3305 | else if (is_data_mapping(flags)) |
84638335 | 3306 | mm->data_vm += npages; |
119f657c | 3307 | } |
fa5dc22f | 3308 | |
b3ec9f33 | 3309 | static vm_fault_t special_mapping_fault(struct vm_fault *vmf); |
a62c34bd AL |
3310 | |
3311 | /* | |
3312 | * Having a close hook prevents vma merging regardless of flags. | |
3313 | */ | |
3314 | static void special_mapping_close(struct vm_area_struct *vma) | |
3315 | { | |
3316 | } | |
3317 | ||
3318 | static const char *special_mapping_name(struct vm_area_struct *vma) | |
3319 | { | |
3320 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; | |
3321 | } | |
3322 | ||
14d07113 | 3323 | static int special_mapping_mremap(struct vm_area_struct *new_vma) |
b059a453 DS |
3324 | { |
3325 | struct vm_special_mapping *sm = new_vma->vm_private_data; | |
3326 | ||
280e87e9 DS |
3327 | if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) |
3328 | return -EFAULT; | |
3329 | ||
b059a453 DS |
3330 | if (sm->mremap) |
3331 | return sm->mremap(sm, new_vma); | |
280e87e9 | 3332 | |
b059a453 DS |
3333 | return 0; |
3334 | } | |
3335 | ||
871402e0 DS |
3336 | static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) |
3337 | { | |
3338 | /* | |
3339 | * Forbid splitting special mappings - kernel has expectations over | |
3340 | * the number of pages in mapping. Together with VM_DONTEXPAND | |
3341 | * the size of vma should stay the same over the special mapping's | |
3342 | * lifetime. | |
3343 | */ | |
3344 | return -EINVAL; | |
3345 | } | |
3346 | ||
a62c34bd AL |
3347 | static const struct vm_operations_struct special_mapping_vmops = { |
3348 | .close = special_mapping_close, | |
3349 | .fault = special_mapping_fault, | |
b059a453 | 3350 | .mremap = special_mapping_mremap, |
a62c34bd | 3351 | .name = special_mapping_name, |
af34ebeb DS |
3352 | /* vDSO code relies that VVAR can't be accessed remotely */ |
3353 | .access = NULL, | |
871402e0 | 3354 | .may_split = special_mapping_split, |
a62c34bd AL |
3355 | }; |
3356 | ||
3357 | static const struct vm_operations_struct legacy_special_mapping_vmops = { | |
3358 | .close = special_mapping_close, | |
3359 | .fault = special_mapping_fault, | |
3360 | }; | |
fa5dc22f | 3361 | |
b3ec9f33 | 3362 | static vm_fault_t special_mapping_fault(struct vm_fault *vmf) |
fa5dc22f | 3363 | { |
11bac800 | 3364 | struct vm_area_struct *vma = vmf->vma; |
b1d0e4f5 | 3365 | pgoff_t pgoff; |
fa5dc22f RM |
3366 | struct page **pages; |
3367 | ||
f872f540 | 3368 | if (vma->vm_ops == &legacy_special_mapping_vmops) { |
a62c34bd | 3369 | pages = vma->vm_private_data; |
f872f540 AL |
3370 | } else { |
3371 | struct vm_special_mapping *sm = vma->vm_private_data; | |
3372 | ||
3373 | if (sm->fault) | |
11bac800 | 3374 | return sm->fault(sm, vmf->vma, vmf); |
f872f540 AL |
3375 | |
3376 | pages = sm->pages; | |
3377 | } | |
a62c34bd | 3378 | |
8a9cc3b5 | 3379 | for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) |
b1d0e4f5 | 3380 | pgoff--; |
fa5dc22f RM |
3381 | |
3382 | if (*pages) { | |
3383 | struct page *page = *pages; | |
3384 | get_page(page); | |
b1d0e4f5 NP |
3385 | vmf->page = page; |
3386 | return 0; | |
fa5dc22f RM |
3387 | } |
3388 | ||
b1d0e4f5 | 3389 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
3390 | } |
3391 | ||
a62c34bd AL |
3392 | static struct vm_area_struct *__install_special_mapping( |
3393 | struct mm_struct *mm, | |
3394 | unsigned long addr, unsigned long len, | |
27f28b97 CG |
3395 | unsigned long vm_flags, void *priv, |
3396 | const struct vm_operations_struct *ops) | |
fa5dc22f | 3397 | { |
462e635e | 3398 | int ret; |
fa5dc22f RM |
3399 | struct vm_area_struct *vma; |
3400 | ||
b50e195f | 3401 | validate_mm(mm); |
490fc053 | 3402 | vma = vm_area_alloc(mm); |
fa5dc22f | 3403 | if (unlikely(vma == NULL)) |
3935ed6a | 3404 | return ERR_PTR(-ENOMEM); |
fa5dc22f | 3405 | |
fa5dc22f RM |
3406 | vma->vm_start = addr; |
3407 | vma->vm_end = addr + len; | |
3408 | ||
e430a95a SB |
3409 | vm_flags_init(vma, (vm_flags | mm->def_flags | |
3410 | VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); | |
3ed75eb8 | 3411 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f | 3412 | |
a62c34bd AL |
3413 | vma->vm_ops = ops; |
3414 | vma->vm_private_data = priv; | |
fa5dc22f | 3415 | |
462e635e TO |
3416 | ret = insert_vm_struct(mm, vma); |
3417 | if (ret) | |
3418 | goto out; | |
fa5dc22f | 3419 | |
84638335 | 3420 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); |
fa5dc22f | 3421 | |
cdd6c482 | 3422 | perf_event_mmap(vma); |
089dd79d | 3423 | |
b50e195f | 3424 | validate_mm(mm); |
3935ed6a | 3425 | return vma; |
462e635e TO |
3426 | |
3427 | out: | |
3928d4f5 | 3428 | vm_area_free(vma); |
b50e195f | 3429 | validate_mm(mm); |
3935ed6a SS |
3430 | return ERR_PTR(ret); |
3431 | } | |
3432 | ||
2eefd878 DS |
3433 | bool vma_is_special_mapping(const struct vm_area_struct *vma, |
3434 | const struct vm_special_mapping *sm) | |
3435 | { | |
3436 | return vma->vm_private_data == sm && | |
3437 | (vma->vm_ops == &special_mapping_vmops || | |
3438 | vma->vm_ops == &legacy_special_mapping_vmops); | |
3439 | } | |
3440 | ||
a62c34bd | 3441 | /* |
c1e8d7c6 | 3442 | * Called with mm->mmap_lock held for writing. |
a62c34bd AL |
3443 | * Insert a new vma covering the given region, with the given flags. |
3444 | * Its pages are supplied by the given array of struct page *. | |
3445 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
3446 | * The region past the last page supplied will always produce SIGBUS. | |
3447 | * The array pointer and the pages it points to are assumed to stay alive | |
3448 | * for as long as this mapping might exist. | |
3449 | */ | |
3450 | struct vm_area_struct *_install_special_mapping( | |
3451 | struct mm_struct *mm, | |
3452 | unsigned long addr, unsigned long len, | |
3453 | unsigned long vm_flags, const struct vm_special_mapping *spec) | |
3454 | { | |
27f28b97 CG |
3455 | return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, |
3456 | &special_mapping_vmops); | |
a62c34bd AL |
3457 | } |
3458 | ||
3935ed6a SS |
3459 | int install_special_mapping(struct mm_struct *mm, |
3460 | unsigned long addr, unsigned long len, | |
3461 | unsigned long vm_flags, struct page **pages) | |
3462 | { | |
a62c34bd | 3463 | struct vm_area_struct *vma = __install_special_mapping( |
27f28b97 CG |
3464 | mm, addr, len, vm_flags, (void *)pages, |
3465 | &legacy_special_mapping_vmops); | |
3935ed6a | 3466 | |
14bd5b45 | 3467 | return PTR_ERR_OR_ZERO(vma); |
fa5dc22f | 3468 | } |
7906d00c AA |
3469 | |
3470 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3471 | ||
454ed842 | 3472 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3473 | { |
f808c13f | 3474 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { |
7906d00c AA |
3475 | /* |
3476 | * The LSB of head.next can't change from under us | |
3477 | * because we hold the mm_all_locks_mutex. | |
3478 | */ | |
da1c55f1 | 3479 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); |
7906d00c AA |
3480 | /* |
3481 | * We can safely modify head.next after taking the | |
5a505085 | 3482 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3483 | * the same anon_vma we won't take it again. |
3484 | * | |
3485 | * No need of atomic instructions here, head.next | |
3486 | * can't change from under us thanks to the | |
5a505085 | 3487 | * anon_vma->root->rwsem. |
7906d00c AA |
3488 | */ |
3489 | if (__test_and_set_bit(0, (unsigned long *) | |
f808c13f | 3490 | &anon_vma->root->rb_root.rb_root.rb_node)) |
7906d00c AA |
3491 | BUG(); |
3492 | } | |
3493 | } | |
3494 | ||
454ed842 | 3495 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3496 | { |
3497 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3498 | /* | |
3499 | * AS_MM_ALL_LOCKS can't change from under us because | |
3500 | * we hold the mm_all_locks_mutex. | |
3501 | * | |
3502 | * Operations on ->flags have to be atomic because | |
3503 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3504 | * mm_all_locks_mutex, there may be other cpus | |
3505 | * changing other bitflags in parallel to us. | |
3506 | */ | |
3507 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3508 | BUG(); | |
da1c55f1 | 3509 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); |
7906d00c AA |
3510 | } |
3511 | } | |
3512 | ||
3513 | /* | |
3514 | * This operation locks against the VM for all pte/vma/mm related | |
3515 | * operations that could ever happen on a certain mm. This includes | |
3516 | * vmtruncate, try_to_unmap, and all page faults. | |
3517 | * | |
c1e8d7c6 | 3518 | * The caller must take the mmap_lock in write mode before calling |
7906d00c | 3519 | * mm_take_all_locks(). The caller isn't allowed to release the |
c1e8d7c6 | 3520 | * mmap_lock until mm_drop_all_locks() returns. |
7906d00c | 3521 | * |
c1e8d7c6 | 3522 | * mmap_lock in write mode is required in order to block all operations |
7906d00c | 3523 | * that could modify pagetables and free pages without need of |
27ba0644 | 3524 | * altering the vma layout. It's also needed in write mode to avoid new |
7906d00c AA |
3525 | * anon_vmas to be associated with existing vmas. |
3526 | * | |
3527 | * A single task can't take more than one mm_take_all_locks() in a row | |
3528 | * or it would deadlock. | |
3529 | * | |
bf181b9f | 3530 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3531 | * mapping->flags avoid to take the same lock twice, if more than one |
3532 | * vma in this mm is backed by the same anon_vma or address_space. | |
3533 | * | |
88f306b6 KS |
3534 | * We take locks in following order, accordingly to comment at beginning |
3535 | * of mm/rmap.c: | |
3536 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for | |
3537 | * hugetlb mapping); | |
eeff9a5d | 3538 | * - all vmas marked locked |
88f306b6 KS |
3539 | * - all i_mmap_rwsem locks; |
3540 | * - all anon_vma->rwseml | |
3541 | * | |
3542 | * We can take all locks within these types randomly because the VM code | |
3543 | * doesn't nest them and we protected from parallel mm_take_all_locks() by | |
3544 | * mm_all_locks_mutex. | |
7906d00c AA |
3545 | * |
3546 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3547 | * that may have to take thousand of locks. | |
3548 | * | |
3549 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3550 | */ | |
3551 | int mm_take_all_locks(struct mm_struct *mm) | |
3552 | { | |
3553 | struct vm_area_struct *vma; | |
5beb4930 | 3554 | struct anon_vma_chain *avc; |
763ecb03 | 3555 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
7906d00c | 3556 | |
325bca1f | 3557 | mmap_assert_write_locked(mm); |
7906d00c AA |
3558 | |
3559 | mutex_lock(&mm_all_locks_mutex); | |
3560 | ||
eeff9a5d SB |
3561 | mas_for_each(&mas, vma, ULONG_MAX) { |
3562 | if (signal_pending(current)) | |
3563 | goto out_unlock; | |
3564 | vma_start_write(vma); | |
3565 | } | |
3566 | ||
3567 | mas_set(&mas, 0); | |
763ecb03 | 3568 | mas_for_each(&mas, vma, ULONG_MAX) { |
7906d00c AA |
3569 | if (signal_pending(current)) |
3570 | goto out_unlock; | |
88f306b6 KS |
3571 | if (vma->vm_file && vma->vm_file->f_mapping && |
3572 | is_vm_hugetlb_page(vma)) | |
3573 | vm_lock_mapping(mm, vma->vm_file->f_mapping); | |
3574 | } | |
3575 | ||
763ecb03 LH |
3576 | mas_set(&mas, 0); |
3577 | mas_for_each(&mas, vma, ULONG_MAX) { | |
88f306b6 KS |
3578 | if (signal_pending(current)) |
3579 | goto out_unlock; | |
3580 | if (vma->vm_file && vma->vm_file->f_mapping && | |
3581 | !is_vm_hugetlb_page(vma)) | |
454ed842 | 3582 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3583 | } |
7cd5a02f | 3584 | |
763ecb03 LH |
3585 | mas_set(&mas, 0); |
3586 | mas_for_each(&mas, vma, ULONG_MAX) { | |
7cd5a02f PZ |
3587 | if (signal_pending(current)) |
3588 | goto out_unlock; | |
3589 | if (vma->anon_vma) | |
5beb4930 RR |
3590 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3591 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3592 | } |
7cd5a02f | 3593 | |
584cff54 | 3594 | return 0; |
7906d00c AA |
3595 | |
3596 | out_unlock: | |
584cff54 KC |
3597 | mm_drop_all_locks(mm); |
3598 | return -EINTR; | |
7906d00c AA |
3599 | } |
3600 | ||
3601 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3602 | { | |
f808c13f | 3603 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { |
7906d00c AA |
3604 | /* |
3605 | * The LSB of head.next can't change to 0 from under | |
3606 | * us because we hold the mm_all_locks_mutex. | |
3607 | * | |
3608 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3609 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3610 | * never see our bitflag. |
3611 | * | |
3612 | * No need of atomic instructions here, head.next | |
3613 | * can't change from under us until we release the | |
5a505085 | 3614 | * anon_vma->root->rwsem. |
7906d00c AA |
3615 | */ |
3616 | if (!__test_and_clear_bit(0, (unsigned long *) | |
f808c13f | 3617 | &anon_vma->root->rb_root.rb_root.rb_node)) |
7906d00c | 3618 | BUG(); |
08b52706 | 3619 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3620 | } |
3621 | } | |
3622 | ||
3623 | static void vm_unlock_mapping(struct address_space *mapping) | |
3624 | { | |
3625 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3626 | /* | |
3627 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3628 | * because we hold the mm_all_locks_mutex. | |
3629 | */ | |
83cde9e8 | 3630 | i_mmap_unlock_write(mapping); |
7906d00c AA |
3631 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3632 | &mapping->flags)) | |
3633 | BUG(); | |
3634 | } | |
3635 | } | |
3636 | ||
3637 | /* | |
c1e8d7c6 | 3638 | * The mmap_lock cannot be released by the caller until |
7906d00c AA |
3639 | * mm_drop_all_locks() returns. |
3640 | */ | |
3641 | void mm_drop_all_locks(struct mm_struct *mm) | |
3642 | { | |
3643 | struct vm_area_struct *vma; | |
5beb4930 | 3644 | struct anon_vma_chain *avc; |
763ecb03 | 3645 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
7906d00c | 3646 | |
325bca1f | 3647 | mmap_assert_write_locked(mm); |
7906d00c AA |
3648 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); |
3649 | ||
763ecb03 | 3650 | mas_for_each(&mas, vma, ULONG_MAX) { |
7906d00c | 3651 | if (vma->anon_vma) |
5beb4930 RR |
3652 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3653 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3654 | if (vma->vm_file && vma->vm_file->f_mapping) |
3655 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3656 | } | |
eeff9a5d | 3657 | vma_end_write_all(mm); |
7906d00c AA |
3658 | |
3659 | mutex_unlock(&mm_all_locks_mutex); | |
3660 | } | |
8feae131 DH |
3661 | |
3662 | /* | |
3edf41d8 | 3663 | * initialise the percpu counter for VM |
8feae131 DH |
3664 | */ |
3665 | void __init mmap_init(void) | |
3666 | { | |
00a62ce9 KM |
3667 | int ret; |
3668 | ||
908c7f19 | 3669 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 3670 | VM_BUG_ON(ret); |
8feae131 | 3671 | } |
c9b1d098 AS |
3672 | |
3673 | /* | |
3674 | * Initialise sysctl_user_reserve_kbytes. | |
3675 | * | |
3676 | * This is intended to prevent a user from starting a single memory hogging | |
3677 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3678 | * mode. | |
3679 | * | |
3680 | * The default value is min(3% of free memory, 128MB) | |
3681 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3682 | */ | |
1640879a | 3683 | static int init_user_reserve(void) |
c9b1d098 AS |
3684 | { |
3685 | unsigned long free_kbytes; | |
3686 | ||
c41f012a | 3687 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
c9b1d098 AS |
3688 | |
3689 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3690 | return 0; | |
3691 | } | |
a64fb3cd | 3692 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
3693 | |
3694 | /* | |
3695 | * Initialise sysctl_admin_reserve_kbytes. | |
3696 | * | |
3697 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3698 | * to log in and kill a memory hogging process. | |
3699 | * | |
3700 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3701 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3702 | * only reserve 3% of free pages by default. | |
3703 | */ | |
1640879a | 3704 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3705 | { |
3706 | unsigned long free_kbytes; | |
3707 | ||
c41f012a | 3708 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
4eeab4f5 AS |
3709 | |
3710 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3711 | return 0; | |
3712 | } | |
a64fb3cd | 3713 | subsys_initcall(init_admin_reserve); |
1640879a AS |
3714 | |
3715 | /* | |
3716 | * Reinititalise user and admin reserves if memory is added or removed. | |
3717 | * | |
3718 | * The default user reserve max is 128MB, and the default max for the | |
3719 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3720 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3721 | * and tools like top. It may make sense to increase or even disable the | |
3722 | * reserve depending on the existence of swap or variations in the recovery | |
3723 | * tools. So, the admin may have changed them. | |
3724 | * | |
3725 | * If memory is added and the reserves have been eliminated or increased above | |
3726 | * the default max, then we'll trust the admin. | |
3727 | * | |
3728 | * If memory is removed and there isn't enough free memory, then we | |
3729 | * need to reset the reserves. | |
3730 | * | |
3731 | * Otherwise keep the reserve set by the admin. | |
3732 | */ | |
3733 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3734 | unsigned long action, void *data) | |
3735 | { | |
3736 | unsigned long tmp, free_kbytes; | |
3737 | ||
3738 | switch (action) { | |
3739 | case MEM_ONLINE: | |
3740 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3741 | tmp = sysctl_user_reserve_kbytes; | |
3742 | if (0 < tmp && tmp < (1UL << 17)) | |
3743 | init_user_reserve(); | |
3744 | ||
3745 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3746 | tmp = sysctl_admin_reserve_kbytes; | |
3747 | if (0 < tmp && tmp < (1UL << 13)) | |
3748 | init_admin_reserve(); | |
3749 | ||
3750 | break; | |
3751 | case MEM_OFFLINE: | |
c41f012a | 3752 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
1640879a AS |
3753 | |
3754 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3755 | init_user_reserve(); | |
3756 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3757 | sysctl_user_reserve_kbytes); | |
3758 | } | |
3759 | ||
3760 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3761 | init_admin_reserve(); | |
3762 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3763 | sysctl_admin_reserve_kbytes); | |
3764 | } | |
3765 | break; | |
3766 | default: | |
3767 | break; | |
3768 | } | |
3769 | return NOTIFY_OK; | |
3770 | } | |
3771 | ||
1640879a AS |
3772 | static int __meminit init_reserve_notifier(void) |
3773 | { | |
1eeaa4fd | 3774 | if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) |
b1de0d13 | 3775 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
1640879a AS |
3776 | |
3777 | return 0; | |
3778 | } | |
a64fb3cd | 3779 | subsys_initcall(init_reserve_notifier); |