]>
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 | ||
b0cc5e89 | 185 | return mlock_future_ok(current->mm, current->mm->def_flags, len) |
3c54a298 | 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; |
408579cd | 196 | bool populate = false; |
897ab3e0 | 197 | LIST_HEAD(uf); |
92fed820 | 198 | struct vma_iterator vmi; |
1da177e4 | 199 | |
d8ed45c5 | 200 | if (mmap_write_lock_killable(mm)) |
dc0ef0df | 201 | return -EINTR; |
1da177e4 | 202 | |
9bc8039e YS |
203 | origbrk = mm->brk; |
204 | ||
a5b4592c | 205 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
206 | /* |
207 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
208 | * randomize_va_space to 2, which will still cause mm->start_brk | |
209 | * to be arbitrarily shifted | |
210 | */ | |
4471a675 | 211 | if (current->brk_randomized) |
5520e894 JK |
212 | min_brk = mm->start_brk; |
213 | else | |
214 | min_brk = mm->end_data; | |
a5b4592c JK |
215 | #else |
216 | min_brk = mm->start_brk; | |
217 | #endif | |
218 | if (brk < min_brk) | |
1da177e4 | 219 | goto out; |
1e624196 RG |
220 | |
221 | /* | |
222 | * Check against rlimit here. If this check is done later after the test | |
223 | * of oldbrk with newbrk then it can escape the test and let the data | |
224 | * segment grow beyond its set limit the in case where the limit is | |
225 | * not page aligned -Ram Gupta | |
226 | */ | |
8764b338 CG |
227 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
228 | mm->end_data, mm->start_data)) | |
1e624196 RG |
229 | goto out; |
230 | ||
1da177e4 LT |
231 | newbrk = PAGE_ALIGN(brk); |
232 | oldbrk = PAGE_ALIGN(mm->brk); | |
9bc8039e YS |
233 | if (oldbrk == newbrk) { |
234 | mm->brk = brk; | |
235 | goto success; | |
236 | } | |
1da177e4 | 237 | |
408579cd | 238 | /* Always allow shrinking brk. */ |
1da177e4 | 239 | if (brk <= mm->brk) { |
2e7ce7d3 | 240 | /* Search one past newbrk */ |
92fed820 LH |
241 | vma_iter_init(&vmi, mm, newbrk); |
242 | brkvma = vma_find(&vmi, oldbrk); | |
f5ad5083 | 243 | if (!brkvma || brkvma->vm_start >= oldbrk) |
2e7ce7d3 | 244 | goto out; /* mapping intersects with an existing non-brk vma. */ |
9bc8039e | 245 | /* |
2e7ce7d3 | 246 | * mm->brk must be protected by write mmap_lock. |
408579cd | 247 | * do_vma_munmap() will drop the lock on success, so update it |
27b26701 | 248 | * before calling do_vma_munmap(). |
9bc8039e YS |
249 | */ |
250 | mm->brk = brk; | |
408579cd LH |
251 | if (do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true)) |
252 | goto out; | |
253 | ||
254 | goto success_unlocked; | |
1da177e4 LT |
255 | } |
256 | ||
2e7ce7d3 LH |
257 | if (check_brk_limits(oldbrk, newbrk - oldbrk)) |
258 | goto out; | |
259 | ||
260 | /* | |
261 | * Only check if the next VMA is within the stack_guard_gap of the | |
262 | * expansion area | |
263 | */ | |
92fed820 LH |
264 | vma_iter_init(&vmi, mm, oldbrk); |
265 | next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap); | |
1be7107f | 266 | if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) |
1da177e4 LT |
267 | goto out; |
268 | ||
92fed820 | 269 | brkvma = vma_prev_limit(&vmi, mm->start_brk); |
1da177e4 | 270 | /* Ok, looks good - let it rip. */ |
92fed820 | 271 | if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0) |
1da177e4 | 272 | goto out; |
2e7ce7d3 | 273 | |
1da177e4 | 274 | mm->brk = brk; |
408579cd LH |
275 | if (mm->def_flags & VM_LOCKED) |
276 | populate = true; | |
9bc8039e YS |
277 | |
278 | success: | |
408579cd LH |
279 | mmap_write_unlock(mm); |
280 | success_unlocked: | |
897ab3e0 | 281 | userfaultfd_unmap_complete(mm, &uf); |
128557ff ML |
282 | if (populate) |
283 | mm_populate(oldbrk, newbrk - oldbrk); | |
284 | return brk; | |
285 | ||
1da177e4 | 286 | out: |
408579cd | 287 | mm->brk = origbrk; |
d8ed45c5 | 288 | mmap_write_unlock(mm); |
b7204006 | 289 | return origbrk; |
1da177e4 LT |
290 | } |
291 | ||
d4af56c5 | 292 | #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) |
eafd4dc4 | 293 | static void validate_mm(struct mm_struct *mm) |
1da177e4 LT |
294 | { |
295 | int bug = 0; | |
296 | int i = 0; | |
763ecb03 | 297 | struct vm_area_struct *vma; |
b50e195f | 298 | VMA_ITERATOR(vmi, mm, 0); |
524e00b3 | 299 | |
b50e195f LH |
300 | mt_validate(&mm->mm_mt); |
301 | for_each_vma(vmi, vma) { | |
524e00b3 | 302 | #ifdef CONFIG_DEBUG_VM_RB |
12352d3c | 303 | struct anon_vma *anon_vma = vma->anon_vma; |
ed8ea815 | 304 | struct anon_vma_chain *avc; |
b50e195f LH |
305 | #endif |
306 | unsigned long vmi_start, vmi_end; | |
307 | bool warn = 0; | |
ff26f70f | 308 | |
b50e195f LH |
309 | vmi_start = vma_iter_addr(&vmi); |
310 | vmi_end = vma_iter_end(&vmi); | |
311 | if (VM_WARN_ON_ONCE_MM(vma->vm_end != vmi_end, mm)) | |
312 | warn = 1; | |
ff26f70f | 313 | |
b50e195f LH |
314 | if (VM_WARN_ON_ONCE_MM(vma->vm_start != vmi_start, mm)) |
315 | warn = 1; | |
316 | ||
317 | if (warn) { | |
318 | pr_emerg("issue in %s\n", current->comm); | |
319 | dump_stack(); | |
320 | dump_vma(vma); | |
321 | pr_emerg("tree range: %px start %lx end %lx\n", vma, | |
322 | vmi_start, vmi_end - 1); | |
323 | vma_iter_dump_tree(&vmi); | |
324 | } | |
325 | ||
326 | #ifdef CONFIG_DEBUG_VM_RB | |
12352d3c KK |
327 | if (anon_vma) { |
328 | anon_vma_lock_read(anon_vma); | |
329 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
330 | anon_vma_interval_tree_verify(avc); | |
331 | anon_vma_unlock_read(anon_vma); | |
332 | } | |
524e00b3 | 333 | #endif |
1da177e4 LT |
334 | i++; |
335 | } | |
5a0768f6 | 336 | if (i != mm->map_count) { |
b50e195f | 337 | pr_emerg("map_count %d vma iterator %d\n", mm->map_count, i); |
5a0768f6 ML |
338 | bug = 1; |
339 | } | |
96dad67f | 340 | VM_BUG_ON_MM(bug, mm); |
1da177e4 | 341 | } |
524e00b3 LH |
342 | |
343 | #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */ | |
1da177e4 | 344 | #define validate_mm(mm) do { } while (0) |
524e00b3 | 345 | #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */ |
8f26e0b1 | 346 | |
bf181b9f ML |
347 | /* |
348 | * vma has some anon_vma assigned, and is already inserted on that | |
349 | * anon_vma's interval trees. | |
350 | * | |
351 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
352 | * vma must be removed from the anon_vma's interval trees using | |
353 | * anon_vma_interval_tree_pre_update_vma(). | |
354 | * | |
355 | * After the update, the vma will be reinserted using | |
356 | * anon_vma_interval_tree_post_update_vma(). | |
357 | * | |
c1e8d7c6 | 358 | * The entire update must be protected by exclusive mmap_lock and by |
bf181b9f ML |
359 | * the root anon_vma's mutex. |
360 | */ | |
361 | static inline void | |
362 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
363 | { | |
364 | struct anon_vma_chain *avc; | |
365 | ||
366 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
367 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
368 | } | |
369 | ||
370 | static inline void | |
371 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
372 | { | |
373 | struct anon_vma_chain *avc; | |
374 | ||
375 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
376 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
377 | } | |
378 | ||
e8420a8e CH |
379 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
380 | unsigned long addr, unsigned long end) | |
381 | { | |
2e3af1db | 382 | VMA_ITERATOR(vmi, mm, addr); |
e8420a8e | 383 | struct vm_area_struct *vma; |
2e3af1db | 384 | unsigned long nr_pages = 0; |
e8420a8e | 385 | |
2e3af1db MWO |
386 | for_each_vma_range(vmi, vma, end) { |
387 | unsigned long vm_start = max(addr, vma->vm_start); | |
388 | unsigned long vm_end = min(end, vma->vm_end); | |
e8420a8e | 389 | |
2e3af1db | 390 | nr_pages += PHYS_PFN(vm_end - vm_start); |
e8420a8e CH |
391 | } |
392 | ||
393 | return nr_pages; | |
394 | } | |
395 | ||
c154124f LH |
396 | static void __vma_link_file(struct vm_area_struct *vma, |
397 | struct address_space *mapping) | |
1da177e4 | 398 | { |
c154124f LH |
399 | if (vma->vm_flags & VM_SHARED) |
400 | mapping_allow_writable(mapping); | |
1da177e4 | 401 | |
c154124f LH |
402 | flush_dcache_mmap_lock(mapping); |
403 | vma_interval_tree_insert(vma, &mapping->i_mmap); | |
404 | flush_dcache_mmap_unlock(mapping); | |
1da177e4 LT |
405 | } |
406 | ||
763ecb03 | 407 | static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 408 | { |
79e4f2ca | 409 | VMA_ITERATOR(vmi, mm, 0); |
1da177e4 LT |
410 | struct address_space *mapping = NULL; |
411 | ||
79e4f2ca | 412 | if (vma_iter_prealloc(&vmi)) |
d4af56c5 LH |
413 | return -ENOMEM; |
414 | ||
6852c46c YM |
415 | vma_iter_store(&vmi, vma); |
416 | ||
64ac4940 | 417 | if (vma->vm_file) { |
1da177e4 | 418 | mapping = vma->vm_file->f_mapping; |
83cde9e8 | 419 | i_mmap_lock_write(mapping); |
c154124f | 420 | __vma_link_file(vma, mapping); |
83cde9e8 | 421 | i_mmap_unlock_write(mapping); |
c154124f | 422 | } |
1da177e4 LT |
423 | |
424 | mm->map_count++; | |
425 | validate_mm(mm); | |
d4af56c5 | 426 | return 0; |
1da177e4 LT |
427 | } |
428 | ||
68cefec5 LH |
429 | /* |
430 | * init_multi_vma_prep() - Initializer for struct vma_prepare | |
431 | * @vp: The vma_prepare struct | |
432 | * @vma: The vma that will be altered once locked | |
433 | * @next: The next vma if it is to be adjusted | |
434 | * @remove: The first vma to be removed | |
435 | * @remove2: The second vma to be removed | |
436 | */ | |
437 | static inline void init_multi_vma_prep(struct vma_prepare *vp, | |
438 | struct vm_area_struct *vma, struct vm_area_struct *next, | |
439 | struct vm_area_struct *remove, struct vm_area_struct *remove2) | |
440 | { | |
441 | memset(vp, 0, sizeof(struct vma_prepare)); | |
442 | vp->vma = vma; | |
443 | vp->anon_vma = vma->anon_vma; | |
444 | vp->remove = remove; | |
445 | vp->remove2 = remove2; | |
446 | vp->adj_next = next; | |
447 | if (!vp->anon_vma && next) | |
448 | vp->anon_vma = next->anon_vma; | |
449 | ||
450 | vp->file = vma->vm_file; | |
451 | if (vp->file) | |
452 | vp->mapping = vma->vm_file->f_mapping; | |
453 | ||
454 | } | |
455 | ||
456 | /* | |
457 | * init_vma_prep() - Initializer wrapper for vma_prepare struct | |
458 | * @vp: The vma_prepare struct | |
459 | * @vma: The vma that will be altered once locked | |
460 | */ | |
461 | static inline void init_vma_prep(struct vma_prepare *vp, | |
462 | struct vm_area_struct *vma) | |
463 | { | |
464 | init_multi_vma_prep(vp, vma, NULL, NULL, NULL); | |
465 | } | |
466 | ||
467 | ||
440703e0 LH |
468 | /* |
469 | * vma_prepare() - Helper function for handling locking VMAs prior to altering | |
470 | * @vp: The initialized vma_prepare struct | |
471 | */ | |
472 | static inline void vma_prepare(struct vma_prepare *vp) | |
473 | { | |
10fca64a SB |
474 | vma_start_write(vp->vma); |
475 | if (vp->adj_next) | |
476 | vma_start_write(vp->adj_next); | |
477 | /* vp->insert is always a newly created VMA, no need for locking */ | |
478 | if (vp->remove) | |
479 | vma_start_write(vp->remove); | |
480 | if (vp->remove2) | |
481 | vma_start_write(vp->remove2); | |
482 | ||
440703e0 LH |
483 | if (vp->file) { |
484 | uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); | |
485 | ||
486 | if (vp->adj_next) | |
487 | uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, | |
488 | vp->adj_next->vm_end); | |
489 | ||
490 | i_mmap_lock_write(vp->mapping); | |
491 | if (vp->insert && vp->insert->vm_file) { | |
492 | /* | |
493 | * Put into interval tree now, so instantiated pages | |
494 | * are visible to arm/parisc __flush_dcache_page | |
495 | * throughout; but we cannot insert into address | |
496 | * space until vma start or end is updated. | |
497 | */ | |
498 | __vma_link_file(vp->insert, | |
499 | vp->insert->vm_file->f_mapping); | |
500 | } | |
501 | } | |
502 | ||
503 | if (vp->anon_vma) { | |
504 | anon_vma_lock_write(vp->anon_vma); | |
505 | anon_vma_interval_tree_pre_update_vma(vp->vma); | |
506 | if (vp->adj_next) | |
507 | anon_vma_interval_tree_pre_update_vma(vp->adj_next); | |
508 | } | |
509 | ||
510 | if (vp->file) { | |
511 | flush_dcache_mmap_lock(vp->mapping); | |
512 | vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); | |
513 | if (vp->adj_next) | |
514 | vma_interval_tree_remove(vp->adj_next, | |
515 | &vp->mapping->i_mmap); | |
516 | } | |
517 | ||
518 | } | |
519 | ||
520 | /* | |
521 | * vma_complete- Helper function for handling the unlocking after altering VMAs, | |
522 | * or for inserting a VMA. | |
523 | * | |
524 | * @vp: The vma_prepare struct | |
525 | * @vmi: The vma iterator | |
526 | * @mm: The mm_struct | |
527 | */ | |
528 | static inline void vma_complete(struct vma_prepare *vp, | |
529 | struct vma_iterator *vmi, struct mm_struct *mm) | |
530 | { | |
531 | if (vp->file) { | |
532 | if (vp->adj_next) | |
533 | vma_interval_tree_insert(vp->adj_next, | |
534 | &vp->mapping->i_mmap); | |
535 | vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); | |
536 | flush_dcache_mmap_unlock(vp->mapping); | |
537 | } | |
538 | ||
539 | if (vp->remove && vp->file) { | |
540 | __remove_shared_vm_struct(vp->remove, vp->file, vp->mapping); | |
541 | if (vp->remove2) | |
542 | __remove_shared_vm_struct(vp->remove2, vp->file, | |
543 | vp->mapping); | |
544 | } else if (vp->insert) { | |
545 | /* | |
546 | * split_vma has split insert from vma, and needs | |
547 | * us to insert it before dropping the locks | |
548 | * (it may either follow vma or precede it). | |
549 | */ | |
550 | vma_iter_store(vmi, vp->insert); | |
551 | mm->map_count++; | |
552 | } | |
553 | ||
554 | if (vp->anon_vma) { | |
555 | anon_vma_interval_tree_post_update_vma(vp->vma); | |
556 | if (vp->adj_next) | |
557 | anon_vma_interval_tree_post_update_vma(vp->adj_next); | |
558 | anon_vma_unlock_write(vp->anon_vma); | |
559 | } | |
560 | ||
561 | if (vp->file) { | |
562 | i_mmap_unlock_write(vp->mapping); | |
563 | uprobe_mmap(vp->vma); | |
564 | ||
565 | if (vp->adj_next) | |
566 | uprobe_mmap(vp->adj_next); | |
567 | } | |
568 | ||
569 | if (vp->remove) { | |
570 | again: | |
457f67be | 571 | vma_mark_detached(vp->remove, true); |
440703e0 LH |
572 | if (vp->file) { |
573 | uprobe_munmap(vp->remove, vp->remove->vm_start, | |
574 | vp->remove->vm_end); | |
575 | fput(vp->file); | |
576 | } | |
577 | if (vp->remove->anon_vma) | |
578 | anon_vma_merge(vp->vma, vp->remove); | |
579 | mm->map_count--; | |
580 | mpol_put(vma_policy(vp->remove)); | |
581 | if (!vp->remove2) | |
582 | WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); | |
583 | vm_area_free(vp->remove); | |
584 | ||
585 | /* | |
586 | * In mprotect's case 6 (see comments on vma_merge), | |
5ff783f1 | 587 | * we are removing both mid and next vmas |
440703e0 LH |
588 | */ |
589 | if (vp->remove2) { | |
590 | vp->remove = vp->remove2; | |
591 | vp->remove2 = NULL; | |
592 | goto again; | |
593 | } | |
594 | } | |
595 | if (vp->insert && vp->file) | |
596 | uprobe_mmap(vp->insert); | |
597 | } | |
598 | ||
04241ffe LH |
599 | /* |
600 | * dup_anon_vma() - Helper function to duplicate anon_vma | |
601 | * @dst: The destination VMA | |
602 | * @src: The source VMA | |
603 | * | |
604 | * Returns: 0 on success. | |
605 | */ | |
606 | static inline int dup_anon_vma(struct vm_area_struct *dst, | |
607 | struct vm_area_struct *src) | |
608 | { | |
609 | /* | |
610 | * Easily overlooked: when mprotect shifts the boundary, make sure the | |
611 | * expanding vma has anon_vma set if the shrinking vma had, to cover any | |
612 | * anon pages imported. | |
613 | */ | |
614 | if (src->anon_vma && !dst->anon_vma) { | |
d8ab9f7b | 615 | vma_start_write(dst); |
04241ffe LH |
616 | dst->anon_vma = src->anon_vma; |
617 | return anon_vma_clone(dst, src); | |
618 | } | |
619 | ||
620 | return 0; | |
621 | } | |
622 | ||
9303d3e1 LH |
623 | /* |
624 | * vma_expand - Expand an existing VMA | |
625 | * | |
626 | * @vmi: The vma iterator | |
627 | * @vma: The vma to expand | |
628 | * @start: The start of the vma | |
629 | * @end: The exclusive end of the vma | |
630 | * @pgoff: The page offset of vma | |
631 | * @next: The current of next vma. | |
632 | * | |
633 | * Expand @vma to @start and @end. Can expand off the start and end. Will | |
634 | * expand over @next if it's different from @vma and @end == @next->vm_end. | |
635 | * Checking if the @vma can expand and merge with @next needs to be handled by | |
636 | * the caller. | |
637 | * | |
638 | * Returns: 0 on success | |
639 | */ | |
7c9813e8 LH |
640 | int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, |
641 | unsigned long start, unsigned long end, pgoff_t pgoff, | |
642 | struct vm_area_struct *next) | |
9303d3e1 | 643 | { |
68cefec5 | 644 | bool remove_next = false; |
9303d3e1 LH |
645 | struct vma_prepare vp; |
646 | ||
9303d3e1 | 647 | if (next && (vma != next) && (end == next->vm_end)) { |
04241ffe | 648 | int ret; |
9303d3e1 | 649 | |
04241ffe LH |
650 | remove_next = true; |
651 | ret = dup_anon_vma(vma, next); | |
652 | if (ret) | |
653 | return ret; | |
9303d3e1 LH |
654 | } |
655 | ||
68cefec5 | 656 | init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); |
9303d3e1 LH |
657 | /* Not merging but overwriting any part of next is not handled. */ |
658 | VM_WARN_ON(next && !vp.remove && | |
659 | next != vma && end > next->vm_start); | |
660 | /* Only handles expanding */ | |
661 | VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); | |
662 | ||
663 | if (vma_iter_prealloc(vmi)) | |
664 | goto nomem; | |
665 | ||
ccf1d78d | 666 | vma_prepare(&vp); |
9303d3e1 | 667 | vma_adjust_trans_huge(vma, start, end, 0); |
9303d3e1 LH |
668 | /* VMA iterator points to previous, so set to start if necessary */ |
669 | if (vma_iter_addr(vmi) != start) | |
670 | vma_iter_set(vmi, start); | |
671 | ||
9303d3e1 LH |
672 | vma->vm_start = start; |
673 | vma->vm_end = end; | |
674 | vma->vm_pgoff = pgoff; | |
675 | /* Note: mas must be pointing to the expanding VMA */ | |
676 | vma_iter_store(vmi, vma); | |
677 | ||
678 | vma_complete(&vp, vmi, vma->vm_mm); | |
679 | validate_mm(vma->vm_mm); | |
680 | return 0; | |
681 | ||
682 | nomem: | |
683 | return -ENOMEM; | |
684 | } | |
cf51e86d LH |
685 | |
686 | /* | |
687 | * vma_shrink() - Reduce an existing VMAs memory area | |
688 | * @vmi: The vma iterator | |
689 | * @vma: The VMA to modify | |
690 | * @start: The new start | |
691 | * @end: The new end | |
692 | * | |
693 | * Returns: 0 on success, -ENOMEM otherwise | |
694 | */ | |
695 | int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, | |
696 | unsigned long start, unsigned long end, pgoff_t pgoff) | |
697 | { | |
698 | struct vma_prepare vp; | |
699 | ||
700 | WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); | |
701 | ||
702 | if (vma_iter_prealloc(vmi)) | |
703 | return -ENOMEM; | |
704 | ||
705 | init_vma_prep(&vp, vma); | |
cf51e86d | 706 | vma_prepare(&vp); |
ccf1d78d | 707 | vma_adjust_trans_huge(vma, start, end, 0); |
cf51e86d LH |
708 | |
709 | if (vma->vm_start < start) | |
710 | vma_iter_clear(vmi, vma->vm_start, start); | |
711 | ||
712 | if (vma->vm_end > end) | |
713 | vma_iter_clear(vmi, end, vma->vm_end); | |
714 | ||
715 | vma->vm_start = start; | |
716 | vma->vm_end = end; | |
717 | vma->vm_pgoff = pgoff; | |
718 | vma_complete(&vp, vmi, vma->vm_mm); | |
719 | validate_mm(vma->vm_mm); | |
720 | return 0; | |
721 | } | |
722 | ||
1da177e4 LT |
723 | /* |
724 | * If the vma has a ->close operation then the driver probably needs to release | |
714965ca VB |
725 | * per-vma resources, so we don't attempt to merge those if the caller indicates |
726 | * the current vma may be removed as part of the merge. | |
1da177e4 | 727 | */ |
2dbf4010 VB |
728 | static inline bool is_mergeable_vma(struct vm_area_struct *vma, |
729 | struct file *file, unsigned long vm_flags, | |
730 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, | |
714965ca | 731 | struct anon_vma_name *anon_name, bool may_remove_vma) |
1da177e4 | 732 | { |
34228d47 CG |
733 | /* |
734 | * VM_SOFTDIRTY should not prevent from VMA merging, if we | |
735 | * match the flags but dirty bit -- the caller should mark | |
736 | * merged VMA as dirty. If dirty bit won't be excluded from | |
8bb4e7a2 | 737 | * comparison, we increase pressure on the memory system forcing |
34228d47 CG |
738 | * the kernel to generate new VMAs when old one could be |
739 | * extended instead. | |
740 | */ | |
741 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) | |
2dbf4010 | 742 | return false; |
1da177e4 | 743 | if (vma->vm_file != file) |
2dbf4010 | 744 | return false; |
714965ca | 745 | if (may_remove_vma && vma->vm_ops && vma->vm_ops->close) |
2dbf4010 | 746 | return false; |
19a809af | 747 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) |
2dbf4010 | 748 | return false; |
5c26f6ac | 749 | if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) |
2dbf4010 VB |
750 | return false; |
751 | return true; | |
1da177e4 LT |
752 | } |
753 | ||
2dbf4010 VB |
754 | static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, |
755 | struct anon_vma *anon_vma2, struct vm_area_struct *vma) | |
1da177e4 | 756 | { |
965f55de SL |
757 | /* |
758 | * The list_is_singular() test is to avoid merging VMA cloned from | |
759 | * parents. This can improve scalability caused by anon_vma lock. | |
760 | */ | |
761 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
762 | list_is_singular(&vma->anon_vma_chain))) | |
2dbf4010 | 763 | return true; |
965f55de | 764 | return anon_vma1 == anon_vma2; |
1da177e4 LT |
765 | } |
766 | ||
767 | /* | |
768 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
769 | * in front of (at a lower virtual address and file offset than) the vma. | |
770 | * | |
771 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
772 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
773 | * | |
774 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
45e55300 | 775 | * indices (16TB on ia32) because do_mmap() does not permit mmap's which |
1da177e4 | 776 | * wrap, nor mmaps which cover the final page at index -1UL. |
714965ca VB |
777 | * |
778 | * We assume the vma may be removed as part of the merge. | |
1da177e4 | 779 | */ |
2dbf4010 | 780 | static bool |
1da177e4 | 781 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, |
2dbf4010 VB |
782 | struct anon_vma *anon_vma, struct file *file, |
783 | pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, | |
784 | struct anon_vma_name *anon_name) | |
1da177e4 | 785 | { |
714965ca | 786 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) && |
965f55de | 787 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 788 | if (vma->vm_pgoff == vm_pgoff) |
2dbf4010 | 789 | return true; |
1da177e4 | 790 | } |
2dbf4010 | 791 | return false; |
1da177e4 LT |
792 | } |
793 | ||
794 | /* | |
795 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
796 | * beyond (at a higher virtual address and file offset than) the vma. | |
797 | * | |
798 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
799 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
714965ca VB |
800 | * |
801 | * We assume that vma is not removed as part of the merge. | |
1da177e4 | 802 | */ |
2dbf4010 | 803 | static bool |
1da177e4 | 804 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, |
2dbf4010 VB |
805 | struct anon_vma *anon_vma, struct file *file, |
806 | pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, | |
807 | struct anon_vma_name *anon_name) | |
1da177e4 | 808 | { |
714965ca | 809 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) && |
965f55de | 810 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 811 | pgoff_t vm_pglen; |
d6e93217 | 812 | vm_pglen = vma_pages(vma); |
1da177e4 | 813 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
2dbf4010 | 814 | return true; |
1da177e4 | 815 | } |
2dbf4010 | 816 | return false; |
1da177e4 LT |
817 | } |
818 | ||
819 | /* | |
9a10064f CC |
820 | * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), |
821 | * figure out whether that can be merged with its predecessor or its | |
822 | * successor. Or both (it neatly fills a hole). | |
1da177e4 LT |
823 | * |
824 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
825 | * certain not to be mapped by the time vma_merge is called; but when | |
826 | * called for mprotect, it is certain to be already mapped (either at | |
827 | * an offset within prev, or at the start of next), and the flags of | |
828 | * this area are about to be changed to vm_flags - and the no-change | |
829 | * case has already been eliminated. | |
830 | * | |
fcfccd91 | 831 | * The following mprotect cases have to be considered, where **** is |
1da177e4 | 832 | * the area passed down from mprotect_fixup, never extending beyond one |
fcfccd91 LS |
833 | * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts |
834 | * at the same address as **** and is of the same or larger span, and | |
835 | * NNNN the next vma after ****: | |
1da177e4 | 836 | * |
fcfccd91 LS |
837 | * **** **** **** |
838 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC | |
5d42ab29 | 839 | * cannot merge might become might become |
fcfccd91 | 840 | * PPNNNNNNNNNN PPPPPPPPPPCC |
5d42ab29 WY |
841 | * mmap, brk or case 4 below case 5 below |
842 | * mremap move: | |
fcfccd91 LS |
843 | * **** **** |
844 | * PPPP NNNN PPPPCCCCNNNN | |
5d42ab29 WY |
845 | * might become might become |
846 | * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or | |
fcfccd91 LS |
847 | * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or |
848 | * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8 | |
1da177e4 | 849 | * |
fcfccd91 LS |
850 | * It is important for case 8 that the vma CCCC overlapping the |
851 | * region **** is never going to extended over NNNN. Instead NNNN must | |
852 | * be extended in region **** and CCCC must be removed. This way in | |
0503ea8f | 853 | * all cases where vma_merge succeeds, the moment vma_merge drops the |
e86f15ee AA |
854 | * rmap_locks, the properties of the merged vma will be already |
855 | * correct for the whole merged range. Some of those properties like | |
856 | * vm_page_prot/vm_flags may be accessed by rmap_walks and they must | |
857 | * be correct for the whole merged range immediately after the | |
fcfccd91 LS |
858 | * rmap_locks are released. Otherwise if NNNN would be removed and |
859 | * CCCC would be extended over the NNNN range, remove_migration_ptes | |
e86f15ee | 860 | * or other rmap walkers (if working on addresses beyond the "end" |
fcfccd91 LS |
861 | * parameter) may establish ptes with the wrong permissions of CCCC |
862 | * instead of the right permissions of NNNN. | |
0503ea8f LH |
863 | * |
864 | * In the code below: | |
865 | * PPPP is represented by *prev | |
fcfccd91 LS |
866 | * CCCC is represented by *curr or not represented at all (NULL) |
867 | * NNNN is represented by *next or not represented at all (NULL) | |
868 | * **** is not represented - it will be merged and the vma containing the | |
9e8a39d2 | 869 | * area is returned, or the function will return NULL |
1da177e4 | 870 | */ |
9760ebff | 871 | struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, |
1da177e4 LT |
872 | struct vm_area_struct *prev, unsigned long addr, |
873 | unsigned long end, unsigned long vm_flags, | |
cc71aba3 | 874 | struct anon_vma *anon_vma, struct file *file, |
19a809af | 875 | pgoff_t pgoff, struct mempolicy *policy, |
9a10064f | 876 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, |
5c26f6ac | 877 | struct anon_vma_name *anon_name) |
1da177e4 | 878 | { |
b0729ae0 | 879 | struct vm_area_struct *curr, *next, *res; |
0503ea8f | 880 | struct vm_area_struct *vma, *adjust, *remove, *remove2; |
0173db4f LS |
881 | struct vma_prepare vp; |
882 | pgoff_t vma_pgoff; | |
883 | int err = 0; | |
eef19944 JM |
884 | bool merge_prev = false; |
885 | bool merge_next = false; | |
0503ea8f | 886 | bool vma_expanded = false; |
0503ea8f | 887 | unsigned long vma_start = addr; |
0503ea8f | 888 | unsigned long vma_end = end; |
0173db4f | 889 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; |
1e76454f | 890 | long adj_start = 0; |
1da177e4 | 891 | |
0503ea8f | 892 | validate_mm(mm); |
1da177e4 LT |
893 | /* |
894 | * We later require that vma->vm_flags == vm_flags, | |
895 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
896 | */ | |
897 | if (vm_flags & VM_SPECIAL) | |
898 | return NULL; | |
899 | ||
00cd00a6 LS |
900 | /* Does the input range span an existing VMA? (cases 5 - 8) */ |
901 | curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end); | |
1da177e4 | 902 | |
00cd00a6 LS |
903 | if (!curr || /* cases 1 - 4 */ |
904 | end == curr->vm_end) /* cases 6 - 8, adjacent VMA */ | |
905 | next = vma_lookup(mm, end); | |
906 | else | |
907 | next = NULL; /* case 5 */ | |
e86f15ee | 908 | |
0503ea8f | 909 | if (prev) { |
0503ea8f LH |
910 | vma_start = prev->vm_start; |
911 | vma_pgoff = prev->vm_pgoff; | |
0173db4f | 912 | |
0503ea8f | 913 | /* Can we merge the predecessor? */ |
0173db4f | 914 | if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy) |
0503ea8f | 915 | && can_vma_merge_after(prev, vm_flags, anon_vma, file, |
0173db4f | 916 | pgoff, vm_userfaultfd_ctx, anon_name)) { |
0503ea8f | 917 | merge_prev = true; |
18b098af | 918 | vma_prev(vmi); |
0503ea8f | 919 | } |
1da177e4 | 920 | } |
b0729ae0 | 921 | |
eef19944 | 922 | /* Can we merge the successor? */ |
00cd00a6 | 923 | if (next && mpol_equal(policy, vma_policy(next)) && |
0173db4f | 924 | can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen, |
00cd00a6 | 925 | vm_userfaultfd_ctx, anon_name)) { |
eef19944 JM |
926 | merge_next = true; |
927 | } | |
0503ea8f | 928 | |
29417d29 LS |
929 | /* Verify some invariant that must be enforced by the caller. */ |
930 | VM_WARN_ON(prev && addr <= prev->vm_start); | |
931 | VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end)); | |
932 | VM_WARN_ON(addr >= end); | |
933 | ||
0173db4f LS |
934 | if (!merge_prev && !merge_next) |
935 | return NULL; /* Not mergeable. */ | |
936 | ||
937 | res = vma = prev; | |
0503ea8f | 938 | remove = remove2 = adjust = NULL; |
0173db4f | 939 | |
eef19944 JM |
940 | /* Can we merge both the predecessor and the successor? */ |
941 | if (merge_prev && merge_next && | |
0503ea8f | 942 | is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { |
5ff783f1 | 943 | remove = next; /* case 1 */ |
0503ea8f | 944 | vma_end = next->vm_end; |
5ff783f1 | 945 | err = dup_anon_vma(prev, next); |
fcfccd91 LS |
946 | if (curr) { /* case 6 */ |
947 | remove = curr; | |
0503ea8f | 948 | remove2 = next; |
5ff783f1 | 949 | if (!next->anon_vma) |
fcfccd91 | 950 | err = dup_anon_vma(prev, curr); |
0503ea8f | 951 | } |
0173db4f | 952 | } else if (merge_prev) { /* case 2 */ |
fcfccd91 LS |
953 | if (curr) { |
954 | err = dup_anon_vma(prev, curr); | |
955 | if (end == curr->vm_end) { /* case 7 */ | |
956 | remove = curr; | |
0503ea8f | 957 | } else { /* case 5 */ |
fcfccd91 LS |
958 | adjust = curr; |
959 | adj_start = (end - curr->vm_start); | |
0503ea8f LH |
960 | } |
961 | } | |
0173db4f | 962 | } else { /* merge_next */ |
eef19944 | 963 | res = next; |
0503ea8f LH |
964 | if (prev && addr < prev->vm_end) { /* case 4 */ |
965 | vma_end = addr; | |
183b7a60 | 966 | adjust = next; |
1e76454f | 967 | adj_start = -(prev->vm_end - addr); |
183b7a60 | 968 | err = dup_anon_vma(next, prev); |
0503ea8f | 969 | } else { |
b0729ae0 LS |
970 | /* |
971 | * Note that cases 3 and 8 are the ONLY ones where prev | |
972 | * is permitted to be (but is not necessarily) NULL. | |
973 | */ | |
0503ea8f LH |
974 | vma = next; /* case 3 */ |
975 | vma_start = addr; | |
976 | vma_end = next->vm_end; | |
7e775787 | 977 | vma_pgoff = next->vm_pgoff - pglen; |
fcfccd91 LS |
978 | if (curr) { /* case 8 */ |
979 | vma_pgoff = curr->vm_pgoff; | |
980 | remove = curr; | |
981 | err = dup_anon_vma(next, curr); | |
0503ea8f LH |
982 | } |
983 | } | |
1da177e4 LT |
984 | } |
985 | ||
0173db4f | 986 | /* Error in anon_vma clone. */ |
eef19944 JM |
987 | if (err) |
988 | return NULL; | |
0503ea8f LH |
989 | |
990 | if (vma_iter_prealloc(vmi)) | |
991 | return NULL; | |
992 | ||
0503ea8f LH |
993 | init_multi_vma_prep(&vp, vma, adjust, remove, remove2); |
994 | VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && | |
995 | vp.anon_vma != adjust->anon_vma); | |
996 | ||
997 | vma_prepare(&vp); | |
ccf1d78d | 998 | vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); |
0503ea8f LH |
999 | if (vma_start < vma->vm_start || vma_end > vma->vm_end) |
1000 | vma_expanded = true; | |
1001 | ||
1002 | vma->vm_start = vma_start; | |
1003 | vma->vm_end = vma_end; | |
1004 | vma->vm_pgoff = vma_pgoff; | |
1005 | ||
1006 | if (vma_expanded) | |
1007 | vma_iter_store(vmi, vma); | |
1008 | ||
1e76454f VB |
1009 | if (adj_start) { |
1010 | adjust->vm_start += adj_start; | |
1011 | adjust->vm_pgoff += adj_start >> PAGE_SHIFT; | |
1012 | if (adj_start < 0) { | |
0503ea8f LH |
1013 | WARN_ON(vma_expanded); |
1014 | vma_iter_store(vmi, next); | |
1015 | } | |
1016 | } | |
1017 | ||
1018 | vma_complete(&vp, vmi, mm); | |
1019 | vma_iter_free(vmi); | |
1020 | validate_mm(mm); | |
eef19944 | 1021 | khugepaged_enter_vma(res, vm_flags); |
1da177e4 | 1022 | |
9760ebff | 1023 | return res; |
f2ebfe43 LH |
1024 | } |
1025 | ||
d0e9fe17 | 1026 | /* |
b4f315b4 | 1027 | * Rough compatibility check to quickly see if it's even worth looking |
d0e9fe17 LT |
1028 | * at sharing an anon_vma. |
1029 | * | |
1030 | * They need to have the same vm_file, and the flags can only differ | |
1031 | * in things that mprotect may change. | |
1032 | * | |
1033 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1034 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1035 | * there is a vm_ops->close() function, because that indicates that the | |
1036 | * driver is doing some kind of reference counting. But that doesn't | |
1037 | * really matter for the anon_vma sharing case. | |
1038 | */ | |
1039 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1040 | { | |
1041 | return a->vm_end == b->vm_start && | |
1042 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1043 | a->vm_file == b->vm_file && | |
6cb4d9a2 | 1044 | !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && |
d0e9fe17 LT |
1045 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1050 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1051 | * the same as 'old', the other will be the new one that is trying | |
1052 | * to share the anon_vma. | |
1053 | * | |
5b449489 | 1054 | * NOTE! This runs with mmap_lock held for reading, so it is possible that |
d0e9fe17 LT |
1055 | * the anon_vma of 'old' is concurrently in the process of being set up |
1056 | * by another page fault trying to merge _that_. But that's ok: if it | |
1057 | * is being set up, that automatically means that it will be a singleton | |
1058 | * acceptable for merging, so we can do all of this optimistically. But | |
4db0c3c2 | 1059 | * we do that READ_ONCE() to make sure that we never re-load the pointer. |
d0e9fe17 LT |
1060 | * |
1061 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1062 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1063 | * is to return an anon_vma that is "complex" due to having gone through | |
1064 | * a fork). | |
1065 | * | |
1066 | * We also make sure that the two vma's are compatible (adjacent, | |
1067 | * and with the same memory policies). That's all stable, even with just | |
5b449489 | 1068 | * a read lock on the mmap_lock. |
d0e9fe17 LT |
1069 | */ |
1070 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1071 | { | |
1072 | if (anon_vma_compatible(a, b)) { | |
4db0c3c2 | 1073 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); |
d0e9fe17 LT |
1074 | |
1075 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1076 | return anon_vma; | |
1077 | } | |
1078 | return NULL; | |
1079 | } | |
1080 | ||
1da177e4 LT |
1081 | /* |
1082 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1083 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1084 | * to allocate a new anon_vma. It checks because a repetitive | |
1085 | * sequence of mprotects and faults may otherwise lead to distinct | |
1086 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1087 | * mprotect. | |
1088 | */ | |
1089 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1090 | { | |
763ecb03 | 1091 | MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); |
a67c8caa | 1092 | struct anon_vma *anon_vma = NULL; |
763ecb03 | 1093 | struct vm_area_struct *prev, *next; |
a67c8caa ML |
1094 | |
1095 | /* Try next first. */ | |
763ecb03 LH |
1096 | next = mas_walk(&mas); |
1097 | if (next) { | |
1098 | anon_vma = reusable_anon_vma(next, vma, next); | |
a67c8caa ML |
1099 | if (anon_vma) |
1100 | return anon_vma; | |
1101 | } | |
1102 | ||
763ecb03 LH |
1103 | prev = mas_prev(&mas, 0); |
1104 | VM_BUG_ON_VMA(prev != vma, vma); | |
1105 | prev = mas_prev(&mas, 0); | |
a67c8caa | 1106 | /* Try prev next. */ |
763ecb03 LH |
1107 | if (prev) |
1108 | anon_vma = reusable_anon_vma(prev, prev, vma); | |
a67c8caa | 1109 | |
1da177e4 | 1110 | /* |
a67c8caa ML |
1111 | * We might reach here with anon_vma == NULL if we can't find |
1112 | * any reusable anon_vma. | |
1da177e4 LT |
1113 | * There's no absolute need to look only at touching neighbours: |
1114 | * we could search further afield for "compatible" anon_vmas. | |
1115 | * But it would probably just be a waste of time searching, | |
1116 | * or lead to too many vmas hanging off the same anon_vma. | |
1117 | * We're trying to allow mprotect remerging later on, | |
1118 | * not trying to minimize memory used for anon_vmas. | |
1119 | */ | |
a67c8caa | 1120 | return anon_vma; |
1da177e4 LT |
1121 | } |
1122 | ||
40401530 AV |
1123 | /* |
1124 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1125 | * low as possible but still greater than mmap_min_addr | |
1126 | */ | |
1127 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1128 | { | |
1129 | hint &= PAGE_MASK; | |
1130 | if (((void *)hint != NULL) && | |
1131 | (hint < mmap_min_addr)) | |
1132 | return PAGE_ALIGN(mmap_min_addr); | |
1133 | return hint; | |
1134 | } | |
1135 | ||
b0cc5e89 | 1136 | bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, |
3c54a298 | 1137 | unsigned long bytes) |
363ee17f | 1138 | { |
3c54a298 | 1139 | unsigned long locked_pages, limit_pages; |
363ee17f | 1140 | |
3c54a298 LS |
1141 | if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) |
1142 | return true; | |
1143 | ||
1144 | locked_pages = bytes >> PAGE_SHIFT; | |
1145 | locked_pages += mm->locked_vm; | |
1146 | ||
1147 | limit_pages = rlimit(RLIMIT_MEMLOCK); | |
1148 | limit_pages >>= PAGE_SHIFT; | |
1149 | ||
1150 | return locked_pages <= limit_pages; | |
363ee17f DB |
1151 | } |
1152 | ||
be83bbf8 LT |
1153 | static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) |
1154 | { | |
1155 | if (S_ISREG(inode->i_mode)) | |
423913ad | 1156 | return MAX_LFS_FILESIZE; |
be83bbf8 LT |
1157 | |
1158 | if (S_ISBLK(inode->i_mode)) | |
1159 | return MAX_LFS_FILESIZE; | |
1160 | ||
76f34950 IK |
1161 | if (S_ISSOCK(inode->i_mode)) |
1162 | return MAX_LFS_FILESIZE; | |
1163 | ||
be83bbf8 LT |
1164 | /* Special "we do even unsigned file positions" case */ |
1165 | if (file->f_mode & FMODE_UNSIGNED_OFFSET) | |
1166 | return 0; | |
1167 | ||
1168 | /* Yes, random drivers might want more. But I'm tired of buggy drivers */ | |
1169 | return ULONG_MAX; | |
1170 | } | |
1171 | ||
1172 | static inline bool file_mmap_ok(struct file *file, struct inode *inode, | |
1173 | unsigned long pgoff, unsigned long len) | |
1174 | { | |
1175 | u64 maxsize = file_mmap_size_max(file, inode); | |
1176 | ||
1177 | if (maxsize && len > maxsize) | |
1178 | return false; | |
1179 | maxsize -= len; | |
1180 | if (pgoff > maxsize >> PAGE_SHIFT) | |
1181 | return false; | |
1182 | return true; | |
1183 | } | |
1184 | ||
1da177e4 | 1185 | /* |
3e4e28c5 | 1186 | * The caller must write-lock current->mm->mmap_lock. |
1da177e4 | 1187 | */ |
1fcfd8db | 1188 | unsigned long do_mmap(struct file *file, unsigned long addr, |
1da177e4 | 1189 | unsigned long len, unsigned long prot, |
45e55300 PC |
1190 | unsigned long flags, unsigned long pgoff, |
1191 | unsigned long *populate, struct list_head *uf) | |
1da177e4 | 1192 | { |
cc71aba3 | 1193 | struct mm_struct *mm = current->mm; |
45e55300 | 1194 | vm_flags_t vm_flags; |
62b5f7d0 | 1195 | int pkey = 0; |
1da177e4 | 1196 | |
524e00b3 | 1197 | validate_mm(mm); |
41badc15 | 1198 | *populate = 0; |
bebeb3d6 | 1199 | |
e37609bb PK |
1200 | if (!len) |
1201 | return -EINVAL; | |
1202 | ||
1da177e4 LT |
1203 | /* |
1204 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1205 | * | |
1206 | * (the exception is when the underlying filesystem is noexec | |
1207 | * mounted, in which case we dont add PROT_EXEC.) | |
1208 | */ | |
1209 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
90f8572b | 1210 | if (!(file && path_noexec(&file->f_path))) |
1da177e4 LT |
1211 | prot |= PROT_EXEC; |
1212 | ||
a4ff8e86 MH |
1213 | /* force arch specific MAP_FIXED handling in get_unmapped_area */ |
1214 | if (flags & MAP_FIXED_NOREPLACE) | |
1215 | flags |= MAP_FIXED; | |
1216 | ||
7cd94146 EP |
1217 | if (!(flags & MAP_FIXED)) |
1218 | addr = round_hint_to_min(addr); | |
1219 | ||
1da177e4 LT |
1220 | /* Careful about overflows.. */ |
1221 | len = PAGE_ALIGN(len); | |
9206de95 | 1222 | if (!len) |
1da177e4 LT |
1223 | return -ENOMEM; |
1224 | ||
1225 | /* offset overflow? */ | |
1226 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
cc71aba3 | 1227 | return -EOVERFLOW; |
1da177e4 LT |
1228 | |
1229 | /* Too many mappings? */ | |
1230 | if (mm->map_count > sysctl_max_map_count) | |
1231 | return -ENOMEM; | |
1232 | ||
1233 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1234 | * that it represents a valid section of the address space. | |
1235 | */ | |
1236 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
ff68dac6 | 1237 | if (IS_ERR_VALUE(addr)) |
1da177e4 LT |
1238 | return addr; |
1239 | ||
a4ff8e86 | 1240 | if (flags & MAP_FIXED_NOREPLACE) { |
35e43c5f | 1241 | if (find_vma_intersection(mm, addr, addr + len)) |
a4ff8e86 MH |
1242 | return -EEXIST; |
1243 | } | |
1244 | ||
62b5f7d0 DH |
1245 | if (prot == PROT_EXEC) { |
1246 | pkey = execute_only_pkey(mm); | |
1247 | if (pkey < 0) | |
1248 | pkey = 0; | |
1249 | } | |
1250 | ||
1da177e4 LT |
1251 | /* Do simple checking here so the lower-level routines won't have |
1252 | * to. we assume access permissions have been handled by the open | |
1253 | * of the memory object, so we don't do any here. | |
1254 | */ | |
45e55300 | 1255 | vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | |
1da177e4 LT |
1256 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1257 | ||
cdf7b341 | 1258 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1259 | if (!can_do_mlock()) |
1260 | return -EPERM; | |
ba470de4 | 1261 | |
b0cc5e89 | 1262 | if (!mlock_future_ok(mm, vm_flags, len)) |
363ee17f | 1263 | return -EAGAIN; |
1da177e4 | 1264 | |
1da177e4 | 1265 | if (file) { |
077bf22b | 1266 | struct inode *inode = file_inode(file); |
1c972597 DW |
1267 | unsigned long flags_mask; |
1268 | ||
be83bbf8 LT |
1269 | if (!file_mmap_ok(file, inode, pgoff, len)) |
1270 | return -EOVERFLOW; | |
1271 | ||
1c972597 | 1272 | flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; |
077bf22b | 1273 | |
1da177e4 LT |
1274 | switch (flags & MAP_TYPE) { |
1275 | case MAP_SHARED: | |
1c972597 DW |
1276 | /* |
1277 | * Force use of MAP_SHARED_VALIDATE with non-legacy | |
1278 | * flags. E.g. MAP_SYNC is dangerous to use with | |
1279 | * MAP_SHARED as you don't know which consistency model | |
1280 | * you will get. We silently ignore unsupported flags | |
1281 | * with MAP_SHARED to preserve backward compatibility. | |
1282 | */ | |
1283 | flags &= LEGACY_MAP_MASK; | |
e4a9bc58 | 1284 | fallthrough; |
1c972597 DW |
1285 | case MAP_SHARED_VALIDATE: |
1286 | if (flags & ~flags_mask) | |
1287 | return -EOPNOTSUPP; | |
dc617f29 DW |
1288 | if (prot & PROT_WRITE) { |
1289 | if (!(file->f_mode & FMODE_WRITE)) | |
1290 | return -EACCES; | |
1291 | if (IS_SWAPFILE(file->f_mapping->host)) | |
1292 | return -ETXTBSY; | |
1293 | } | |
1da177e4 LT |
1294 | |
1295 | /* | |
1296 | * Make sure we don't allow writing to an append-only | |
1297 | * file.. | |
1298 | */ | |
1299 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1300 | return -EACCES; | |
1301 | ||
1da177e4 LT |
1302 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1303 | if (!(file->f_mode & FMODE_WRITE)) | |
1304 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
e4a9bc58 | 1305 | fallthrough; |
1da177e4 LT |
1306 | case MAP_PRIVATE: |
1307 | if (!(file->f_mode & FMODE_READ)) | |
1308 | return -EACCES; | |
90f8572b | 1309 | if (path_noexec(&file->f_path)) { |
80c5606c LT |
1310 | if (vm_flags & VM_EXEC) |
1311 | return -EPERM; | |
1312 | vm_flags &= ~VM_MAYEXEC; | |
1313 | } | |
80c5606c | 1314 | |
72c2d531 | 1315 | if (!file->f_op->mmap) |
80c5606c | 1316 | return -ENODEV; |
b2c56e4f ON |
1317 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1318 | return -EINVAL; | |
1da177e4 LT |
1319 | break; |
1320 | ||
1321 | default: | |
1322 | return -EINVAL; | |
1323 | } | |
1324 | } else { | |
1325 | switch (flags & MAP_TYPE) { | |
1326 | case MAP_SHARED: | |
b2c56e4f ON |
1327 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1328 | return -EINVAL; | |
ce363942 TH |
1329 | /* |
1330 | * Ignore pgoff. | |
1331 | */ | |
1332 | pgoff = 0; | |
1da177e4 LT |
1333 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1334 | break; | |
1335 | case MAP_PRIVATE: | |
1336 | /* | |
1337 | * Set pgoff according to addr for anon_vma. | |
1338 | */ | |
1339 | pgoff = addr >> PAGE_SHIFT; | |
1340 | break; | |
1341 | default: | |
1342 | return -EINVAL; | |
1343 | } | |
1344 | } | |
1345 | ||
c22c0d63 ML |
1346 | /* |
1347 | * Set 'VM_NORESERVE' if we should not account for the | |
1348 | * memory use of this mapping. | |
1349 | */ | |
1350 | if (flags & MAP_NORESERVE) { | |
1351 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1352 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1353 | vm_flags |= VM_NORESERVE; | |
1354 | ||
1355 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1356 | if (file && is_file_hugepages(file)) | |
1357 | vm_flags |= VM_NORESERVE; | |
1358 | } | |
1359 | ||
897ab3e0 | 1360 | addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); |
09a9f1d2 ML |
1361 | if (!IS_ERR_VALUE(addr) && |
1362 | ((vm_flags & VM_LOCKED) || | |
1363 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1364 | *populate = len; |
bebeb3d6 | 1365 | return addr; |
0165ab44 | 1366 | } |
6be5ceb0 | 1367 | |
a90f590a DB |
1368 | unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, |
1369 | unsigned long prot, unsigned long flags, | |
1370 | unsigned long fd, unsigned long pgoff) | |
66f0dc48 HD |
1371 | { |
1372 | struct file *file = NULL; | |
1e3ee14b | 1373 | unsigned long retval; |
66f0dc48 HD |
1374 | |
1375 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1376 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1377 | file = fget(fd); |
1378 | if (!file) | |
1e3ee14b | 1379 | return -EBADF; |
7bba8f0e | 1380 | if (is_file_hugepages(file)) { |
af73e4d9 | 1381 | len = ALIGN(len, huge_page_size(hstate_file(file))); |
7bba8f0e ZL |
1382 | } else if (unlikely(flags & MAP_HUGETLB)) { |
1383 | retval = -EINVAL; | |
493af578 | 1384 | goto out_fput; |
7bba8f0e | 1385 | } |
66f0dc48 | 1386 | } else if (flags & MAP_HUGETLB) { |
c103a4dc | 1387 | struct hstate *hs; |
af73e4d9 | 1388 | |
20ac2893 | 1389 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); |
091d0d55 LZ |
1390 | if (!hs) |
1391 | return -EINVAL; | |
1392 | ||
1393 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1394 | /* |
1395 | * VM_NORESERVE is used because the reservations will be | |
1396 | * taken when vm_ops->mmap() is called | |
66f0dc48 | 1397 | */ |
af73e4d9 | 1398 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f | 1399 | VM_NORESERVE, |
83c1fd76 | 1400 | HUGETLB_ANONHUGE_INODE, |
42d7395f | 1401 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); |
66f0dc48 HD |
1402 | if (IS_ERR(file)) |
1403 | return PTR_ERR(file); | |
1404 | } | |
1405 | ||
9fbeb5ab | 1406 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
493af578 | 1407 | out_fput: |
66f0dc48 HD |
1408 | if (file) |
1409 | fput(file); | |
66f0dc48 HD |
1410 | return retval; |
1411 | } | |
1412 | ||
a90f590a DB |
1413 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1414 | unsigned long, prot, unsigned long, flags, | |
1415 | unsigned long, fd, unsigned long, pgoff) | |
1416 | { | |
1417 | return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); | |
1418 | } | |
1419 | ||
a4679373 CH |
1420 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1421 | struct mmap_arg_struct { | |
1422 | unsigned long addr; | |
1423 | unsigned long len; | |
1424 | unsigned long prot; | |
1425 | unsigned long flags; | |
1426 | unsigned long fd; | |
1427 | unsigned long offset; | |
1428 | }; | |
1429 | ||
1430 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1431 | { | |
1432 | struct mmap_arg_struct a; | |
1433 | ||
1434 | if (copy_from_user(&a, arg, sizeof(a))) | |
1435 | return -EFAULT; | |
de1741a1 | 1436 | if (offset_in_page(a.offset)) |
a4679373 CH |
1437 | return -EINVAL; |
1438 | ||
a90f590a DB |
1439 | return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, |
1440 | a.offset >> PAGE_SHIFT); | |
a4679373 CH |
1441 | } |
1442 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1443 | ||
54cbbbf3 LS |
1444 | static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops) |
1445 | { | |
1446 | return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite); | |
1447 | } | |
1448 | ||
1449 | static bool vma_is_shared_writable(struct vm_area_struct *vma) | |
1450 | { | |
1451 | return (vma->vm_flags & (VM_WRITE | VM_SHARED)) == | |
1452 | (VM_WRITE | VM_SHARED); | |
1453 | } | |
1454 | ||
1455 | static bool vma_fs_can_writeback(struct vm_area_struct *vma) | |
1456 | { | |
1457 | /* No managed pages to writeback. */ | |
1458 | if (vma->vm_flags & VM_PFNMAP) | |
1459 | return false; | |
1460 | ||
1461 | return vma->vm_file && vma->vm_file->f_mapping && | |
1462 | mapping_can_writeback(vma->vm_file->f_mapping); | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Does this VMA require the underlying folios to have their dirty state | |
1467 | * tracked? | |
1468 | */ | |
1469 | bool vma_needs_dirty_tracking(struct vm_area_struct *vma) | |
1470 | { | |
1471 | /* Only shared, writable VMAs require dirty tracking. */ | |
1472 | if (!vma_is_shared_writable(vma)) | |
1473 | return false; | |
1474 | ||
1475 | /* Does the filesystem need to be notified? */ | |
1476 | if (vm_ops_needs_writenotify(vma->vm_ops)) | |
1477 | return true; | |
1478 | ||
1479 | /* | |
1480 | * Even if the filesystem doesn't indicate a need for writenotify, if it | |
1481 | * can writeback, dirty tracking is still required. | |
1482 | */ | |
1483 | return vma_fs_can_writeback(vma); | |
1484 | } | |
1485 | ||
4e950f6f | 1486 | /* |
8bb4e7a2 | 1487 | * Some shared mappings will want the pages marked read-only |
4e950f6f AD |
1488 | * to track write events. If so, we'll downgrade vm_page_prot |
1489 | * to the private version (using protection_map[] without the | |
1490 | * VM_SHARED bit). | |
1491 | */ | |
6d2329f8 | 1492 | int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) |
4e950f6f | 1493 | { |
4e950f6f | 1494 | /* If it was private or non-writable, the write bit is already clear */ |
54cbbbf3 | 1495 | if (!vma_is_shared_writable(vma)) |
4e950f6f AD |
1496 | return 0; |
1497 | ||
1498 | /* The backer wishes to know when pages are first written to? */ | |
54cbbbf3 | 1499 | if (vm_ops_needs_writenotify(vma->vm_ops)) |
4e950f6f AD |
1500 | return 1; |
1501 | ||
64e45507 PF |
1502 | /* The open routine did something to the protections that pgprot_modify |
1503 | * won't preserve? */ | |
6d2329f8 | 1504 | if (pgprot_val(vm_page_prot) != |
54cbbbf3 | 1505 | pgprot_val(vm_pgprot_modify(vm_page_prot, vma->vm_flags))) |
4e950f6f AD |
1506 | return 0; |
1507 | ||
f96f7a40 DH |
1508 | /* |
1509 | * Do we need to track softdirty? hugetlb does not support softdirty | |
1510 | * tracking yet. | |
1511 | */ | |
1512 | if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) | |
64e45507 PF |
1513 | return 1; |
1514 | ||
51d3d5eb DH |
1515 | /* Do we need write faults for uffd-wp tracking? */ |
1516 | if (userfaultfd_wp(vma)) | |
1517 | return 1; | |
1518 | ||
4e950f6f | 1519 | /* Can the mapping track the dirty pages? */ |
54cbbbf3 | 1520 | return vma_fs_can_writeback(vma); |
4e950f6f AD |
1521 | } |
1522 | ||
fc8744ad LT |
1523 | /* |
1524 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1525 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1526 | */ |
ca16d140 | 1527 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1528 | { |
5a6fe125 MG |
1529 | /* |
1530 | * hugetlb has its own accounting separate from the core VM | |
1531 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1532 | */ | |
1533 | if (file && is_file_hugepages(file)) | |
1534 | return 0; | |
1535 | ||
fc8744ad LT |
1536 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1537 | } | |
1538 | ||
3499a131 LH |
1539 | /** |
1540 | * unmapped_area() - Find an area between the low_limit and the high_limit with | |
1541 | * the correct alignment and offset, all from @info. Note: current->mm is used | |
1542 | * for the search. | |
1543 | * | |
82b24936 VY |
1544 | * @info: The unmapped area information including the range [low_limit - |
1545 | * high_limit), the alignment offset and mask. | |
3499a131 LH |
1546 | * |
1547 | * Return: A memory address or -ENOMEM. | |
1548 | */ | |
baceaf1c | 1549 | static unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
db4fbfb9 | 1550 | { |
6b008640 LT |
1551 | unsigned long length, gap; |
1552 | unsigned long low_limit, high_limit; | |
58c5d0d6 | 1553 | struct vm_area_struct *tmp; |
db4fbfb9 | 1554 | |
3499a131 | 1555 | MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); |
db4fbfb9 ML |
1556 | |
1557 | /* Adjust search length to account for worst case alignment overhead */ | |
1558 | length = info->length + info->align_mask; | |
1559 | if (length < info->length) | |
1560 | return -ENOMEM; | |
1561 | ||
58c5d0d6 | 1562 | low_limit = info->low_limit; |
6b008640 LT |
1563 | if (low_limit < mmap_min_addr) |
1564 | low_limit = mmap_min_addr; | |
1565 | high_limit = info->high_limit; | |
58c5d0d6 | 1566 | retry: |
6b008640 | 1567 | if (mas_empty_area(&mas, low_limit, high_limit - 1, length)) |
db4fbfb9 ML |
1568 | return -ENOMEM; |
1569 | ||
3499a131 LH |
1570 | gap = mas.index; |
1571 | gap += (info->align_offset - gap) & info->align_mask; | |
58c5d0d6 LH |
1572 | tmp = mas_next(&mas, ULONG_MAX); |
1573 | if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ | |
1574 | if (vm_start_gap(tmp) < gap + length - 1) { | |
1575 | low_limit = tmp->vm_end; | |
1576 | mas_reset(&mas); | |
1577 | goto retry; | |
1578 | } | |
1579 | } else { | |
1580 | tmp = mas_prev(&mas, 0); | |
1581 | if (tmp && vm_end_gap(tmp) > gap) { | |
1582 | low_limit = vm_end_gap(tmp); | |
1583 | mas_reset(&mas); | |
1584 | goto retry; | |
1585 | } | |
1586 | } | |
1587 | ||
3499a131 | 1588 | return gap; |
db4fbfb9 ML |
1589 | } |
1590 | ||
3499a131 LH |
1591 | /** |
1592 | * unmapped_area_topdown() - Find an area between the low_limit and the | |
82b24936 | 1593 | * high_limit with the correct alignment and offset at the highest available |
3499a131 LH |
1594 | * address, all from @info. Note: current->mm is used for the search. |
1595 | * | |
82b24936 VY |
1596 | * @info: The unmapped area information including the range [low_limit - |
1597 | * high_limit), the alignment offset and mask. | |
3499a131 LH |
1598 | * |
1599 | * Return: A memory address or -ENOMEM. | |
1600 | */ | |
baceaf1c | 1601 | static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) |
db4fbfb9 | 1602 | { |
6b008640 LT |
1603 | unsigned long length, gap, gap_end; |
1604 | unsigned long low_limit, high_limit; | |
58c5d0d6 | 1605 | struct vm_area_struct *tmp; |
db4fbfb9 | 1606 | |
3499a131 | 1607 | MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); |
db4fbfb9 ML |
1608 | /* Adjust search length to account for worst case alignment overhead */ |
1609 | length = info->length + info->align_mask; | |
1610 | if (length < info->length) | |
1611 | return -ENOMEM; | |
1612 | ||
6b008640 LT |
1613 | low_limit = info->low_limit; |
1614 | if (low_limit < mmap_min_addr) | |
1615 | low_limit = mmap_min_addr; | |
58c5d0d6 LH |
1616 | high_limit = info->high_limit; |
1617 | retry: | |
6b008640 | 1618 | if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length)) |
db4fbfb9 | 1619 | return -ENOMEM; |
db4fbfb9 | 1620 | |
3499a131 LH |
1621 | gap = mas.last + 1 - info->length; |
1622 | gap -= (gap - info->align_offset) & info->align_mask; | |
58c5d0d6 LH |
1623 | gap_end = mas.last; |
1624 | tmp = mas_next(&mas, ULONG_MAX); | |
1625 | if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ | |
1626 | if (vm_start_gap(tmp) <= gap_end) { | |
1627 | high_limit = vm_start_gap(tmp); | |
1628 | mas_reset(&mas); | |
1629 | goto retry; | |
1630 | } | |
1631 | } else { | |
1632 | tmp = mas_prev(&mas, 0); | |
1633 | if (tmp && vm_end_gap(tmp) > gap) { | |
1634 | high_limit = tmp->vm_start; | |
1635 | mas_reset(&mas); | |
1636 | goto retry; | |
1637 | } | |
1638 | } | |
1639 | ||
3499a131 | 1640 | return gap; |
db4fbfb9 ML |
1641 | } |
1642 | ||
baceaf1c JK |
1643 | /* |
1644 | * Search for an unmapped address range. | |
1645 | * | |
1646 | * We are looking for a range that: | |
1647 | * - does not intersect with any VMA; | |
1648 | * - is contained within the [low_limit, high_limit) interval; | |
1649 | * - is at least the desired size. | |
1650 | * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) | |
1651 | */ | |
1652 | unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) | |
1653 | { | |
df529cab JK |
1654 | unsigned long addr; |
1655 | ||
baceaf1c | 1656 | if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) |
df529cab | 1657 | addr = unmapped_area_topdown(info); |
baceaf1c | 1658 | else |
df529cab JK |
1659 | addr = unmapped_area(info); |
1660 | ||
1661 | trace_vm_unmapped_area(addr, info); | |
1662 | return addr; | |
baceaf1c | 1663 | } |
f6795053 | 1664 | |
1da177e4 LT |
1665 | /* Get an address range which is currently unmapped. |
1666 | * For shmat() with addr=0. | |
1667 | * | |
1668 | * Ugly calling convention alert: | |
1669 | * Return value with the low bits set means error value, | |
1670 | * ie | |
1671 | * if (ret & ~PAGE_MASK) | |
1672 | * error = ret; | |
1673 | * | |
1674 | * This function "knows" that -ENOMEM has the bits set. | |
1675 | */ | |
1da177e4 | 1676 | unsigned long |
4b439e25 CL |
1677 | generic_get_unmapped_area(struct file *filp, unsigned long addr, |
1678 | unsigned long len, unsigned long pgoff, | |
1679 | unsigned long flags) | |
1da177e4 LT |
1680 | { |
1681 | struct mm_struct *mm = current->mm; | |
1be7107f | 1682 | struct vm_area_struct *vma, *prev; |
db4fbfb9 | 1683 | struct vm_unmapped_area_info info; |
2cb4de08 | 1684 | const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); |
1da177e4 | 1685 | |
f6795053 | 1686 | if (len > mmap_end - mmap_min_addr) |
1da177e4 LT |
1687 | return -ENOMEM; |
1688 | ||
06abdfb4 BH |
1689 | if (flags & MAP_FIXED) |
1690 | return addr; | |
1691 | ||
1da177e4 LT |
1692 | if (addr) { |
1693 | addr = PAGE_ALIGN(addr); | |
1be7107f | 1694 | vma = find_vma_prev(mm, addr, &prev); |
f6795053 | 1695 | if (mmap_end - len >= addr && addr >= mmap_min_addr && |
1be7107f HD |
1696 | (!vma || addr + len <= vm_start_gap(vma)) && |
1697 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
1698 | return addr; |
1699 | } | |
1da177e4 | 1700 | |
db4fbfb9 ML |
1701 | info.flags = 0; |
1702 | info.length = len; | |
4e99b021 | 1703 | info.low_limit = mm->mmap_base; |
f6795053 | 1704 | info.high_limit = mmap_end; |
db4fbfb9 | 1705 | info.align_mask = 0; |
09ef5283 | 1706 | info.align_offset = 0; |
db4fbfb9 | 1707 | return vm_unmapped_area(&info); |
1da177e4 | 1708 | } |
4b439e25 CL |
1709 | |
1710 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1711 | unsigned long | |
1712 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1713 | unsigned long len, unsigned long pgoff, | |
1714 | unsigned long flags) | |
1715 | { | |
1716 | return generic_get_unmapped_area(filp, addr, len, pgoff, flags); | |
1717 | } | |
cc71aba3 | 1718 | #endif |
1da177e4 | 1719 | |
1da177e4 LT |
1720 | /* |
1721 | * This mmap-allocator allocates new areas top-down from below the | |
1722 | * stack's low limit (the base): | |
1723 | */ | |
1da177e4 | 1724 | unsigned long |
4b439e25 CL |
1725 | generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, |
1726 | unsigned long len, unsigned long pgoff, | |
1727 | unsigned long flags) | |
1da177e4 | 1728 | { |
1be7107f | 1729 | struct vm_area_struct *vma, *prev; |
1da177e4 | 1730 | struct mm_struct *mm = current->mm; |
db4fbfb9 | 1731 | struct vm_unmapped_area_info info; |
2cb4de08 | 1732 | const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); |
1da177e4 LT |
1733 | |
1734 | /* requested length too big for entire address space */ | |
f6795053 | 1735 | if (len > mmap_end - mmap_min_addr) |
1da177e4 LT |
1736 | return -ENOMEM; |
1737 | ||
06abdfb4 BH |
1738 | if (flags & MAP_FIXED) |
1739 | return addr; | |
1740 | ||
1da177e4 LT |
1741 | /* requesting a specific address */ |
1742 | if (addr) { | |
1743 | addr = PAGE_ALIGN(addr); | |
1be7107f | 1744 | vma = find_vma_prev(mm, addr, &prev); |
f6795053 | 1745 | if (mmap_end - len >= addr && addr >= mmap_min_addr && |
1be7107f HD |
1746 | (!vma || addr + len <= vm_start_gap(vma)) && |
1747 | (!prev || addr >= vm_end_gap(prev))) | |
1da177e4 LT |
1748 | return addr; |
1749 | } | |
1750 | ||
db4fbfb9 ML |
1751 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
1752 | info.length = len; | |
6b008640 | 1753 | info.low_limit = PAGE_SIZE; |
f6795053 | 1754 | info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); |
db4fbfb9 | 1755 | info.align_mask = 0; |
09ef5283 | 1756 | info.align_offset = 0; |
db4fbfb9 | 1757 | addr = vm_unmapped_area(&info); |
b716ad95 | 1758 | |
1da177e4 LT |
1759 | /* |
1760 | * A failed mmap() very likely causes application failure, | |
1761 | * so fall back to the bottom-up function here. This scenario | |
1762 | * can happen with large stack limits and large mmap() | |
1763 | * allocations. | |
1764 | */ | |
de1741a1 | 1765 | if (offset_in_page(addr)) { |
db4fbfb9 ML |
1766 | VM_BUG_ON(addr != -ENOMEM); |
1767 | info.flags = 0; | |
1768 | info.low_limit = TASK_UNMAPPED_BASE; | |
f6795053 | 1769 | info.high_limit = mmap_end; |
db4fbfb9 ML |
1770 | addr = vm_unmapped_area(&info); |
1771 | } | |
1da177e4 LT |
1772 | |
1773 | return addr; | |
1774 | } | |
4b439e25 CL |
1775 | |
1776 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1777 | unsigned long | |
1778 | arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, | |
1779 | unsigned long len, unsigned long pgoff, | |
1780 | unsigned long flags) | |
1781 | { | |
1782 | return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); | |
1783 | } | |
1da177e4 LT |
1784 | #endif |
1785 | ||
1da177e4 LT |
1786 | unsigned long |
1787 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1788 | unsigned long pgoff, unsigned long flags) | |
1789 | { | |
06abdfb4 BH |
1790 | unsigned long (*get_area)(struct file *, unsigned long, |
1791 | unsigned long, unsigned long, unsigned long); | |
1792 | ||
9206de95 AV |
1793 | unsigned long error = arch_mmap_check(addr, len, flags); |
1794 | if (error) | |
1795 | return error; | |
1796 | ||
1797 | /* Careful about overflows.. */ | |
1798 | if (len > TASK_SIZE) | |
1799 | return -ENOMEM; | |
1800 | ||
06abdfb4 | 1801 | get_area = current->mm->get_unmapped_area; |
c01d5b30 HD |
1802 | if (file) { |
1803 | if (file->f_op->get_unmapped_area) | |
1804 | get_area = file->f_op->get_unmapped_area; | |
1805 | } else if (flags & MAP_SHARED) { | |
1806 | /* | |
1807 | * mmap_region() will call shmem_zero_setup() to create a file, | |
1808 | * so use shmem's get_unmapped_area in case it can be huge. | |
45e55300 | 1809 | * do_mmap() will clear pgoff, so match alignment. |
c01d5b30 HD |
1810 | */ |
1811 | pgoff = 0; | |
1812 | get_area = shmem_get_unmapped_area; | |
1813 | } | |
1814 | ||
06abdfb4 BH |
1815 | addr = get_area(file, addr, len, pgoff, flags); |
1816 | if (IS_ERR_VALUE(addr)) | |
1817 | return addr; | |
1da177e4 | 1818 | |
07ab67c8 LT |
1819 | if (addr > TASK_SIZE - len) |
1820 | return -ENOMEM; | |
de1741a1 | 1821 | if (offset_in_page(addr)) |
07ab67c8 | 1822 | return -EINVAL; |
06abdfb4 | 1823 | |
9ac4ed4b AV |
1824 | error = security_mmap_addr(addr); |
1825 | return error ? error : addr; | |
1da177e4 LT |
1826 | } |
1827 | ||
1828 | EXPORT_SYMBOL(get_unmapped_area); | |
1829 | ||
abdba2dd LH |
1830 | /** |
1831 | * find_vma_intersection() - Look up the first VMA which intersects the interval | |
1832 | * @mm: The process address space. | |
1833 | * @start_addr: The inclusive start user address. | |
1834 | * @end_addr: The exclusive end user address. | |
1835 | * | |
1836 | * Returns: The first VMA within the provided range, %NULL otherwise. Assumes | |
1837 | * start_addr < end_addr. | |
1838 | */ | |
1839 | struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, | |
1840 | unsigned long start_addr, | |
1841 | unsigned long end_addr) | |
1842 | { | |
abdba2dd LH |
1843 | unsigned long index = start_addr; |
1844 | ||
1845 | mmap_assert_locked(mm); | |
7964cf8c | 1846 | return mt_find(&mm->mm_mt, &index, end_addr - 1); |
abdba2dd LH |
1847 | } |
1848 | EXPORT_SYMBOL(find_vma_intersection); | |
1849 | ||
be8432e7 LH |
1850 | /** |
1851 | * find_vma() - Find the VMA for a given address, or the next VMA. | |
1852 | * @mm: The mm_struct to check | |
1853 | * @addr: The address | |
1854 | * | |
1855 | * Returns: The VMA associated with addr, or the next VMA. | |
1856 | * May return %NULL in the case of no VMA at addr or above. | |
1857 | */ | |
48aae425 | 1858 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 1859 | { |
be8432e7 | 1860 | unsigned long index = addr; |
1da177e4 | 1861 | |
5b78ed24 | 1862 | mmap_assert_locked(mm); |
7964cf8c | 1863 | return mt_find(&mm->mm_mt, &index, ULONG_MAX); |
1da177e4 | 1864 | } |
1da177e4 LT |
1865 | EXPORT_SYMBOL(find_vma); |
1866 | ||
7fdbd37d LH |
1867 | /** |
1868 | * find_vma_prev() - Find the VMA for a given address, or the next vma and | |
1869 | * set %pprev to the previous VMA, if any. | |
1870 | * @mm: The mm_struct to check | |
1871 | * @addr: The address | |
1872 | * @pprev: The pointer to set to the previous VMA | |
1873 | * | |
1874 | * Note that RCU lock is missing here since the external mmap_lock() is used | |
1875 | * instead. | |
1876 | * | |
1877 | * Returns: The VMA associated with @addr, or the next vma. | |
1878 | * May return %NULL in the case of no vma at addr or above. | |
6bd4837d | 1879 | */ |
1da177e4 LT |
1880 | struct vm_area_struct * |
1881 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
1882 | struct vm_area_struct **pprev) | |
1883 | { | |
6bd4837d | 1884 | struct vm_area_struct *vma; |
7fdbd37d | 1885 | MA_STATE(mas, &mm->mm_mt, addr, addr); |
1da177e4 | 1886 | |
7fdbd37d LH |
1887 | vma = mas_walk(&mas); |
1888 | *pprev = mas_prev(&mas, 0); | |
1889 | if (!vma) | |
1890 | vma = mas_next(&mas, ULONG_MAX); | |
6bd4837d | 1891 | return vma; |
1da177e4 LT |
1892 | } |
1893 | ||
1894 | /* | |
1895 | * Verify that the stack growth is acceptable and | |
1896 | * update accounting. This is shared with both the | |
1897 | * grow-up and grow-down cases. | |
1898 | */ | |
1be7107f HD |
1899 | static int acct_stack_growth(struct vm_area_struct *vma, |
1900 | unsigned long size, unsigned long grow) | |
1da177e4 LT |
1901 | { |
1902 | struct mm_struct *mm = vma->vm_mm; | |
1be7107f | 1903 | unsigned long new_start; |
1da177e4 LT |
1904 | |
1905 | /* address space limit tests */ | |
84638335 | 1906 | if (!may_expand_vm(mm, vma->vm_flags, grow)) |
1da177e4 LT |
1907 | return -ENOMEM; |
1908 | ||
1909 | /* Stack limit test */ | |
24c79d8e | 1910 | if (size > rlimit(RLIMIT_STACK)) |
1da177e4 LT |
1911 | return -ENOMEM; |
1912 | ||
1913 | /* mlock limit tests */ | |
b0cc5e89 | 1914 | if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT)) |
c5d8a364 | 1915 | return -ENOMEM; |
1da177e4 | 1916 | |
0d59a01b AL |
1917 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
1918 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
1919 | vma->vm_end - size; | |
1920 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
1921 | return -EFAULT; | |
1922 | ||
1da177e4 LT |
1923 | /* |
1924 | * Overcommit.. This must be the final test, as it will | |
1925 | * update security statistics. | |
1926 | */ | |
05fa199d | 1927 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
1928 | return -ENOMEM; |
1929 | ||
1da177e4 LT |
1930 | return 0; |
1931 | } | |
1932 | ||
46dea3d0 | 1933 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 1934 | /* |
46dea3d0 HD |
1935 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
1936 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 1937 | */ |
8d7071af | 1938 | static int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 1939 | { |
09357814 | 1940 | struct mm_struct *mm = vma->vm_mm; |
1be7107f HD |
1941 | struct vm_area_struct *next; |
1942 | unsigned long gap_addr; | |
12352d3c | 1943 | int error = 0; |
d4af56c5 | 1944 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
1da177e4 LT |
1945 | |
1946 | if (!(vma->vm_flags & VM_GROWSUP)) | |
1947 | return -EFAULT; | |
1948 | ||
bd726c90 | 1949 | /* Guard against exceeding limits of the address space. */ |
1be7107f | 1950 | address &= PAGE_MASK; |
37511fb5 | 1951 | if (address >= (TASK_SIZE & PAGE_MASK)) |
12352d3c | 1952 | return -ENOMEM; |
bd726c90 | 1953 | address += PAGE_SIZE; |
12352d3c | 1954 | |
1be7107f HD |
1955 | /* Enforce stack_guard_gap */ |
1956 | gap_addr = address + stack_guard_gap; | |
bd726c90 HD |
1957 | |
1958 | /* Guard against overflow */ | |
1959 | if (gap_addr < address || gap_addr > TASK_SIZE) | |
1960 | gap_addr = TASK_SIZE; | |
1961 | ||
763ecb03 LH |
1962 | next = find_vma_intersection(mm, vma->vm_end, gap_addr); |
1963 | if (next && vma_is_accessible(next)) { | |
1be7107f HD |
1964 | if (!(next->vm_flags & VM_GROWSUP)) |
1965 | return -ENOMEM; | |
1966 | /* Check that both stack segments have the same anon_vma? */ | |
1967 | } | |
1968 | ||
c5d5546e | 1969 | if (mas_preallocate(&mas, GFP_KERNEL)) |
d4af56c5 LH |
1970 | return -ENOMEM; |
1971 | ||
12352d3c | 1972 | /* We must make sure the anon_vma is allocated. */ |
d4af56c5 LH |
1973 | if (unlikely(anon_vma_prepare(vma))) { |
1974 | mas_destroy(&mas); | |
1da177e4 | 1975 | return -ENOMEM; |
d4af56c5 | 1976 | } |
1da177e4 | 1977 | |
c137381f SB |
1978 | /* Lock the VMA before expanding to prevent concurrent page faults */ |
1979 | vma_start_write(vma); | |
1da177e4 LT |
1980 | /* |
1981 | * vma->vm_start/vm_end cannot change under us because the caller | |
c1e8d7c6 | 1982 | * is required to hold the mmap_lock in read mode. We need the |
1da177e4 LT |
1983 | * anon_vma lock to serialize against concurrent expand_stacks. |
1984 | */ | |
12352d3c | 1985 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
1986 | |
1987 | /* Somebody else might have raced and expanded it already */ | |
1988 | if (address > vma->vm_end) { | |
1989 | unsigned long size, grow; | |
1990 | ||
1991 | size = address - vma->vm_start; | |
1992 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
1993 | ||
42c36f63 HD |
1994 | error = -ENOMEM; |
1995 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
1996 | error = acct_stack_growth(vma, size, grow); | |
1997 | if (!error) { | |
4128997b | 1998 | /* |
524e00b3 LH |
1999 | * We only hold a shared mmap_lock lock here, so |
2000 | * we need to protect against concurrent vma | |
2001 | * expansions. anon_vma_lock_write() doesn't | |
2002 | * help here, as we don't guarantee that all | |
2003 | * growable vmas in a mm share the same root | |
2004 | * anon vma. So, we reuse mm->page_table_lock | |
2005 | * to guard against concurrent vma expansions. | |
4128997b | 2006 | */ |
09357814 | 2007 | spin_lock(&mm->page_table_lock); |
87e8827b | 2008 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2009 | mm->locked_vm += grow; |
84638335 | 2010 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2011 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 2012 | vma->vm_end = address; |
d4af56c5 | 2013 | /* Overwrite old entry in mtree. */ |
fbcc3104 LH |
2014 | mas_set_range(&mas, vma->vm_start, address - 1); |
2015 | mas_store_prealloc(&mas, vma); | |
bf181b9f | 2016 | anon_vma_interval_tree_post_update_vma(vma); |
09357814 | 2017 | spin_unlock(&mm->page_table_lock); |
4128997b | 2018 | |
42c36f63 HD |
2019 | perf_event_mmap(vma); |
2020 | } | |
3af9e859 | 2021 | } |
1da177e4 | 2022 | } |
12352d3c | 2023 | anon_vma_unlock_write(vma->anon_vma); |
c791576c | 2024 | khugepaged_enter_vma(vma, vma->vm_flags); |
d4af56c5 | 2025 | mas_destroy(&mas); |
1da177e4 LT |
2026 | return error; |
2027 | } | |
46dea3d0 HD |
2028 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2029 | ||
1da177e4 LT |
2030 | /* |
2031 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
8d7071af | 2032 | * mmap_lock held for writing. |
1da177e4 | 2033 | */ |
524e00b3 | 2034 | int expand_downwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 | 2035 | { |
09357814 | 2036 | struct mm_struct *mm = vma->vm_mm; |
763ecb03 | 2037 | MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); |
1be7107f | 2038 | struct vm_area_struct *prev; |
0a1d5299 | 2039 | int error = 0; |
1da177e4 | 2040 | |
8d7071af LT |
2041 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
2042 | return -EFAULT; | |
2043 | ||
8869477a | 2044 | address &= PAGE_MASK; |
8b35ca3e | 2045 | if (address < mmap_min_addr || address < FIRST_USER_ADDRESS) |
0a1d5299 | 2046 | return -EPERM; |
8869477a | 2047 | |
1be7107f | 2048 | /* Enforce stack_guard_gap */ |
763ecb03 | 2049 | prev = mas_prev(&mas, 0); |
32e4e6d5 | 2050 | /* Check that both stack segments have the same anon_vma? */ |
f440fa1a LH |
2051 | if (prev) { |
2052 | if (!(prev->vm_flags & VM_GROWSDOWN) && | |
2053 | vma_is_accessible(prev) && | |
2054 | (address - prev->vm_end < stack_guard_gap)) | |
1be7107f | 2055 | return -ENOMEM; |
1be7107f HD |
2056 | } |
2057 | ||
c5d5546e | 2058 | if (mas_preallocate(&mas, GFP_KERNEL)) |
d4af56c5 LH |
2059 | return -ENOMEM; |
2060 | ||
12352d3c | 2061 | /* We must make sure the anon_vma is allocated. */ |
d4af56c5 LH |
2062 | if (unlikely(anon_vma_prepare(vma))) { |
2063 | mas_destroy(&mas); | |
12352d3c | 2064 | return -ENOMEM; |
d4af56c5 | 2065 | } |
1da177e4 | 2066 | |
c137381f SB |
2067 | /* Lock the VMA before expanding to prevent concurrent page faults */ |
2068 | vma_start_write(vma); | |
1da177e4 LT |
2069 | /* |
2070 | * vma->vm_start/vm_end cannot change under us because the caller | |
c1e8d7c6 | 2071 | * is required to hold the mmap_lock in read mode. We need the |
1da177e4 LT |
2072 | * anon_vma lock to serialize against concurrent expand_stacks. |
2073 | */ | |
12352d3c | 2074 | anon_vma_lock_write(vma->anon_vma); |
1da177e4 LT |
2075 | |
2076 | /* Somebody else might have raced and expanded it already */ | |
2077 | if (address < vma->vm_start) { | |
2078 | unsigned long size, grow; | |
2079 | ||
2080 | size = vma->vm_end - address; | |
2081 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2082 | ||
a626ca6a LT |
2083 | error = -ENOMEM; |
2084 | if (grow <= vma->vm_pgoff) { | |
2085 | error = acct_stack_growth(vma, size, grow); | |
2086 | if (!error) { | |
4128997b | 2087 | /* |
524e00b3 LH |
2088 | * We only hold a shared mmap_lock lock here, so |
2089 | * we need to protect against concurrent vma | |
2090 | * expansions. anon_vma_lock_write() doesn't | |
2091 | * help here, as we don't guarantee that all | |
2092 | * growable vmas in a mm share the same root | |
2093 | * anon vma. So, we reuse mm->page_table_lock | |
2094 | * to guard against concurrent vma expansions. | |
4128997b | 2095 | */ |
09357814 | 2096 | spin_lock(&mm->page_table_lock); |
87e8827b | 2097 | if (vma->vm_flags & VM_LOCKED) |
09357814 | 2098 | mm->locked_vm += grow; |
84638335 | 2099 | vm_stat_account(mm, vma->vm_flags, grow); |
bf181b9f | 2100 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2101 | vma->vm_start = address; |
2102 | vma->vm_pgoff -= grow; | |
d4af56c5 | 2103 | /* Overwrite old entry in mtree. */ |
fbcc3104 LH |
2104 | mas_set_range(&mas, address, vma->vm_end - 1); |
2105 | mas_store_prealloc(&mas, vma); | |
bf181b9f | 2106 | anon_vma_interval_tree_post_update_vma(vma); |
09357814 | 2107 | spin_unlock(&mm->page_table_lock); |
4128997b | 2108 | |
a626ca6a LT |
2109 | perf_event_mmap(vma); |
2110 | } | |
1da177e4 LT |
2111 | } |
2112 | } | |
12352d3c | 2113 | anon_vma_unlock_write(vma->anon_vma); |
c791576c | 2114 | khugepaged_enter_vma(vma, vma->vm_flags); |
d4af56c5 | 2115 | mas_destroy(&mas); |
1da177e4 LT |
2116 | return error; |
2117 | } | |
2118 | ||
1be7107f HD |
2119 | /* enforced gap between the expanding stack and other mappings. */ |
2120 | unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; | |
2121 | ||
2122 | static int __init cmdline_parse_stack_guard_gap(char *p) | |
2123 | { | |
2124 | unsigned long val; | |
2125 | char *endptr; | |
2126 | ||
2127 | val = simple_strtoul(p, &endptr, 10); | |
2128 | if (!*endptr) | |
2129 | stack_guard_gap = val << PAGE_SHIFT; | |
2130 | ||
e6d09493 | 2131 | return 1; |
1be7107f HD |
2132 | } |
2133 | __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); | |
2134 | ||
b6a2fea3 | 2135 | #ifdef CONFIG_STACK_GROWSUP |
8d7071af | 2136 | int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) |
b6a2fea3 OW |
2137 | { |
2138 | return expand_upwards(vma, address); | |
2139 | } | |
2140 | ||
8d7071af | 2141 | struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) |
b6a2fea3 OW |
2142 | { |
2143 | struct vm_area_struct *vma, *prev; | |
2144 | ||
2145 | addr &= PAGE_MASK; | |
2146 | vma = find_vma_prev(mm, addr, &prev); | |
2147 | if (vma && (vma->vm_start <= addr)) | |
2148 | return vma; | |
f440fa1a LH |
2149 | if (!prev) |
2150 | return NULL; | |
8d7071af | 2151 | if (expand_stack_locked(prev, addr)) |
b6a2fea3 | 2152 | return NULL; |
cea10a19 | 2153 | if (prev->vm_flags & VM_LOCKED) |
fc05f566 | 2154 | populate_vma_page_range(prev, addr, prev->vm_end, NULL); |
b6a2fea3 OW |
2155 | return prev; |
2156 | } | |
2157 | #else | |
8d7071af | 2158 | int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) |
b6a2fea3 | 2159 | { |
f440fa1a LH |
2160 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) |
2161 | return -EINVAL; | |
b6a2fea3 OW |
2162 | return expand_downwards(vma, address); |
2163 | } | |
2164 | ||
8d7071af | 2165 | struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2166 | { |
cc71aba3 | 2167 | struct vm_area_struct *vma; |
1da177e4 LT |
2168 | unsigned long start; |
2169 | ||
2170 | addr &= PAGE_MASK; | |
cc71aba3 | 2171 | vma = find_vma(mm, addr); |
1da177e4 LT |
2172 | if (!vma) |
2173 | return NULL; | |
2174 | if (vma->vm_start <= addr) | |
2175 | return vma; | |
1da177e4 | 2176 | start = vma->vm_start; |
8d7071af | 2177 | if (expand_stack_locked(vma, addr)) |
1da177e4 | 2178 | return NULL; |
cea10a19 | 2179 | if (vma->vm_flags & VM_LOCKED) |
fc05f566 | 2180 | populate_vma_page_range(vma, addr, start, NULL); |
1da177e4 LT |
2181 | return vma; |
2182 | } | |
2183 | #endif | |
2184 | ||
8d7071af LT |
2185 | /* |
2186 | * IA64 has some horrid mapping rules: it can expand both up and down, | |
2187 | * but with various special rules. | |
2188 | * | |
2189 | * We'll get rid of this architecture eventually, so the ugliness is | |
2190 | * temporary. | |
2191 | */ | |
2192 | #ifdef CONFIG_IA64 | |
2193 | static inline bool vma_expand_ok(struct vm_area_struct *vma, unsigned long addr) | |
2194 | { | |
2195 | return REGION_NUMBER(addr) == REGION_NUMBER(vma->vm_start) && | |
2196 | REGION_OFFSET(addr) < RGN_MAP_LIMIT; | |
2197 | } | |
2198 | ||
2199 | /* | |
2200 | * IA64 stacks grow down, but there's a special register backing store | |
2201 | * that can grow up. Only sequentially, though, so the new address must | |
2202 | * match vm_end. | |
2203 | */ | |
2204 | static inline int vma_expand_up(struct vm_area_struct *vma, unsigned long addr) | |
2205 | { | |
2206 | if (!vma_expand_ok(vma, addr)) | |
2207 | return -EFAULT; | |
2208 | if (vma->vm_end != (addr & PAGE_MASK)) | |
2209 | return -EFAULT; | |
2210 | return expand_upwards(vma, addr); | |
2211 | } | |
2212 | ||
2213 | static inline bool vma_expand_down(struct vm_area_struct *vma, unsigned long addr) | |
2214 | { | |
2215 | if (!vma_expand_ok(vma, addr)) | |
2216 | return -EFAULT; | |
2217 | return expand_downwards(vma, addr); | |
2218 | } | |
2219 | ||
2220 | #elif defined(CONFIG_STACK_GROWSUP) | |
2221 | ||
2222 | #define vma_expand_up(vma,addr) expand_upwards(vma, addr) | |
2223 | #define vma_expand_down(vma, addr) (-EFAULT) | |
2224 | ||
2225 | #else | |
2226 | ||
2227 | #define vma_expand_up(vma,addr) (-EFAULT) | |
2228 | #define vma_expand_down(vma, addr) expand_downwards(vma, addr) | |
2229 | ||
2230 | #endif | |
2231 | ||
2232 | /* | |
2233 | * expand_stack(): legacy interface for page faulting. Don't use unless | |
2234 | * you have to. | |
2235 | * | |
2236 | * This is called with the mm locked for reading, drops the lock, takes | |
2237 | * the lock for writing, tries to look up a vma again, expands it if | |
2238 | * necessary, and downgrades the lock to reading again. | |
2239 | * | |
2240 | * If no vma is found or it can't be expanded, it returns NULL and has | |
2241 | * dropped the lock. | |
2242 | */ | |
2243 | struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr) | |
f440fa1a | 2244 | { |
8d7071af LT |
2245 | struct vm_area_struct *vma, *prev; |
2246 | ||
2247 | mmap_read_unlock(mm); | |
2248 | if (mmap_write_lock_killable(mm)) | |
2249 | return NULL; | |
2250 | ||
2251 | vma = find_vma_prev(mm, addr, &prev); | |
2252 | if (vma && vma->vm_start <= addr) | |
2253 | goto success; | |
2254 | ||
2255 | if (prev && !vma_expand_up(prev, addr)) { | |
2256 | vma = prev; | |
2257 | goto success; | |
2258 | } | |
2259 | ||
2260 | if (vma && !vma_expand_down(vma, addr)) | |
2261 | goto success; | |
2262 | ||
2263 | mmap_write_unlock(mm); | |
2264 | return NULL; | |
2265 | ||
2266 | success: | |
2267 | mmap_write_downgrade(mm); | |
2268 | return vma; | |
f440fa1a | 2269 | } |
e1d6d01a | 2270 | |
1da177e4 | 2271 | /* |
763ecb03 LH |
2272 | * Ok - we have the memory areas we should free on a maple tree so release them, |
2273 | * and do the vma updates. | |
2c0b3814 HD |
2274 | * |
2275 | * Called with the mm semaphore held. | |
1da177e4 | 2276 | */ |
763ecb03 | 2277 | static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) |
1da177e4 | 2278 | { |
4f74d2c8 | 2279 | unsigned long nr_accounted = 0; |
763ecb03 | 2280 | struct vm_area_struct *vma; |
4f74d2c8 | 2281 | |
365e9c87 HD |
2282 | /* Update high watermark before we lower total_vm */ |
2283 | update_hiwater_vm(mm); | |
763ecb03 | 2284 | mas_for_each(mas, vma, ULONG_MAX) { |
2c0b3814 HD |
2285 | long nrpages = vma_pages(vma); |
2286 | ||
4f74d2c8 LT |
2287 | if (vma->vm_flags & VM_ACCOUNT) |
2288 | nr_accounted += nrpages; | |
84638335 | 2289 | vm_stat_account(mm, vma->vm_flags, -nrpages); |
0d2ebf9c | 2290 | remove_vma(vma, false); |
763ecb03 | 2291 | } |
4f74d2c8 | 2292 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2293 | validate_mm(mm); |
2294 | } | |
2295 | ||
2296 | /* | |
2297 | * Get rid of page table information in the indicated region. | |
2298 | * | |
f10df686 | 2299 | * Called with the mm semaphore held. |
1da177e4 | 2300 | */ |
763ecb03 | 2301 | static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, |
e0da382c | 2302 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
763ecb03 | 2303 | struct vm_area_struct *next, |
68f48381 | 2304 | unsigned long start, unsigned long end, bool mm_wr_locked) |
1da177e4 | 2305 | { |
d16dfc55 | 2306 | struct mmu_gather tlb; |
1da177e4 LT |
2307 | |
2308 | lru_add_drain(); | |
a72afd87 | 2309 | tlb_gather_mmu(&tlb, mm); |
365e9c87 | 2310 | update_hiwater_rss(mm); |
68f48381 | 2311 | unmap_vmas(&tlb, mt, vma, start, end, mm_wr_locked); |
763ecb03 | 2312 | free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
98e51a22 SB |
2313 | next ? next->vm_start : USER_PGTABLES_CEILING, |
2314 | mm_wr_locked); | |
ae8eba8b | 2315 | tlb_finish_mmu(&tlb); |
1da177e4 LT |
2316 | } |
2317 | ||
1da177e4 | 2318 | /* |
def5efe0 DR |
2319 | * __split_vma() bypasses sysctl_max_map_count checking. We use this where it |
2320 | * has already been checked or doesn't make sense to fail. | |
0fd5a9e2 | 2321 | * VMA Iterator will point to the end VMA. |
1da177e4 | 2322 | */ |
9760ebff | 2323 | int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, |
def5efe0 | 2324 | unsigned long addr, int new_below) |
1da177e4 | 2325 | { |
b2b3b886 | 2326 | struct vma_prepare vp; |
1da177e4 | 2327 | struct vm_area_struct *new; |
e3975891 | 2328 | int err; |
9760ebff | 2329 | |
b50e195f | 2330 | validate_mm(vma->vm_mm); |
1da177e4 | 2331 | |
b2b3b886 LH |
2332 | WARN_ON(vma->vm_start >= addr); |
2333 | WARN_ON(vma->vm_end <= addr); | |
2334 | ||
dd3b614f DS |
2335 | if (vma->vm_ops && vma->vm_ops->may_split) { |
2336 | err = vma->vm_ops->may_split(vma, addr); | |
31383c68 DW |
2337 | if (err) |
2338 | return err; | |
2339 | } | |
1da177e4 | 2340 | |
3928d4f5 | 2341 | new = vm_area_dup(vma); |
1da177e4 | 2342 | if (!new) |
e3975891 | 2343 | return -ENOMEM; |
1da177e4 | 2344 | |
b2b3b886 LH |
2345 | err = -ENOMEM; |
2346 | if (vma_iter_prealloc(vmi)) | |
2347 | goto out_free_vma; | |
2348 | ||
2349 | if (new_below) { | |
1da177e4 | 2350 | new->vm_end = addr; |
b2b3b886 | 2351 | } else { |
1da177e4 LT |
2352 | new->vm_start = addr; |
2353 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2354 | } | |
2355 | ||
ef0855d3 ON |
2356 | err = vma_dup_policy(vma, new); |
2357 | if (err) | |
b2b3b886 | 2358 | goto out_free_vmi; |
1da177e4 | 2359 | |
c4ea95d7 DF |
2360 | err = anon_vma_clone(new, vma); |
2361 | if (err) | |
5beb4930 RR |
2362 | goto out_free_mpol; |
2363 | ||
e9714acf | 2364 | if (new->vm_file) |
1da177e4 LT |
2365 | get_file(new->vm_file); |
2366 | ||
2367 | if (new->vm_ops && new->vm_ops->open) | |
2368 | new->vm_ops->open(new); | |
2369 | ||
b2b3b886 LH |
2370 | init_vma_prep(&vp, vma); |
2371 | vp.insert = new; | |
2372 | vma_prepare(&vp); | |
ccf1d78d | 2373 | vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); |
1da177e4 | 2374 | |
b2b3b886 LH |
2375 | if (new_below) { |
2376 | vma->vm_start = addr; | |
2377 | vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; | |
2378 | } else { | |
2379 | vma->vm_end = addr; | |
9760ebff | 2380 | } |
5beb4930 | 2381 | |
b2b3b886 LH |
2382 | /* vma_complete stores the new vma */ |
2383 | vma_complete(&vp, vmi, vma->vm_mm); | |
2384 | ||
2385 | /* Success. */ | |
2386 | if (new_below) | |
2387 | vma_next(vmi); | |
b50e195f | 2388 | validate_mm(vma->vm_mm); |
b2b3b886 LH |
2389 | return 0; |
2390 | ||
2391 | out_free_mpol: | |
ef0855d3 | 2392 | mpol_put(vma_policy(new)); |
b2b3b886 LH |
2393 | out_free_vmi: |
2394 | vma_iter_free(vmi); | |
2395 | out_free_vma: | |
3928d4f5 | 2396 | vm_area_free(new); |
b50e195f | 2397 | validate_mm(vma->vm_mm); |
5beb4930 | 2398 | return err; |
1da177e4 LT |
2399 | } |
2400 | ||
659ace58 KM |
2401 | /* |
2402 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2403 | * either for the first part or the tail. | |
2404 | */ | |
9760ebff | 2405 | int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, |
659ace58 KM |
2406 | unsigned long addr, int new_below) |
2407 | { | |
9760ebff | 2408 | if (vma->vm_mm->map_count >= sysctl_max_map_count) |
659ace58 KM |
2409 | return -ENOMEM; |
2410 | ||
9760ebff | 2411 | return __split_vma(vmi, vma, addr, new_below); |
f2ebfe43 LH |
2412 | } |
2413 | ||
11f9a21a | 2414 | /* |
183654ce LH |
2415 | * do_vmi_align_munmap() - munmap the aligned region from @start to @end. |
2416 | * @vmi: The vma iterator | |
11f9a21a LH |
2417 | * @vma: The starting vm_area_struct |
2418 | * @mm: The mm_struct | |
2419 | * @start: The aligned start address to munmap. | |
2420 | * @end: The aligned end address to munmap. | |
2421 | * @uf: The userfaultfd list_head | |
408579cd LH |
2422 | * @unlock: Set to true to drop the mmap_lock. unlocking only happens on |
2423 | * success. | |
11f9a21a | 2424 | * |
408579cd LH |
2425 | * Return: 0 on success and drops the lock if so directed, error and leaves the |
2426 | * lock held otherwise. | |
11f9a21a LH |
2427 | */ |
2428 | static int | |
183654ce | 2429 | do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, |
11f9a21a | 2430 | struct mm_struct *mm, unsigned long start, |
408579cd | 2431 | unsigned long end, struct list_head *uf, bool unlock) |
11f9a21a | 2432 | { |
763ecb03 LH |
2433 | struct vm_area_struct *prev, *next = NULL; |
2434 | struct maple_tree mt_detach; | |
2435 | int count = 0; | |
11f9a21a | 2436 | int error = -ENOMEM; |
606c812e | 2437 | unsigned long locked_vm = 0; |
763ecb03 | 2438 | MA_STATE(mas_detach, &mt_detach, 0, 0); |
3dd44325 | 2439 | mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK); |
763ecb03 | 2440 | mt_set_external_lock(&mt_detach, &mm->mmap_lock); |
d4af56c5 | 2441 | |
1da177e4 LT |
2442 | /* |
2443 | * If we need to split any vma, do it now to save pain later. | |
2444 | * | |
2445 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2446 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2447 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2448 | */ | |
763ecb03 LH |
2449 | |
2450 | /* Does it split the first one? */ | |
146425a3 | 2451 | if (start > vma->vm_start) { |
659ace58 KM |
2452 | |
2453 | /* | |
2454 | * Make sure that map_count on return from munmap() will | |
2455 | * not exceed its limit; but let map_count go just above | |
2456 | * its limit temporarily, to help free resources as expected. | |
2457 | */ | |
2458 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
d4af56c5 | 2459 | goto map_count_exceeded; |
659ace58 | 2460 | |
9760ebff | 2461 | error = __split_vma(vmi, vma, start, 0); |
1da177e4 | 2462 | if (error) |
763ecb03 | 2463 | goto start_split_failed; |
11f9a21a | 2464 | |
0fd5a9e2 | 2465 | vma = vma_iter_load(vmi); |
1da177e4 LT |
2466 | } |
2467 | ||
183654ce | 2468 | prev = vma_prev(vmi); |
763ecb03 | 2469 | if (unlikely((!prev))) |
183654ce | 2470 | vma_iter_set(vmi, start); |
763ecb03 LH |
2471 | |
2472 | /* | |
2473 | * Detach a range of VMAs from the mm. Using next as a temp variable as | |
2474 | * it is always overwritten. | |
2475 | */ | |
183654ce | 2476 | for_each_vma_range(*vmi, next, end) { |
763ecb03 LH |
2477 | /* Does it split the end? */ |
2478 | if (next->vm_end > end) { | |
6b73cff2 | 2479 | error = __split_vma(vmi, next, end, 0); |
763ecb03 LH |
2480 | if (error) |
2481 | goto end_split_failed; | |
763ecb03 | 2482 | } |
606c812e LH |
2483 | vma_start_write(next); |
2484 | mas_set_range(&mas_detach, next->vm_start, next->vm_end - 1); | |
6c26bd43 DW |
2485 | error = mas_store_gfp(&mas_detach, next, GFP_KERNEL); |
2486 | if (error) | |
606c812e LH |
2487 | goto munmap_gather_failed; |
2488 | vma_mark_detached(next, true); | |
2489 | if (next->vm_flags & VM_LOCKED) | |
2490 | locked_vm += vma_pages(next); | |
11f9a21a | 2491 | |
763ecb03 | 2492 | count++; |
65ac1320 LH |
2493 | if (unlikely(uf)) { |
2494 | /* | |
2495 | * If userfaultfd_unmap_prep returns an error the vmas | |
2496 | * will remain split, but userland will get a | |
2497 | * highly unexpected error anyway. This is no | |
2498 | * different than the case where the first of the two | |
2499 | * __split_vma fails, but we don't undo the first | |
2500 | * split, despite we could. This is unlikely enough | |
2501 | * failure that it's not worth optimizing it for. | |
2502 | */ | |
2503 | error = userfaultfd_unmap_prep(next, start, end, uf); | |
2504 | ||
2505 | if (error) | |
2506 | goto userfaultfd_error; | |
2507 | } | |
763ecb03 LH |
2508 | #ifdef CONFIG_DEBUG_VM_MAPLE_TREE |
2509 | BUG_ON(next->vm_start < start); | |
2510 | BUG_ON(next->vm_start > end); | |
2511 | #endif | |
1da177e4 | 2512 | } |
1da177e4 | 2513 | |
15c0c60b LH |
2514 | if (vma_iter_end(vmi) > end) |
2515 | next = vma_iter_load(vmi); | |
11f9a21a | 2516 | |
15c0c60b LH |
2517 | if (!next) |
2518 | next = vma_next(vmi); | |
2376dd7c | 2519 | |
763ecb03 LH |
2520 | #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) |
2521 | /* Make sure no VMAs are about to be lost. */ | |
2522 | { | |
2523 | MA_STATE(test, &mt_detach, start, end - 1); | |
2524 | struct vm_area_struct *vma_mas, *vma_test; | |
2525 | int test_count = 0; | |
2526 | ||
183654ce | 2527 | vma_iter_set(vmi, start); |
763ecb03 LH |
2528 | rcu_read_lock(); |
2529 | vma_test = mas_find(&test, end - 1); | |
183654ce | 2530 | for_each_vma_range(*vmi, vma_mas, end) { |
763ecb03 LH |
2531 | BUG_ON(vma_mas != vma_test); |
2532 | test_count++; | |
2533 | vma_test = mas_next(&test, end - 1); | |
2534 | } | |
2535 | rcu_read_unlock(); | |
2536 | BUG_ON(count != test_count); | |
763ecb03 LH |
2537 | } |
2538 | #endif | |
183654ce | 2539 | vma_iter_set(vmi, start); |
6c26bd43 DW |
2540 | error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL); |
2541 | if (error) | |
606c812e | 2542 | goto clear_tree_failed; |
0378c0a0 | 2543 | |
6c26bd43 | 2544 | /* Point of no return */ |
606c812e | 2545 | mm->locked_vm -= locked_vm; |
763ecb03 | 2546 | mm->map_count -= count; |
408579cd | 2547 | if (unlock) |
e4bd84c0 | 2548 | mmap_write_downgrade(mm); |
dd2283f2 | 2549 | |
68f48381 SB |
2550 | /* |
2551 | * We can free page tables without write-locking mmap_lock because VMAs | |
2552 | * were isolated before we downgraded mmap_lock. | |
2553 | */ | |
408579cd | 2554 | unmap_region(mm, &mt_detach, vma, prev, next, start, end, !unlock); |
763ecb03 LH |
2555 | /* Statistics and freeing VMAs */ |
2556 | mas_set(&mas_detach, start); | |
2557 | remove_mt(mm, &mas_detach); | |
2558 | __mt_destroy(&mt_detach); | |
ae80b404 | 2559 | validate_mm(mm); |
408579cd LH |
2560 | if (unlock) |
2561 | mmap_read_unlock(mm); | |
1da177e4 | 2562 | |
408579cd | 2563 | return 0; |
d4af56c5 | 2564 | |
606c812e | 2565 | clear_tree_failed: |
d4af56c5 | 2566 | userfaultfd_error: |
606c812e | 2567 | munmap_gather_failed: |
763ecb03 | 2568 | end_split_failed: |
606c812e LH |
2569 | mas_set(&mas_detach, 0); |
2570 | mas_for_each(&mas_detach, next, end) | |
2571 | vma_mark_detached(next, false); | |
2572 | ||
763ecb03 LH |
2573 | __mt_destroy(&mt_detach); |
2574 | start_split_failed: | |
2575 | map_count_exceeded: | |
b5641a5d | 2576 | validate_mm(mm); |
d4af56c5 | 2577 | return error; |
1da177e4 | 2578 | } |
1da177e4 | 2579 | |
11f9a21a | 2580 | /* |
183654ce LH |
2581 | * do_vmi_munmap() - munmap a given range. |
2582 | * @vmi: The vma iterator | |
11f9a21a LH |
2583 | * @mm: The mm_struct |
2584 | * @start: The start address to munmap | |
2585 | * @len: The length of the range to munmap | |
2586 | * @uf: The userfaultfd list_head | |
408579cd | 2587 | * @unlock: set to true if the user wants to drop the mmap_lock on success |
11f9a21a LH |
2588 | * |
2589 | * This function takes a @mas that is either pointing to the previous VMA or set | |
2590 | * to MA_START and sets it up to remove the mapping(s). The @len will be | |
2591 | * aligned and any arch_unmap work will be preformed. | |
2592 | * | |
408579cd LH |
2593 | * Return: 0 on success and drops the lock if so directed, error and leaves the |
2594 | * lock held otherwise. | |
11f9a21a | 2595 | */ |
183654ce | 2596 | int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, |
11f9a21a | 2597 | unsigned long start, size_t len, struct list_head *uf, |
408579cd | 2598 | bool unlock) |
11f9a21a LH |
2599 | { |
2600 | unsigned long end; | |
2601 | struct vm_area_struct *vma; | |
2602 | ||
2603 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) | |
2604 | return -EINVAL; | |
2605 | ||
2606 | end = start + PAGE_ALIGN(len); | |
2607 | if (end == start) | |
2608 | return -EINVAL; | |
2609 | ||
2610 | /* arch_unmap() might do unmaps itself. */ | |
2611 | arch_unmap(mm, start, end); | |
2612 | ||
2613 | /* Find the first overlapping VMA */ | |
183654ce | 2614 | vma = vma_find(vmi, end); |
408579cd LH |
2615 | if (!vma) { |
2616 | if (unlock) | |
2617 | mmap_write_unlock(mm); | |
11f9a21a | 2618 | return 0; |
408579cd | 2619 | } |
11f9a21a | 2620 | |
408579cd | 2621 | return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); |
11f9a21a LH |
2622 | } |
2623 | ||
2624 | /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. | |
2625 | * @mm: The mm_struct | |
2626 | * @start: The start address to munmap | |
2627 | * @len: The length to be munmapped. | |
2628 | * @uf: The userfaultfd list_head | |
408579cd LH |
2629 | * |
2630 | * Return: 0 on success, error otherwise. | |
11f9a21a | 2631 | */ |
dd2283f2 YS |
2632 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, |
2633 | struct list_head *uf) | |
2634 | { | |
183654ce | 2635 | VMA_ITERATOR(vmi, mm, start); |
11f9a21a | 2636 | |
183654ce | 2637 | return do_vmi_munmap(&vmi, mm, start, len, uf, false); |
dd2283f2 YS |
2638 | } |
2639 | ||
e99668a5 LH |
2640 | unsigned long mmap_region(struct file *file, unsigned long addr, |
2641 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, | |
2642 | struct list_head *uf) | |
2643 | { | |
2644 | struct mm_struct *mm = current->mm; | |
2645 | struct vm_area_struct *vma = NULL; | |
2646 | struct vm_area_struct *next, *prev, *merge; | |
2647 | pgoff_t pglen = len >> PAGE_SHIFT; | |
2648 | unsigned long charged = 0; | |
2649 | unsigned long end = addr + len; | |
2650 | unsigned long merge_start = addr, merge_end = end; | |
2651 | pgoff_t vm_pgoff; | |
2652 | int error; | |
183654ce | 2653 | VMA_ITERATOR(vmi, mm, addr); |
e99668a5 LH |
2654 | |
2655 | /* Check against address space limit. */ | |
2656 | if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { | |
2657 | unsigned long nr_pages; | |
2658 | ||
2659 | /* | |
2660 | * MAP_FIXED may remove pages of mappings that intersects with | |
2661 | * requested mapping. Account for the pages it would unmap. | |
2662 | */ | |
2663 | nr_pages = count_vma_pages_range(mm, addr, end); | |
2664 | ||
2665 | if (!may_expand_vm(mm, vm_flags, | |
2666 | (len >> PAGE_SHIFT) - nr_pages)) | |
2667 | return -ENOMEM; | |
2668 | } | |
2669 | ||
2670 | /* Unmap any existing mapping in the area */ | |
183654ce | 2671 | if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) |
e99668a5 LH |
2672 | return -ENOMEM; |
2673 | ||
2674 | /* | |
2675 | * Private writable mapping: check memory availability | |
2676 | */ | |
2677 | if (accountable_mapping(file, vm_flags)) { | |
2678 | charged = len >> PAGE_SHIFT; | |
2679 | if (security_vm_enough_memory_mm(mm, charged)) | |
2680 | return -ENOMEM; | |
2681 | vm_flags |= VM_ACCOUNT; | |
2682 | } | |
2683 | ||
183654ce LH |
2684 | next = vma_next(&vmi); |
2685 | prev = vma_prev(&vmi); | |
e99668a5 LH |
2686 | if (vm_flags & VM_SPECIAL) |
2687 | goto cannot_expand; | |
2688 | ||
2689 | /* Attempt to expand an old mapping */ | |
2690 | /* Check next */ | |
2691 | if (next && next->vm_start == end && !vma_policy(next) && | |
2692 | can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, | |
2693 | NULL_VM_UFFD_CTX, NULL)) { | |
2694 | merge_end = next->vm_end; | |
2695 | vma = next; | |
2696 | vm_pgoff = next->vm_pgoff - pglen; | |
2697 | } | |
2698 | ||
2699 | /* Check prev */ | |
2700 | if (prev && prev->vm_end == addr && !vma_policy(prev) && | |
2701 | (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, | |
2702 | pgoff, vma->vm_userfaultfd_ctx, NULL) : | |
2703 | can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, | |
2704 | NULL_VM_UFFD_CTX, NULL))) { | |
2705 | merge_start = prev->vm_start; | |
2706 | vma = prev; | |
2707 | vm_pgoff = prev->vm_pgoff; | |
2708 | } | |
2709 | ||
2710 | ||
2711 | /* Actually expand, if possible */ | |
2712 | if (vma && | |
3c441ab7 | 2713 | !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { |
e99668a5 LH |
2714 | khugepaged_enter_vma(vma, vm_flags); |
2715 | goto expanded; | |
2716 | } | |
2717 | ||
e99668a5 | 2718 | cannot_expand: |
5c1c03de LH |
2719 | if (prev) |
2720 | vma_iter_next_range(&vmi); | |
2721 | ||
e99668a5 LH |
2722 | /* |
2723 | * Determine the object being mapped and call the appropriate | |
2724 | * specific mapper. the address has already been validated, but | |
2725 | * not unmapped, but the maps are removed from the list. | |
2726 | */ | |
2727 | vma = vm_area_alloc(mm); | |
2728 | if (!vma) { | |
2729 | error = -ENOMEM; | |
2730 | goto unacct_error; | |
2731 | } | |
2732 | ||
0fd5a9e2 | 2733 | vma_iter_set(&vmi, addr); |
e99668a5 LH |
2734 | vma->vm_start = addr; |
2735 | vma->vm_end = end; | |
1c71222e | 2736 | vm_flags_init(vma, vm_flags); |
e99668a5 LH |
2737 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
2738 | vma->vm_pgoff = pgoff; | |
2739 | ||
2740 | if (file) { | |
2741 | if (vm_flags & VM_SHARED) { | |
2742 | error = mapping_map_writable(file->f_mapping); | |
2743 | if (error) | |
2744 | goto free_vma; | |
2745 | } | |
2746 | ||
2747 | vma->vm_file = get_file(file); | |
2748 | error = call_mmap(file, vma); | |
2749 | if (error) | |
2750 | goto unmap_and_free_vma; | |
2751 | ||
a57b7051 LH |
2752 | /* |
2753 | * Expansion is handled above, merging is handled below. | |
2754 | * Drivers should not alter the address of the VMA. | |
e99668a5 | 2755 | */ |
cc8d1b09 LH |
2756 | error = -EINVAL; |
2757 | if (WARN_ON((addr != vma->vm_start))) | |
a57b7051 | 2758 | goto close_and_free_vma; |
e99668a5 | 2759 | |
cc8d1b09 | 2760 | vma_iter_set(&vmi, addr); |
e99668a5 LH |
2761 | /* |
2762 | * If vm_flags changed after call_mmap(), we should try merge | |
2763 | * vma again as we may succeed this time. | |
2764 | */ | |
2765 | if (unlikely(vm_flags != vma->vm_flags && prev)) { | |
9760ebff LH |
2766 | merge = vma_merge(&vmi, mm, prev, vma->vm_start, |
2767 | vma->vm_end, vma->vm_flags, NULL, | |
2768 | vma->vm_file, vma->vm_pgoff, NULL, | |
2769 | NULL_VM_UFFD_CTX, NULL); | |
e99668a5 LH |
2770 | if (merge) { |
2771 | /* | |
2772 | * ->mmap() can change vma->vm_file and fput | |
2773 | * the original file. So fput the vma->vm_file | |
2774 | * here or we would add an extra fput for file | |
2775 | * and cause general protection fault | |
2776 | * ultimately. | |
2777 | */ | |
2778 | fput(vma->vm_file); | |
2779 | vm_area_free(vma); | |
2780 | vma = merge; | |
2781 | /* Update vm_flags to pick up the change. */ | |
e99668a5 LH |
2782 | vm_flags = vma->vm_flags; |
2783 | goto unmap_writable; | |
2784 | } | |
2785 | } | |
2786 | ||
2787 | vm_flags = vma->vm_flags; | |
2788 | } else if (vm_flags & VM_SHARED) { | |
2789 | error = shmem_zero_setup(vma); | |
2790 | if (error) | |
2791 | goto free_vma; | |
2792 | } else { | |
2793 | vma_set_anonymous(vma); | |
2794 | } | |
2795 | ||
b507808e JG |
2796 | if (map_deny_write_exec(vma, vma->vm_flags)) { |
2797 | error = -EACCES; | |
6bbf1090 | 2798 | goto close_and_free_vma; |
b507808e JG |
2799 | } |
2800 | ||
e99668a5 | 2801 | /* Allow architectures to sanity-check the vm_flags */ |
cc8d1b09 LH |
2802 | error = -EINVAL; |
2803 | if (!arch_validate_flags(vma->vm_flags)) | |
2804 | goto close_and_free_vma; | |
e99668a5 | 2805 | |
cc8d1b09 LH |
2806 | error = -ENOMEM; |
2807 | if (vma_iter_prealloc(&vmi)) | |
2808 | goto close_and_free_vma; | |
e99668a5 | 2809 | |
1c7873e3 HD |
2810 | /* Lock the VMA since it is modified after insertion into VMA tree */ |
2811 | vma_start_write(vma); | |
183654ce | 2812 | vma_iter_store(&vmi, vma); |
e99668a5 LH |
2813 | mm->map_count++; |
2814 | if (vma->vm_file) { | |
6852c46c | 2815 | i_mmap_lock_write(vma->vm_file->f_mapping); |
e99668a5 LH |
2816 | if (vma->vm_flags & VM_SHARED) |
2817 | mapping_allow_writable(vma->vm_file->f_mapping); | |
2818 | ||
2819 | flush_dcache_mmap_lock(vma->vm_file->f_mapping); | |
2820 | vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); | |
2821 | flush_dcache_mmap_unlock(vma->vm_file->f_mapping); | |
2822 | i_mmap_unlock_write(vma->vm_file->f_mapping); | |
2823 | } | |
2824 | ||
2825 | /* | |
2826 | * vma_merge() calls khugepaged_enter_vma() either, the below | |
2827 | * call covers the non-merge case. | |
2828 | */ | |
2829 | khugepaged_enter_vma(vma, vma->vm_flags); | |
2830 | ||
2831 | /* Once vma denies write, undo our temporary denial count */ | |
2832 | unmap_writable: | |
2833 | if (file && vm_flags & VM_SHARED) | |
2834 | mapping_unmap_writable(file->f_mapping); | |
2835 | file = vma->vm_file; | |
d7597f59 | 2836 | ksm_add_vma(vma); |
e99668a5 LH |
2837 | expanded: |
2838 | perf_event_mmap(vma); | |
2839 | ||
2840 | vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); | |
2841 | if (vm_flags & VM_LOCKED) { | |
2842 | if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || | |
2843 | is_vm_hugetlb_page(vma) || | |
2844 | vma == get_gate_vma(current->mm)) | |
e430a95a | 2845 | vm_flags_clear(vma, VM_LOCKED_MASK); |
e99668a5 LH |
2846 | else |
2847 | mm->locked_vm += (len >> PAGE_SHIFT); | |
2848 | } | |
2849 | ||
2850 | if (file) | |
2851 | uprobe_mmap(vma); | |
2852 | ||
2853 | /* | |
2854 | * New (or expanded) vma always get soft dirty status. | |
2855 | * Otherwise user-space soft-dirty page tracker won't | |
2856 | * be able to distinguish situation when vma area unmapped, | |
2857 | * then new mapped in-place (which must be aimed as | |
2858 | * a completely new data area). | |
2859 | */ | |
1c71222e | 2860 | vm_flags_set(vma, VM_SOFTDIRTY); |
e99668a5 LH |
2861 | |
2862 | vma_set_page_prot(vma); | |
2863 | ||
2864 | validate_mm(mm); | |
2865 | return addr; | |
2866 | ||
deb0f656 | 2867 | close_and_free_vma: |
cc8d1b09 | 2868 | if (file && vma->vm_ops && vma->vm_ops->close) |
deb0f656 | 2869 | vma->vm_ops->close(vma); |
cc8d1b09 LH |
2870 | |
2871 | if (file || vma->vm_file) { | |
e99668a5 | 2872 | unmap_and_free_vma: |
cc8d1b09 LH |
2873 | fput(vma->vm_file); |
2874 | vma->vm_file = NULL; | |
e99668a5 | 2875 | |
cc8d1b09 LH |
2876 | /* Undo any partial mapping done by a device driver. */ |
2877 | unmap_region(mm, &mm->mm_mt, vma, prev, next, vma->vm_start, | |
68f48381 | 2878 | vma->vm_end, true); |
cc8d1b09 | 2879 | } |
cc674ab3 | 2880 | if (file && (vm_flags & VM_SHARED)) |
e99668a5 LH |
2881 | mapping_unmap_writable(file->f_mapping); |
2882 | free_vma: | |
2883 | vm_area_free(vma); | |
2884 | unacct_error: | |
2885 | if (charged) | |
2886 | vm_unacct_memory(charged); | |
2887 | validate_mm(mm); | |
2888 | return error; | |
2889 | } | |
2890 | ||
408579cd | 2891 | static int __vm_munmap(unsigned long start, size_t len, bool unlock) |
1da177e4 LT |
2892 | { |
2893 | int ret; | |
bfce281c | 2894 | struct mm_struct *mm = current->mm; |
897ab3e0 | 2895 | LIST_HEAD(uf); |
183654ce | 2896 | VMA_ITERATOR(vmi, mm, start); |
1da177e4 | 2897 | |
d8ed45c5 | 2898 | if (mmap_write_lock_killable(mm)) |
ae798783 MH |
2899 | return -EINTR; |
2900 | ||
408579cd LH |
2901 | ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock); |
2902 | if (ret || !unlock) | |
d8ed45c5 | 2903 | mmap_write_unlock(mm); |
dd2283f2 | 2904 | |
897ab3e0 | 2905 | userfaultfd_unmap_complete(mm, &uf); |
1da177e4 LT |
2906 | return ret; |
2907 | } | |
dd2283f2 YS |
2908 | |
2909 | int vm_munmap(unsigned long start, size_t len) | |
2910 | { | |
2911 | return __vm_munmap(start, len, false); | |
2912 | } | |
a46ef99d LT |
2913 | EXPORT_SYMBOL(vm_munmap); |
2914 | ||
2915 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2916 | { | |
ce18d171 | 2917 | addr = untagged_addr(addr); |
dd2283f2 | 2918 | return __vm_munmap(addr, len, true); |
a46ef99d | 2919 | } |
1da177e4 | 2920 | |
c8d78c18 KS |
2921 | |
2922 | /* | |
2923 | * Emulation of deprecated remap_file_pages() syscall. | |
2924 | */ | |
2925 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, | |
2926 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) | |
2927 | { | |
2928 | ||
2929 | struct mm_struct *mm = current->mm; | |
2930 | struct vm_area_struct *vma; | |
2931 | unsigned long populate = 0; | |
2932 | unsigned long ret = -EINVAL; | |
2933 | struct file *file; | |
2934 | ||
ee65728e | 2935 | pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", |
756a025f | 2936 | current->comm, current->pid); |
c8d78c18 KS |
2937 | |
2938 | if (prot) | |
2939 | return ret; | |
2940 | start = start & PAGE_MASK; | |
2941 | size = size & PAGE_MASK; | |
2942 | ||
2943 | if (start + size <= start) | |
2944 | return ret; | |
2945 | ||
2946 | /* Does pgoff wrap? */ | |
2947 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) | |
2948 | return ret; | |
2949 | ||
d8ed45c5 | 2950 | if (mmap_write_lock_killable(mm)) |
dc0ef0df MH |
2951 | return -EINTR; |
2952 | ||
9b593cb2 | 2953 | vma = vma_lookup(mm, start); |
c8d78c18 KS |
2954 | |
2955 | if (!vma || !(vma->vm_flags & VM_SHARED)) | |
2956 | goto out; | |
2957 | ||
48f7df32 | 2958 | if (start + size > vma->vm_end) { |
763ecb03 LH |
2959 | VMA_ITERATOR(vmi, mm, vma->vm_end); |
2960 | struct vm_area_struct *next, *prev = vma; | |
48f7df32 | 2961 | |
763ecb03 | 2962 | for_each_vma_range(vmi, next, start + size) { |
48f7df32 | 2963 | /* hole between vmas ? */ |
763ecb03 | 2964 | if (next->vm_start != prev->vm_end) |
48f7df32 KS |
2965 | goto out; |
2966 | ||
2967 | if (next->vm_file != vma->vm_file) | |
2968 | goto out; | |
2969 | ||
2970 | if (next->vm_flags != vma->vm_flags) | |
2971 | goto out; | |
2972 | ||
1db43d3f LH |
2973 | if (start + size <= next->vm_end) |
2974 | break; | |
2975 | ||
763ecb03 | 2976 | prev = next; |
48f7df32 KS |
2977 | } |
2978 | ||
2979 | if (!next) | |
2980 | goto out; | |
c8d78c18 KS |
2981 | } |
2982 | ||
2983 | prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; | |
2984 | prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; | |
2985 | prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; | |
2986 | ||
2987 | flags &= MAP_NONBLOCK; | |
2988 | flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; | |
fce000b1 | 2989 | if (vma->vm_flags & VM_LOCKED) |
c8d78c18 | 2990 | flags |= MAP_LOCKED; |
48f7df32 | 2991 | |
c8d78c18 | 2992 | file = get_file(vma->vm_file); |
45e55300 | 2993 | ret = do_mmap(vma->vm_file, start, size, |
897ab3e0 | 2994 | prot, flags, pgoff, &populate, NULL); |
c8d78c18 KS |
2995 | fput(file); |
2996 | out: | |
d8ed45c5 | 2997 | mmap_write_unlock(mm); |
c8d78c18 KS |
2998 | if (populate) |
2999 | mm_populate(ret, populate); | |
3000 | if (!IS_ERR_VALUE(ret)) | |
3001 | ret = 0; | |
3002 | return ret; | |
3003 | } | |
3004 | ||
1da177e4 | 3005 | /* |
27b26701 LH |
3006 | * do_vma_munmap() - Unmap a full or partial vma. |
3007 | * @vmi: The vma iterator pointing at the vma | |
3008 | * @vma: The first vma to be munmapped | |
3009 | * @start: the start of the address to unmap | |
3010 | * @end: The end of the address to unmap | |
2e7ce7d3 | 3011 | * @uf: The userfaultfd list_head |
408579cd | 3012 | * @unlock: Drop the lock on success |
2e7ce7d3 | 3013 | * |
27b26701 LH |
3014 | * unmaps a VMA mapping when the vma iterator is already in position. |
3015 | * Does not handle alignment. | |
408579cd LH |
3016 | * |
3017 | * Return: 0 on success drops the lock of so directed, error on failure and will | |
3018 | * still hold the lock. | |
1da177e4 | 3019 | */ |
27b26701 | 3020 | int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, |
408579cd LH |
3021 | unsigned long start, unsigned long end, struct list_head *uf, |
3022 | bool unlock) | |
1da177e4 | 3023 | { |
2e7ce7d3 | 3024 | struct mm_struct *mm = vma->vm_mm; |
3a459756 | 3025 | |
27b26701 | 3026 | arch_unmap(mm, start, end); |
b5641a5d | 3027 | return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); |
2e7ce7d3 | 3028 | } |
1da177e4 | 3029 | |
2e7ce7d3 LH |
3030 | /* |
3031 | * do_brk_flags() - Increase the brk vma if the flags match. | |
92fed820 | 3032 | * @vmi: The vma iterator |
2e7ce7d3 LH |
3033 | * @addr: The start address |
3034 | * @len: The length of the increase | |
3035 | * @vma: The vma, | |
3036 | * @flags: The VMA Flags | |
3037 | * | |
3038 | * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags | |
3039 | * do not match then create a new anonymous VMA. Eventually we may be able to | |
3040 | * do some brk-specific accounting here. | |
3041 | */ | |
92fed820 | 3042 | static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, |
763ecb03 | 3043 | unsigned long addr, unsigned long len, unsigned long flags) |
2e7ce7d3 LH |
3044 | { |
3045 | struct mm_struct *mm = current->mm; | |
287051b1 | 3046 | struct vma_prepare vp; |
1da177e4 | 3047 | |
b50e195f | 3048 | validate_mm(mm); |
2e7ce7d3 LH |
3049 | /* |
3050 | * Check against address space limits by the changed size | |
3051 | * Note: This happens *after* clearing old mappings in some code paths. | |
3052 | */ | |
3053 | flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; | |
84638335 | 3054 | if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) |
1da177e4 LT |
3055 | return -ENOMEM; |
3056 | ||
3057 | if (mm->map_count > sysctl_max_map_count) | |
3058 | return -ENOMEM; | |
3059 | ||
191c5424 | 3060 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
3061 | return -ENOMEM; |
3062 | ||
1da177e4 | 3063 | /* |
2e7ce7d3 LH |
3064 | * Expand the existing vma if possible; Note that singular lists do not |
3065 | * occur after forking, so the expand will only happen on new VMAs. | |
1da177e4 | 3066 | */ |
6c28ca64 LH |
3067 | if (vma && vma->vm_end == addr && !vma_policy(vma) && |
3068 | can_vma_merge_after(vma, flags, NULL, NULL, | |
3069 | addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { | |
92fed820 | 3070 | if (vma_iter_prealloc(vmi)) |
675eaca1 | 3071 | goto unacct_fail; |
28c5609f | 3072 | |
287051b1 LH |
3073 | init_vma_prep(&vp, vma); |
3074 | vma_prepare(&vp); | |
ccf1d78d | 3075 | vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); |
2e7ce7d3 | 3076 | vma->vm_end = addr + len; |
1c71222e | 3077 | vm_flags_set(vma, VM_SOFTDIRTY); |
92fed820 | 3078 | vma_iter_store(vmi, vma); |
2e7ce7d3 | 3079 | |
287051b1 | 3080 | vma_complete(&vp, vmi, mm); |
2e7ce7d3 LH |
3081 | khugepaged_enter_vma(vma, flags); |
3082 | goto out; | |
1da177e4 | 3083 | } |
2e7ce7d3 LH |
3084 | |
3085 | /* create a vma struct for an anonymous mapping */ | |
3086 | vma = vm_area_alloc(mm); | |
3087 | if (!vma) | |
675eaca1 | 3088 | goto unacct_fail; |
1da177e4 | 3089 | |
bfd40eaf | 3090 | vma_set_anonymous(vma); |
1da177e4 LT |
3091 | vma->vm_start = addr; |
3092 | vma->vm_end = addr + len; | |
2e7ce7d3 | 3093 | vma->vm_pgoff = addr >> PAGE_SHIFT; |
1c71222e | 3094 | vm_flags_init(vma, flags); |
3ed75eb8 | 3095 | vma->vm_page_prot = vm_get_page_prot(flags); |
92fed820 | 3096 | if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) |
2e7ce7d3 | 3097 | goto mas_store_fail; |
d4af56c5 | 3098 | |
2e7ce7d3 | 3099 | mm->map_count++; |
d7597f59 | 3100 | ksm_add_vma(vma); |
1da177e4 | 3101 | out: |
3af9e859 | 3102 | perf_event_mmap(vma); |
1da177e4 | 3103 | mm->total_vm += len >> PAGE_SHIFT; |
84638335 | 3104 | mm->data_vm += len >> PAGE_SHIFT; |
128557ff ML |
3105 | if (flags & VM_LOCKED) |
3106 | mm->locked_vm += (len >> PAGE_SHIFT); | |
1c71222e | 3107 | vm_flags_set(vma, VM_SOFTDIRTY); |
763ecb03 | 3108 | validate_mm(mm); |
5d22fc25 | 3109 | return 0; |
d4af56c5 | 3110 | |
2e7ce7d3 | 3111 | mas_store_fail: |
d4af56c5 | 3112 | vm_area_free(vma); |
675eaca1 | 3113 | unacct_fail: |
2e7ce7d3 LH |
3114 | vm_unacct_memory(len >> PAGE_SHIFT); |
3115 | return -ENOMEM; | |
1da177e4 LT |
3116 | } |
3117 | ||
bb177a73 | 3118 | int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) |
e4eb1ff6 LT |
3119 | { |
3120 | struct mm_struct *mm = current->mm; | |
2e7ce7d3 | 3121 | struct vm_area_struct *vma = NULL; |
bb177a73 | 3122 | unsigned long len; |
5d22fc25 | 3123 | int ret; |
128557ff | 3124 | bool populate; |
897ab3e0 | 3125 | LIST_HEAD(uf); |
92fed820 | 3126 | VMA_ITERATOR(vmi, mm, addr); |
e4eb1ff6 | 3127 | |
bb177a73 MH |
3128 | len = PAGE_ALIGN(request); |
3129 | if (len < request) | |
3130 | return -ENOMEM; | |
3131 | if (!len) | |
3132 | return 0; | |
3133 | ||
d8ed45c5 | 3134 | if (mmap_write_lock_killable(mm)) |
2d6c9282 MH |
3135 | return -EINTR; |
3136 | ||
2e7ce7d3 LH |
3137 | /* Until we need other flags, refuse anything except VM_EXEC. */ |
3138 | if ((flags & (~VM_EXEC)) != 0) | |
3139 | return -EINVAL; | |
3140 | ||
3141 | ret = check_brk_limits(addr, len); | |
3142 | if (ret) | |
3143 | goto limits_failed; | |
3144 | ||
183654ce | 3145 | ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); |
2e7ce7d3 LH |
3146 | if (ret) |
3147 | goto munmap_failed; | |
3148 | ||
92fed820 LH |
3149 | vma = vma_prev(&vmi); |
3150 | ret = do_brk_flags(&vmi, vma, addr, len, flags); | |
128557ff | 3151 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
d8ed45c5 | 3152 | mmap_write_unlock(mm); |
897ab3e0 | 3153 | userfaultfd_unmap_complete(mm, &uf); |
5d22fc25 | 3154 | if (populate && !ret) |
128557ff | 3155 | mm_populate(addr, len); |
e4eb1ff6 | 3156 | return ret; |
2e7ce7d3 LH |
3157 | |
3158 | munmap_failed: | |
3159 | limits_failed: | |
3160 | mmap_write_unlock(mm); | |
3161 | return ret; | |
e4eb1ff6 | 3162 | } |
16e72e9b DV |
3163 | EXPORT_SYMBOL(vm_brk_flags); |
3164 | ||
3165 | int vm_brk(unsigned long addr, unsigned long len) | |
3166 | { | |
3167 | return vm_brk_flags(addr, len, 0); | |
3168 | } | |
e4eb1ff6 | 3169 | EXPORT_SYMBOL(vm_brk); |
1da177e4 LT |
3170 | |
3171 | /* Release all mmaps. */ | |
3172 | void exit_mmap(struct mm_struct *mm) | |
3173 | { | |
d16dfc55 | 3174 | struct mmu_gather tlb; |
ba470de4 | 3175 | struct vm_area_struct *vma; |
1da177e4 | 3176 | unsigned long nr_accounted = 0; |
763ecb03 LH |
3177 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
3178 | int count = 0; | |
1da177e4 | 3179 | |
d6dd61c8 | 3180 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 3181 | mmu_notifier_release(mm); |
d6dd61c8 | 3182 | |
bf3980c8 | 3183 | mmap_read_lock(mm); |
9480c53e JF |
3184 | arch_exit_mmap(mm); |
3185 | ||
763ecb03 | 3186 | vma = mas_find(&mas, ULONG_MAX); |
64591e86 SB |
3187 | if (!vma) { |
3188 | /* Can happen if dup_mmap() received an OOM */ | |
bf3980c8 | 3189 | mmap_read_unlock(mm); |
9480c53e | 3190 | return; |
64591e86 | 3191 | } |
9480c53e | 3192 | |
1da177e4 | 3193 | lru_add_drain(); |
1da177e4 | 3194 | flush_cache_mm(mm); |
d8b45053 | 3195 | tlb_gather_mmu_fullmm(&tlb, mm); |
901608d9 | 3196 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
763ecb03 | 3197 | /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ |
68f48381 | 3198 | unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX, false); |
bf3980c8 SB |
3199 | mmap_read_unlock(mm); |
3200 | ||
3201 | /* | |
3202 | * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper | |
b3541d91 | 3203 | * because the memory has been already freed. |
bf3980c8 SB |
3204 | */ |
3205 | set_bit(MMF_OOM_SKIP, &mm->flags); | |
3206 | mmap_write_lock(mm); | |
3dd44325 | 3207 | mt_clear_in_rcu(&mm->mm_mt); |
763ecb03 | 3208 | free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS, |
98e51a22 | 3209 | USER_PGTABLES_CEILING, true); |
ae8eba8b | 3210 | tlb_finish_mmu(&tlb); |
1da177e4 | 3211 | |
763ecb03 LH |
3212 | /* |
3213 | * Walk the list again, actually closing and freeing it, with preemption | |
3214 | * enabled, without holding any MM locks besides the unreachable | |
3215 | * mmap_write_lock. | |
3216 | */ | |
3217 | do { | |
4f74d2c8 LT |
3218 | if (vma->vm_flags & VM_ACCOUNT) |
3219 | nr_accounted += vma_pages(vma); | |
0d2ebf9c | 3220 | remove_vma(vma, true); |
763ecb03 | 3221 | count++; |
0a3b3c25 | 3222 | cond_resched(); |
763ecb03 LH |
3223 | } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); |
3224 | ||
3225 | BUG_ON(count != mm->map_count); | |
d4af56c5 LH |
3226 | |
3227 | trace_exit_mmap(mm); | |
3228 | __mt_destroy(&mm->mm_mt); | |
64591e86 | 3229 | mmap_write_unlock(mm); |
4f74d2c8 | 3230 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
3231 | } |
3232 | ||
3233 | /* Insert vm structure into process list sorted by address | |
3234 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
c8c06efa | 3235 | * then i_mmap_rwsem is taken here. |
1da177e4 | 3236 | */ |
6597d783 | 3237 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 3238 | { |
d4af56c5 | 3239 | unsigned long charged = vma_pages(vma); |
1da177e4 | 3240 | |
d4af56c5 | 3241 | |
d0601a50 | 3242 | if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) |
c9d13f5f | 3243 | return -ENOMEM; |
d4af56c5 | 3244 | |
c9d13f5f | 3245 | if ((vma->vm_flags & VM_ACCOUNT) && |
d4af56c5 | 3246 | security_vm_enough_memory_mm(mm, charged)) |
c9d13f5f CG |
3247 | return -ENOMEM; |
3248 | ||
1da177e4 LT |
3249 | /* |
3250 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
3251 | * until its first write fault, when page's anon_vma and index | |
3252 | * are set. But now set the vm_pgoff it will almost certainly | |
3253 | * end up with (unless mremap moves it elsewhere before that | |
3254 | * first wfault), so /proc/pid/maps tells a consistent story. | |
3255 | * | |
3256 | * By setting it to reflect the virtual start address of the | |
3257 | * vma, merges and splits can happen in a seamless way, just | |
3258 | * using the existing file pgoff checks and manipulations. | |
8332326e | 3259 | * Similarly in do_mmap and in do_brk_flags. |
1da177e4 | 3260 | */ |
8a9cc3b5 | 3261 | if (vma_is_anonymous(vma)) { |
1da177e4 LT |
3262 | BUG_ON(vma->anon_vma); |
3263 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
3264 | } | |
2b144498 | 3265 | |
763ecb03 | 3266 | if (vma_link(mm, vma)) { |
d4af56c5 LH |
3267 | vm_unacct_memory(charged); |
3268 | return -ENOMEM; | |
3269 | } | |
3270 | ||
1da177e4 LT |
3271 | return 0; |
3272 | } | |
3273 | ||
3274 | /* | |
3275 | * Copy the vma structure to a new location in the same mm, | |
3276 | * prior to moving page table entries, to effect an mremap move. | |
3277 | */ | |
3278 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
3279 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
3280 | bool *need_rmap_locks) | |
1da177e4 LT |
3281 | { |
3282 | struct vm_area_struct *vma = *vmap; | |
3283 | unsigned long vma_start = vma->vm_start; | |
3284 | struct mm_struct *mm = vma->vm_mm; | |
3285 | struct vm_area_struct *new_vma, *prev; | |
948f017b | 3286 | bool faulted_in_anon_vma = true; |
076f16bf | 3287 | VMA_ITERATOR(vmi, mm, addr); |
1da177e4 | 3288 | |
b50e195f | 3289 | validate_mm(mm); |
1da177e4 LT |
3290 | /* |
3291 | * If anonymous vma has not yet been faulted, update new pgoff | |
3292 | * to match new location, to increase its chance of merging. | |
3293 | */ | |
ce75799b | 3294 | if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { |
1da177e4 | 3295 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
3296 | faulted_in_anon_vma = false; |
3297 | } | |
1da177e4 | 3298 | |
763ecb03 LH |
3299 | new_vma = find_vma_prev(mm, addr, &prev); |
3300 | if (new_vma && new_vma->vm_start < addr + len) | |
6597d783 | 3301 | return NULL; /* should never get here */ |
524e00b3 | 3302 | |
9760ebff | 3303 | new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, |
19a809af | 3304 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
5c26f6ac | 3305 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); |
1da177e4 LT |
3306 | if (new_vma) { |
3307 | /* | |
3308 | * Source vma may have been merged into new_vma | |
3309 | */ | |
948f017b AA |
3310 | if (unlikely(vma_start >= new_vma->vm_start && |
3311 | vma_start < new_vma->vm_end)) { | |
3312 | /* | |
3313 | * The only way we can get a vma_merge with | |
3314 | * self during an mremap is if the vma hasn't | |
3315 | * been faulted in yet and we were allowed to | |
3316 | * reset the dst vma->vm_pgoff to the | |
3317 | * destination address of the mremap to allow | |
3318 | * the merge to happen. mremap must change the | |
3319 | * vm_pgoff linearity between src and dst vmas | |
3320 | * (in turn preventing a vma_merge) to be | |
3321 | * safe. It is only safe to keep the vm_pgoff | |
3322 | * linear if there are no pages mapped yet. | |
3323 | */ | |
81d1b09c | 3324 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
38a76013 | 3325 | *vmap = vma = new_vma; |
108d6642 | 3326 | } |
38a76013 | 3327 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 3328 | } else { |
3928d4f5 | 3329 | new_vma = vm_area_dup(vma); |
e3975891 CG |
3330 | if (!new_vma) |
3331 | goto out; | |
e3975891 CG |
3332 | new_vma->vm_start = addr; |
3333 | new_vma->vm_end = addr + len; | |
3334 | new_vma->vm_pgoff = pgoff; | |
3335 | if (vma_dup_policy(vma, new_vma)) | |
3336 | goto out_free_vma; | |
e3975891 CG |
3337 | if (anon_vma_clone(new_vma, vma)) |
3338 | goto out_free_mempol; | |
3339 | if (new_vma->vm_file) | |
3340 | get_file(new_vma->vm_file); | |
3341 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
3342 | new_vma->vm_ops->open(new_vma); | |
d6ac235d | 3343 | vma_start_write(new_vma); |
763ecb03 | 3344 | if (vma_link(mm, new_vma)) |
524e00b3 | 3345 | goto out_vma_link; |
e3975891 | 3346 | *need_rmap_locks = false; |
1da177e4 | 3347 | } |
b50e195f | 3348 | validate_mm(mm); |
1da177e4 | 3349 | return new_vma; |
5beb4930 | 3350 | |
524e00b3 LH |
3351 | out_vma_link: |
3352 | if (new_vma->vm_ops && new_vma->vm_ops->close) | |
3353 | new_vma->vm_ops->close(new_vma); | |
92b73996 LH |
3354 | |
3355 | if (new_vma->vm_file) | |
3356 | fput(new_vma->vm_file); | |
3357 | ||
3358 | unlink_anon_vmas(new_vma); | |
e3975891 | 3359 | out_free_mempol: |
ef0855d3 | 3360 | mpol_put(vma_policy(new_vma)); |
e3975891 | 3361 | out_free_vma: |
3928d4f5 | 3362 | vm_area_free(new_vma); |
e3975891 | 3363 | out: |
b50e195f | 3364 | validate_mm(mm); |
5beb4930 | 3365 | return NULL; |
1da177e4 | 3366 | } |
119f657c | 3367 | |
3368 | /* | |
3369 | * Return true if the calling process may expand its vm space by the passed | |
3370 | * number of pages | |
3371 | */ | |
84638335 | 3372 | bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) |
119f657c | 3373 | { |
84638335 KK |
3374 | if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) |
3375 | return false; | |
119f657c | 3376 | |
d977d56c KK |
3377 | if (is_data_mapping(flags) && |
3378 | mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { | |
f4fcd558 KK |
3379 | /* Workaround for Valgrind */ |
3380 | if (rlimit(RLIMIT_DATA) == 0 && | |
3381 | mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) | |
3382 | return true; | |
57a7702b DW |
3383 | |
3384 | pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", | |
3385 | current->comm, current->pid, | |
3386 | (mm->data_vm + npages) << PAGE_SHIFT, | |
3387 | rlimit(RLIMIT_DATA), | |
3388 | ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); | |
3389 | ||
3390 | if (!ignore_rlimit_data) | |
d977d56c KK |
3391 | return false; |
3392 | } | |
119f657c | 3393 | |
84638335 KK |
3394 | return true; |
3395 | } | |
3396 | ||
3397 | void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) | |
3398 | { | |
7866076b | 3399 | WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); |
84638335 | 3400 | |
d977d56c | 3401 | if (is_exec_mapping(flags)) |
84638335 | 3402 | mm->exec_vm += npages; |
d977d56c | 3403 | else if (is_stack_mapping(flags)) |
84638335 | 3404 | mm->stack_vm += npages; |
d977d56c | 3405 | else if (is_data_mapping(flags)) |
84638335 | 3406 | mm->data_vm += npages; |
119f657c | 3407 | } |
fa5dc22f | 3408 | |
b3ec9f33 | 3409 | static vm_fault_t special_mapping_fault(struct vm_fault *vmf); |
a62c34bd AL |
3410 | |
3411 | /* | |
3412 | * Having a close hook prevents vma merging regardless of flags. | |
3413 | */ | |
3414 | static void special_mapping_close(struct vm_area_struct *vma) | |
3415 | { | |
3416 | } | |
3417 | ||
3418 | static const char *special_mapping_name(struct vm_area_struct *vma) | |
3419 | { | |
3420 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; | |
3421 | } | |
3422 | ||
14d07113 | 3423 | static int special_mapping_mremap(struct vm_area_struct *new_vma) |
b059a453 DS |
3424 | { |
3425 | struct vm_special_mapping *sm = new_vma->vm_private_data; | |
3426 | ||
280e87e9 DS |
3427 | if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) |
3428 | return -EFAULT; | |
3429 | ||
b059a453 DS |
3430 | if (sm->mremap) |
3431 | return sm->mremap(sm, new_vma); | |
280e87e9 | 3432 | |
b059a453 DS |
3433 | return 0; |
3434 | } | |
3435 | ||
871402e0 DS |
3436 | static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) |
3437 | { | |
3438 | /* | |
3439 | * Forbid splitting special mappings - kernel has expectations over | |
3440 | * the number of pages in mapping. Together with VM_DONTEXPAND | |
3441 | * the size of vma should stay the same over the special mapping's | |
3442 | * lifetime. | |
3443 | */ | |
3444 | return -EINVAL; | |
3445 | } | |
3446 | ||
a62c34bd AL |
3447 | static const struct vm_operations_struct special_mapping_vmops = { |
3448 | .close = special_mapping_close, | |
3449 | .fault = special_mapping_fault, | |
b059a453 | 3450 | .mremap = special_mapping_mremap, |
a62c34bd | 3451 | .name = special_mapping_name, |
af34ebeb DS |
3452 | /* vDSO code relies that VVAR can't be accessed remotely */ |
3453 | .access = NULL, | |
871402e0 | 3454 | .may_split = special_mapping_split, |
a62c34bd AL |
3455 | }; |
3456 | ||
3457 | static const struct vm_operations_struct legacy_special_mapping_vmops = { | |
3458 | .close = special_mapping_close, | |
3459 | .fault = special_mapping_fault, | |
3460 | }; | |
fa5dc22f | 3461 | |
b3ec9f33 | 3462 | static vm_fault_t special_mapping_fault(struct vm_fault *vmf) |
fa5dc22f | 3463 | { |
11bac800 | 3464 | struct vm_area_struct *vma = vmf->vma; |
b1d0e4f5 | 3465 | pgoff_t pgoff; |
fa5dc22f RM |
3466 | struct page **pages; |
3467 | ||
f872f540 | 3468 | if (vma->vm_ops == &legacy_special_mapping_vmops) { |
a62c34bd | 3469 | pages = vma->vm_private_data; |
f872f540 AL |
3470 | } else { |
3471 | struct vm_special_mapping *sm = vma->vm_private_data; | |
3472 | ||
3473 | if (sm->fault) | |
11bac800 | 3474 | return sm->fault(sm, vmf->vma, vmf); |
f872f540 AL |
3475 | |
3476 | pages = sm->pages; | |
3477 | } | |
a62c34bd | 3478 | |
8a9cc3b5 | 3479 | for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) |
b1d0e4f5 | 3480 | pgoff--; |
fa5dc22f RM |
3481 | |
3482 | if (*pages) { | |
3483 | struct page *page = *pages; | |
3484 | get_page(page); | |
b1d0e4f5 NP |
3485 | vmf->page = page; |
3486 | return 0; | |
fa5dc22f RM |
3487 | } |
3488 | ||
b1d0e4f5 | 3489 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
3490 | } |
3491 | ||
a62c34bd AL |
3492 | static struct vm_area_struct *__install_special_mapping( |
3493 | struct mm_struct *mm, | |
3494 | unsigned long addr, unsigned long len, | |
27f28b97 CG |
3495 | unsigned long vm_flags, void *priv, |
3496 | const struct vm_operations_struct *ops) | |
fa5dc22f | 3497 | { |
462e635e | 3498 | int ret; |
fa5dc22f RM |
3499 | struct vm_area_struct *vma; |
3500 | ||
b50e195f | 3501 | validate_mm(mm); |
490fc053 | 3502 | vma = vm_area_alloc(mm); |
fa5dc22f | 3503 | if (unlikely(vma == NULL)) |
3935ed6a | 3504 | return ERR_PTR(-ENOMEM); |
fa5dc22f | 3505 | |
fa5dc22f RM |
3506 | vma->vm_start = addr; |
3507 | vma->vm_end = addr + len; | |
3508 | ||
e430a95a SB |
3509 | vm_flags_init(vma, (vm_flags | mm->def_flags | |
3510 | VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); | |
3ed75eb8 | 3511 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f | 3512 | |
a62c34bd AL |
3513 | vma->vm_ops = ops; |
3514 | vma->vm_private_data = priv; | |
fa5dc22f | 3515 | |
462e635e TO |
3516 | ret = insert_vm_struct(mm, vma); |
3517 | if (ret) | |
3518 | goto out; | |
fa5dc22f | 3519 | |
84638335 | 3520 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); |
fa5dc22f | 3521 | |
cdd6c482 | 3522 | perf_event_mmap(vma); |
089dd79d | 3523 | |
b50e195f | 3524 | validate_mm(mm); |
3935ed6a | 3525 | return vma; |
462e635e TO |
3526 | |
3527 | out: | |
3928d4f5 | 3528 | vm_area_free(vma); |
b50e195f | 3529 | validate_mm(mm); |
3935ed6a SS |
3530 | return ERR_PTR(ret); |
3531 | } | |
3532 | ||
2eefd878 DS |
3533 | bool vma_is_special_mapping(const struct vm_area_struct *vma, |
3534 | const struct vm_special_mapping *sm) | |
3535 | { | |
3536 | return vma->vm_private_data == sm && | |
3537 | (vma->vm_ops == &special_mapping_vmops || | |
3538 | vma->vm_ops == &legacy_special_mapping_vmops); | |
3539 | } | |
3540 | ||
a62c34bd | 3541 | /* |
c1e8d7c6 | 3542 | * Called with mm->mmap_lock held for writing. |
a62c34bd AL |
3543 | * Insert a new vma covering the given region, with the given flags. |
3544 | * Its pages are supplied by the given array of struct page *. | |
3545 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
3546 | * The region past the last page supplied will always produce SIGBUS. | |
3547 | * The array pointer and the pages it points to are assumed to stay alive | |
3548 | * for as long as this mapping might exist. | |
3549 | */ | |
3550 | struct vm_area_struct *_install_special_mapping( | |
3551 | struct mm_struct *mm, | |
3552 | unsigned long addr, unsigned long len, | |
3553 | unsigned long vm_flags, const struct vm_special_mapping *spec) | |
3554 | { | |
27f28b97 CG |
3555 | return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, |
3556 | &special_mapping_vmops); | |
a62c34bd AL |
3557 | } |
3558 | ||
3935ed6a SS |
3559 | int install_special_mapping(struct mm_struct *mm, |
3560 | unsigned long addr, unsigned long len, | |
3561 | unsigned long vm_flags, struct page **pages) | |
3562 | { | |
a62c34bd | 3563 | struct vm_area_struct *vma = __install_special_mapping( |
27f28b97 CG |
3564 | mm, addr, len, vm_flags, (void *)pages, |
3565 | &legacy_special_mapping_vmops); | |
3935ed6a | 3566 | |
14bd5b45 | 3567 | return PTR_ERR_OR_ZERO(vma); |
fa5dc22f | 3568 | } |
7906d00c AA |
3569 | |
3570 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3571 | ||
454ed842 | 3572 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3573 | { |
f808c13f | 3574 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { |
7906d00c AA |
3575 | /* |
3576 | * The LSB of head.next can't change from under us | |
3577 | * because we hold the mm_all_locks_mutex. | |
3578 | */ | |
da1c55f1 | 3579 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); |
7906d00c AA |
3580 | /* |
3581 | * We can safely modify head.next after taking the | |
5a505085 | 3582 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3583 | * the same anon_vma we won't take it again. |
3584 | * | |
3585 | * No need of atomic instructions here, head.next | |
3586 | * can't change from under us thanks to the | |
5a505085 | 3587 | * anon_vma->root->rwsem. |
7906d00c AA |
3588 | */ |
3589 | if (__test_and_set_bit(0, (unsigned long *) | |
f808c13f | 3590 | &anon_vma->root->rb_root.rb_root.rb_node)) |
7906d00c AA |
3591 | BUG(); |
3592 | } | |
3593 | } | |
3594 | ||
454ed842 | 3595 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3596 | { |
3597 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3598 | /* | |
3599 | * AS_MM_ALL_LOCKS can't change from under us because | |
3600 | * we hold the mm_all_locks_mutex. | |
3601 | * | |
3602 | * Operations on ->flags have to be atomic because | |
3603 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3604 | * mm_all_locks_mutex, there may be other cpus | |
3605 | * changing other bitflags in parallel to us. | |
3606 | */ | |
3607 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3608 | BUG(); | |
da1c55f1 | 3609 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); |
7906d00c AA |
3610 | } |
3611 | } | |
3612 | ||
3613 | /* | |
3614 | * This operation locks against the VM for all pte/vma/mm related | |
3615 | * operations that could ever happen on a certain mm. This includes | |
3616 | * vmtruncate, try_to_unmap, and all page faults. | |
3617 | * | |
c1e8d7c6 | 3618 | * The caller must take the mmap_lock in write mode before calling |
7906d00c | 3619 | * mm_take_all_locks(). The caller isn't allowed to release the |
c1e8d7c6 | 3620 | * mmap_lock until mm_drop_all_locks() returns. |
7906d00c | 3621 | * |
c1e8d7c6 | 3622 | * mmap_lock in write mode is required in order to block all operations |
7906d00c | 3623 | * that could modify pagetables and free pages without need of |
27ba0644 | 3624 | * altering the vma layout. It's also needed in write mode to avoid new |
7906d00c AA |
3625 | * anon_vmas to be associated with existing vmas. |
3626 | * | |
3627 | * A single task can't take more than one mm_take_all_locks() in a row | |
3628 | * or it would deadlock. | |
3629 | * | |
bf181b9f | 3630 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3631 | * mapping->flags avoid to take the same lock twice, if more than one |
3632 | * vma in this mm is backed by the same anon_vma or address_space. | |
3633 | * | |
88f306b6 KS |
3634 | * We take locks in following order, accordingly to comment at beginning |
3635 | * of mm/rmap.c: | |
3636 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for | |
3637 | * hugetlb mapping); | |
eeff9a5d | 3638 | * - all vmas marked locked |
88f306b6 KS |
3639 | * - all i_mmap_rwsem locks; |
3640 | * - all anon_vma->rwseml | |
3641 | * | |
3642 | * We can take all locks within these types randomly because the VM code | |
3643 | * doesn't nest them and we protected from parallel mm_take_all_locks() by | |
3644 | * mm_all_locks_mutex. | |
7906d00c AA |
3645 | * |
3646 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3647 | * that may have to take thousand of locks. | |
3648 | * | |
3649 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3650 | */ | |
3651 | int mm_take_all_locks(struct mm_struct *mm) | |
3652 | { | |
3653 | struct vm_area_struct *vma; | |
5beb4930 | 3654 | struct anon_vma_chain *avc; |
763ecb03 | 3655 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
7906d00c | 3656 | |
325bca1f | 3657 | mmap_assert_write_locked(mm); |
7906d00c AA |
3658 | |
3659 | mutex_lock(&mm_all_locks_mutex); | |
3660 | ||
eeff9a5d SB |
3661 | mas_for_each(&mas, vma, ULONG_MAX) { |
3662 | if (signal_pending(current)) | |
3663 | goto out_unlock; | |
3664 | vma_start_write(vma); | |
3665 | } | |
3666 | ||
3667 | mas_set(&mas, 0); | |
763ecb03 | 3668 | mas_for_each(&mas, vma, ULONG_MAX) { |
7906d00c AA |
3669 | if (signal_pending(current)) |
3670 | goto out_unlock; | |
88f306b6 KS |
3671 | if (vma->vm_file && vma->vm_file->f_mapping && |
3672 | is_vm_hugetlb_page(vma)) | |
3673 | vm_lock_mapping(mm, vma->vm_file->f_mapping); | |
3674 | } | |
3675 | ||
763ecb03 LH |
3676 | mas_set(&mas, 0); |
3677 | mas_for_each(&mas, vma, ULONG_MAX) { | |
88f306b6 KS |
3678 | if (signal_pending(current)) |
3679 | goto out_unlock; | |
3680 | if (vma->vm_file && vma->vm_file->f_mapping && | |
3681 | !is_vm_hugetlb_page(vma)) | |
454ed842 | 3682 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3683 | } |
7cd5a02f | 3684 | |
763ecb03 LH |
3685 | mas_set(&mas, 0); |
3686 | mas_for_each(&mas, vma, ULONG_MAX) { | |
7cd5a02f PZ |
3687 | if (signal_pending(current)) |
3688 | goto out_unlock; | |
3689 | if (vma->anon_vma) | |
5beb4930 RR |
3690 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3691 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3692 | } |
7cd5a02f | 3693 | |
584cff54 | 3694 | return 0; |
7906d00c AA |
3695 | |
3696 | out_unlock: | |
584cff54 KC |
3697 | mm_drop_all_locks(mm); |
3698 | return -EINTR; | |
7906d00c AA |
3699 | } |
3700 | ||
3701 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3702 | { | |
f808c13f | 3703 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { |
7906d00c AA |
3704 | /* |
3705 | * The LSB of head.next can't change to 0 from under | |
3706 | * us because we hold the mm_all_locks_mutex. | |
3707 | * | |
3708 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3709 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3710 | * never see our bitflag. |
3711 | * | |
3712 | * No need of atomic instructions here, head.next | |
3713 | * can't change from under us until we release the | |
5a505085 | 3714 | * anon_vma->root->rwsem. |
7906d00c AA |
3715 | */ |
3716 | if (!__test_and_clear_bit(0, (unsigned long *) | |
f808c13f | 3717 | &anon_vma->root->rb_root.rb_root.rb_node)) |
7906d00c | 3718 | BUG(); |
08b52706 | 3719 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3720 | } |
3721 | } | |
3722 | ||
3723 | static void vm_unlock_mapping(struct address_space *mapping) | |
3724 | { | |
3725 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3726 | /* | |
3727 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3728 | * because we hold the mm_all_locks_mutex. | |
3729 | */ | |
83cde9e8 | 3730 | i_mmap_unlock_write(mapping); |
7906d00c AA |
3731 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3732 | &mapping->flags)) | |
3733 | BUG(); | |
3734 | } | |
3735 | } | |
3736 | ||
3737 | /* | |
c1e8d7c6 | 3738 | * The mmap_lock cannot be released by the caller until |
7906d00c AA |
3739 | * mm_drop_all_locks() returns. |
3740 | */ | |
3741 | void mm_drop_all_locks(struct mm_struct *mm) | |
3742 | { | |
3743 | struct vm_area_struct *vma; | |
5beb4930 | 3744 | struct anon_vma_chain *avc; |
763ecb03 | 3745 | MA_STATE(mas, &mm->mm_mt, 0, 0); |
7906d00c | 3746 | |
325bca1f | 3747 | mmap_assert_write_locked(mm); |
7906d00c AA |
3748 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); |
3749 | ||
763ecb03 | 3750 | mas_for_each(&mas, vma, ULONG_MAX) { |
7906d00c | 3751 | if (vma->anon_vma) |
5beb4930 RR |
3752 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3753 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3754 | if (vma->vm_file && vma->vm_file->f_mapping) |
3755 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3756 | } | |
eeff9a5d | 3757 | vma_end_write_all(mm); |
7906d00c AA |
3758 | |
3759 | mutex_unlock(&mm_all_locks_mutex); | |
3760 | } | |
8feae131 DH |
3761 | |
3762 | /* | |
3edf41d8 | 3763 | * initialise the percpu counter for VM |
8feae131 DH |
3764 | */ |
3765 | void __init mmap_init(void) | |
3766 | { | |
00a62ce9 KM |
3767 | int ret; |
3768 | ||
908c7f19 | 3769 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 3770 | VM_BUG_ON(ret); |
8feae131 | 3771 | } |
c9b1d098 AS |
3772 | |
3773 | /* | |
3774 | * Initialise sysctl_user_reserve_kbytes. | |
3775 | * | |
3776 | * This is intended to prevent a user from starting a single memory hogging | |
3777 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3778 | * mode. | |
3779 | * | |
3780 | * The default value is min(3% of free memory, 128MB) | |
3781 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3782 | */ | |
1640879a | 3783 | static int init_user_reserve(void) |
c9b1d098 AS |
3784 | { |
3785 | unsigned long free_kbytes; | |
3786 | ||
c41f012a | 3787 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
c9b1d098 AS |
3788 | |
3789 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3790 | return 0; | |
3791 | } | |
a64fb3cd | 3792 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
3793 | |
3794 | /* | |
3795 | * Initialise sysctl_admin_reserve_kbytes. | |
3796 | * | |
3797 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3798 | * to log in and kill a memory hogging process. | |
3799 | * | |
3800 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3801 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3802 | * only reserve 3% of free pages by default. | |
3803 | */ | |
1640879a | 3804 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3805 | { |
3806 | unsigned long free_kbytes; | |
3807 | ||
c41f012a | 3808 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
4eeab4f5 AS |
3809 | |
3810 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3811 | return 0; | |
3812 | } | |
a64fb3cd | 3813 | subsys_initcall(init_admin_reserve); |
1640879a AS |
3814 | |
3815 | /* | |
3816 | * Reinititalise user and admin reserves if memory is added or removed. | |
3817 | * | |
3818 | * The default user reserve max is 128MB, and the default max for the | |
3819 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3820 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3821 | * and tools like top. It may make sense to increase or even disable the | |
3822 | * reserve depending on the existence of swap or variations in the recovery | |
3823 | * tools. So, the admin may have changed them. | |
3824 | * | |
3825 | * If memory is added and the reserves have been eliminated or increased above | |
3826 | * the default max, then we'll trust the admin. | |
3827 | * | |
3828 | * If memory is removed and there isn't enough free memory, then we | |
3829 | * need to reset the reserves. | |
3830 | * | |
3831 | * Otherwise keep the reserve set by the admin. | |
3832 | */ | |
3833 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3834 | unsigned long action, void *data) | |
3835 | { | |
3836 | unsigned long tmp, free_kbytes; | |
3837 | ||
3838 | switch (action) { | |
3839 | case MEM_ONLINE: | |
3840 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3841 | tmp = sysctl_user_reserve_kbytes; | |
3842 | if (0 < tmp && tmp < (1UL << 17)) | |
3843 | init_user_reserve(); | |
3844 | ||
3845 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3846 | tmp = sysctl_admin_reserve_kbytes; | |
3847 | if (0 < tmp && tmp < (1UL << 13)) | |
3848 | init_admin_reserve(); | |
3849 | ||
3850 | break; | |
3851 | case MEM_OFFLINE: | |
c41f012a | 3852 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
1640879a AS |
3853 | |
3854 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3855 | init_user_reserve(); | |
3856 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3857 | sysctl_user_reserve_kbytes); | |
3858 | } | |
3859 | ||
3860 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3861 | init_admin_reserve(); | |
3862 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3863 | sysctl_admin_reserve_kbytes); | |
3864 | } | |
3865 | break; | |
3866 | default: | |
3867 | break; | |
3868 | } | |
3869 | return NOTIFY_OK; | |
3870 | } | |
3871 | ||
1640879a AS |
3872 | static int __meminit init_reserve_notifier(void) |
3873 | { | |
1eeaa4fd | 3874 | if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) |
b1de0d13 | 3875 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
1640879a AS |
3876 | |
3877 | return 0; | |
3878 | } | |
a64fb3cd | 3879 | subsys_initcall(init_reserve_notifier); |