]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/mmap.c | |
3 | * | |
4 | * Written by obz. | |
5 | * | |
046c6884 | 6 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
7 | */ |
8 | ||
b1de0d13 MH |
9 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
10 | ||
e8420a8e | 11 | #include <linux/kernel.h> |
1da177e4 | 12 | #include <linux/slab.h> |
4af3c9cc | 13 | #include <linux/backing-dev.h> |
1da177e4 | 14 | #include <linux/mm.h> |
615d6e87 | 15 | #include <linux/vmacache.h> |
1da177e4 LT |
16 | #include <linux/shm.h> |
17 | #include <linux/mman.h> | |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/syscalls.h> | |
c59ede7b | 21 | #include <linux/capability.h> |
1da177e4 LT |
22 | #include <linux/init.h> |
23 | #include <linux/file.h> | |
24 | #include <linux/fs.h> | |
25 | #include <linux/personality.h> | |
26 | #include <linux/security.h> | |
27 | #include <linux/hugetlb.h> | |
28 | #include <linux/profile.h> | |
b95f1b31 | 29 | #include <linux/export.h> |
1da177e4 LT |
30 | #include <linux/mount.h> |
31 | #include <linux/mempolicy.h> | |
32 | #include <linux/rmap.h> | |
cddb8a5c | 33 | #include <linux/mmu_notifier.h> |
82f71ae4 | 34 | #include <linux/mmdebug.h> |
cdd6c482 | 35 | #include <linux/perf_event.h> |
120a795d | 36 | #include <linux/audit.h> |
b15d00b6 | 37 | #include <linux/khugepaged.h> |
2b144498 | 38 | #include <linux/uprobes.h> |
d3737187 | 39 | #include <linux/rbtree_augmented.h> |
cf4aebc2 | 40 | #include <linux/sched/sysctl.h> |
1640879a AS |
41 | #include <linux/notifier.h> |
42 | #include <linux/memory.h> | |
b1de0d13 | 43 | #include <linux/printk.h> |
1da177e4 LT |
44 | |
45 | #include <asm/uaccess.h> | |
46 | #include <asm/cacheflush.h> | |
47 | #include <asm/tlb.h> | |
d6dd61c8 | 48 | #include <asm/mmu_context.h> |
1da177e4 | 49 | |
42b77728 JB |
50 | #include "internal.h" |
51 | ||
3a459756 KK |
52 | #ifndef arch_mmap_check |
53 | #define arch_mmap_check(addr, len, flags) (0) | |
54 | #endif | |
55 | ||
08e7d9b5 MS |
56 | #ifndef arch_rebalance_pgtables |
57 | #define arch_rebalance_pgtables(addr, len) (addr) | |
58 | #endif | |
59 | ||
e0da382c HD |
60 | static void unmap_region(struct mm_struct *mm, |
61 | struct vm_area_struct *vma, struct vm_area_struct *prev, | |
62 | unsigned long start, unsigned long end); | |
63 | ||
1da177e4 LT |
64 | /* description of effects of mapping type and prot in current implementation. |
65 | * this is due to the limited x86 page protection hardware. The expected | |
66 | * behavior is in parens: | |
67 | * | |
68 | * map_type prot | |
69 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC | |
70 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes | |
71 | * w: (no) no w: (no) no w: (yes) yes w: (no) no | |
72 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
cc71aba3 | 73 | * |
1da177e4 LT |
74 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes |
75 | * w: (no) no w: (no) no w: (copy) copy w: (no) no | |
76 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes | |
77 | * | |
78 | */ | |
79 | pgprot_t protection_map[16] = { | |
80 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, | |
81 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 | |
82 | }; | |
83 | ||
804af2cf HD |
84 | pgprot_t vm_get_page_prot(unsigned long vm_flags) |
85 | { | |
b845f313 DK |
86 | return __pgprot(pgprot_val(protection_map[vm_flags & |
87 | (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | | |
88 | pgprot_val(arch_vm_get_page_prot(vm_flags))); | |
804af2cf HD |
89 | } |
90 | EXPORT_SYMBOL(vm_get_page_prot); | |
91 | ||
34679d7e SL |
92 | int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
93 | int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ | |
49f0ce5f | 94 | unsigned long sysctl_overcommit_kbytes __read_mostly; |
c3d8c141 | 95 | int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; |
c9b1d098 | 96 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ |
4eeab4f5 | 97 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ |
34679d7e SL |
98 | /* |
99 | * Make sure vm_committed_as in one cacheline and not cacheline shared with | |
100 | * other variables. It can be updated by several CPUs frequently. | |
101 | */ | |
102 | struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; | |
1da177e4 | 103 | |
997071bc S |
104 | /* |
105 | * The global memory commitment made in the system can be a metric | |
106 | * that can be used to drive ballooning decisions when Linux is hosted | |
107 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically | |
108 | * balancing memory across competing virtual machines that are hosted. | |
109 | * Several metrics drive this policy engine including the guest reported | |
110 | * memory commitment. | |
111 | */ | |
112 | unsigned long vm_memory_committed(void) | |
113 | { | |
114 | return percpu_counter_read_positive(&vm_committed_as); | |
115 | } | |
116 | EXPORT_SYMBOL_GPL(vm_memory_committed); | |
117 | ||
1da177e4 LT |
118 | /* |
119 | * Check that a process has enough memory to allocate a new virtual | |
120 | * mapping. 0 means there is enough memory for the allocation to | |
121 | * succeed and -ENOMEM implies there is not. | |
122 | * | |
123 | * We currently support three overcommit policies, which are set via the | |
124 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
125 | * | |
126 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
127 | * Additional code 2002 Jul 20 by Robert Love. | |
128 | * | |
129 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
130 | * | |
131 | * Note this is a helper function intended to be used by LSMs which | |
132 | * wish to use this logic. | |
133 | */ | |
34b4e4aa | 134 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 | 135 | { |
c9b1d098 | 136 | unsigned long free, allowed, reserve; |
1da177e4 | 137 | |
82f71ae4 KK |
138 | VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < |
139 | -(s64)vm_committed_as_batch * num_online_cpus(), | |
140 | "memory commitment underflow"); | |
141 | ||
1da177e4 LT |
142 | vm_acct_memory(pages); |
143 | ||
144 | /* | |
145 | * Sometimes we want to use more memory than we have | |
146 | */ | |
147 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
148 | return 0; | |
149 | ||
150 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
c15bef30 DF |
151 | free = global_page_state(NR_FREE_PAGES); |
152 | free += global_page_state(NR_FILE_PAGES); | |
153 | ||
154 | /* | |
155 | * shmem pages shouldn't be counted as free in this | |
156 | * case, they can't be purged, only swapped out, and | |
157 | * that won't affect the overall amount of available | |
158 | * memory in the system. | |
159 | */ | |
160 | free -= global_page_state(NR_SHMEM); | |
1da177e4 | 161 | |
ec8acf20 | 162 | free += get_nr_swap_pages(); |
1da177e4 LT |
163 | |
164 | /* | |
165 | * Any slabs which are created with the | |
166 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
167 | * which are reclaimable, under pressure. The dentry | |
168 | * cache and most inode caches should fall into this | |
169 | */ | |
972d1a7b | 170 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 | 171 | |
6d9f7839 HA |
172 | /* |
173 | * Leave reserved pages. The pages are not for anonymous pages. | |
174 | */ | |
c15bef30 | 175 | if (free <= totalreserve_pages) |
6d9f7839 HA |
176 | goto error; |
177 | else | |
c15bef30 | 178 | free -= totalreserve_pages; |
6d9f7839 HA |
179 | |
180 | /* | |
4eeab4f5 | 181 | * Reserve some for root |
6d9f7839 | 182 | */ |
1da177e4 | 183 | if (!cap_sys_admin) |
4eeab4f5 | 184 | free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 LT |
185 | |
186 | if (free > pages) | |
187 | return 0; | |
6d9f7839 HA |
188 | |
189 | goto error; | |
1da177e4 LT |
190 | } |
191 | ||
00619bcc | 192 | allowed = vm_commit_limit(); |
1da177e4 | 193 | /* |
4eeab4f5 | 194 | * Reserve some for root |
1da177e4 LT |
195 | */ |
196 | if (!cap_sys_admin) | |
4eeab4f5 | 197 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 | 198 | |
c9b1d098 AS |
199 | /* |
200 | * Don't let a single process grow so big a user can't recover | |
201 | */ | |
202 | if (mm) { | |
203 | reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); | |
204 | allowed -= min(mm->total_vm / 32, reserve); | |
205 | } | |
1da177e4 | 206 | |
00a62ce9 | 207 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 208 | return 0; |
6d9f7839 | 209 | error: |
1da177e4 LT |
210 | vm_unacct_memory(pages); |
211 | ||
212 | return -ENOMEM; | |
213 | } | |
214 | ||
1da177e4 | 215 | /* |
3d48ae45 | 216 | * Requires inode->i_mapping->i_mmap_mutex |
1da177e4 LT |
217 | */ |
218 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, | |
219 | struct file *file, struct address_space *mapping) | |
220 | { | |
221 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 222 | atomic_inc(&file_inode(file)->i_writecount); |
1da177e4 | 223 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 224 | mapping_unmap_writable(mapping); |
1da177e4 LT |
225 | |
226 | flush_dcache_mmap_lock(mapping); | |
227 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
6b2dbba8 | 228 | list_del_init(&vma->shared.nonlinear); |
1da177e4 | 229 | else |
6b2dbba8 | 230 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 LT |
231 | flush_dcache_mmap_unlock(mapping); |
232 | } | |
233 | ||
234 | /* | |
6b2dbba8 | 235 | * Unlink a file-based vm structure from its interval tree, to hide |
a8fb5618 | 236 | * vma from rmap and vmtruncate before freeing its page tables. |
1da177e4 | 237 | */ |
a8fb5618 | 238 | void unlink_file_vma(struct vm_area_struct *vma) |
1da177e4 LT |
239 | { |
240 | struct file *file = vma->vm_file; | |
241 | ||
1da177e4 LT |
242 | if (file) { |
243 | struct address_space *mapping = file->f_mapping; | |
3d48ae45 | 244 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 | 245 | __remove_shared_vm_struct(vma, file, mapping); |
3d48ae45 | 246 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 | 247 | } |
a8fb5618 HD |
248 | } |
249 | ||
250 | /* | |
251 | * Close a vm structure and free it, returning the next. | |
252 | */ | |
253 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) | |
254 | { | |
255 | struct vm_area_struct *next = vma->vm_next; | |
256 | ||
a8fb5618 | 257 | might_sleep(); |
1da177e4 LT |
258 | if (vma->vm_ops && vma->vm_ops->close) |
259 | vma->vm_ops->close(vma); | |
e9714acf | 260 | if (vma->vm_file) |
a8fb5618 | 261 | fput(vma->vm_file); |
f0be3d32 | 262 | mpol_put(vma_policy(vma)); |
1da177e4 | 263 | kmem_cache_free(vm_area_cachep, vma); |
a8fb5618 | 264 | return next; |
1da177e4 LT |
265 | } |
266 | ||
e4eb1ff6 LT |
267 | static unsigned long do_brk(unsigned long addr, unsigned long len); |
268 | ||
6a6160a7 | 269 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 | 270 | { |
8764b338 | 271 | unsigned long retval; |
1da177e4 LT |
272 | unsigned long newbrk, oldbrk; |
273 | struct mm_struct *mm = current->mm; | |
a5b4592c | 274 | unsigned long min_brk; |
128557ff | 275 | bool populate; |
1da177e4 LT |
276 | |
277 | down_write(&mm->mmap_sem); | |
278 | ||
a5b4592c | 279 | #ifdef CONFIG_COMPAT_BRK |
5520e894 JK |
280 | /* |
281 | * CONFIG_COMPAT_BRK can still be overridden by setting | |
282 | * randomize_va_space to 2, which will still cause mm->start_brk | |
283 | * to be arbitrarily shifted | |
284 | */ | |
4471a675 | 285 | if (current->brk_randomized) |
5520e894 JK |
286 | min_brk = mm->start_brk; |
287 | else | |
288 | min_brk = mm->end_data; | |
a5b4592c JK |
289 | #else |
290 | min_brk = mm->start_brk; | |
291 | #endif | |
292 | if (brk < min_brk) | |
1da177e4 | 293 | goto out; |
1e624196 RG |
294 | |
295 | /* | |
296 | * Check against rlimit here. If this check is done later after the test | |
297 | * of oldbrk with newbrk then it can escape the test and let the data | |
298 | * segment grow beyond its set limit the in case where the limit is | |
299 | * not page aligned -Ram Gupta | |
300 | */ | |
8764b338 CG |
301 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
302 | mm->end_data, mm->start_data)) | |
1e624196 RG |
303 | goto out; |
304 | ||
1da177e4 LT |
305 | newbrk = PAGE_ALIGN(brk); |
306 | oldbrk = PAGE_ALIGN(mm->brk); | |
307 | if (oldbrk == newbrk) | |
308 | goto set_brk; | |
309 | ||
310 | /* Always allow shrinking brk. */ | |
311 | if (brk <= mm->brk) { | |
312 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) | |
313 | goto set_brk; | |
314 | goto out; | |
315 | } | |
316 | ||
1da177e4 LT |
317 | /* Check against existing mmap mappings. */ |
318 | if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) | |
319 | goto out; | |
320 | ||
321 | /* Ok, looks good - let it rip. */ | |
322 | if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) | |
323 | goto out; | |
128557ff | 324 | |
1da177e4 LT |
325 | set_brk: |
326 | mm->brk = brk; | |
128557ff ML |
327 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
328 | up_write(&mm->mmap_sem); | |
329 | if (populate) | |
330 | mm_populate(oldbrk, newbrk - oldbrk); | |
331 | return brk; | |
332 | ||
1da177e4 LT |
333 | out: |
334 | retval = mm->brk; | |
335 | up_write(&mm->mmap_sem); | |
336 | return retval; | |
337 | } | |
338 | ||
d3737187 ML |
339 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
340 | { | |
341 | unsigned long max, subtree_gap; | |
342 | max = vma->vm_start; | |
343 | if (vma->vm_prev) | |
344 | max -= vma->vm_prev->vm_end; | |
345 | if (vma->vm_rb.rb_left) { | |
346 | subtree_gap = rb_entry(vma->vm_rb.rb_left, | |
347 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
348 | if (subtree_gap > max) | |
349 | max = subtree_gap; | |
350 | } | |
351 | if (vma->vm_rb.rb_right) { | |
352 | subtree_gap = rb_entry(vma->vm_rb.rb_right, | |
353 | struct vm_area_struct, vm_rb)->rb_subtree_gap; | |
354 | if (subtree_gap > max) | |
355 | max = subtree_gap; | |
356 | } | |
357 | return max; | |
358 | } | |
359 | ||
ed8ea815 | 360 | #ifdef CONFIG_DEBUG_VM_RB |
1da177e4 LT |
361 | static int browse_rb(struct rb_root *root) |
362 | { | |
5a0768f6 | 363 | int i = 0, j, bug = 0; |
1da177e4 LT |
364 | struct rb_node *nd, *pn = NULL; |
365 | unsigned long prev = 0, pend = 0; | |
366 | ||
367 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
368 | struct vm_area_struct *vma; | |
369 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
5a0768f6 | 370 | if (vma->vm_start < prev) { |
ff26f70f AM |
371 | pr_emerg("vm_start %lx < prev %lx\n", |
372 | vma->vm_start, prev); | |
5a0768f6 ML |
373 | bug = 1; |
374 | } | |
375 | if (vma->vm_start < pend) { | |
ff26f70f AM |
376 | pr_emerg("vm_start %lx < pend %lx\n", |
377 | vma->vm_start, pend); | |
5a0768f6 ML |
378 | bug = 1; |
379 | } | |
380 | if (vma->vm_start > vma->vm_end) { | |
ff26f70f AM |
381 | pr_emerg("vm_start %lx > vm_end %lx\n", |
382 | vma->vm_start, vma->vm_end); | |
5a0768f6 ML |
383 | bug = 1; |
384 | } | |
385 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { | |
8542bdfc | 386 | pr_emerg("free gap %lx, correct %lx\n", |
5a0768f6 ML |
387 | vma->rb_subtree_gap, |
388 | vma_compute_subtree_gap(vma)); | |
389 | bug = 1; | |
390 | } | |
1da177e4 LT |
391 | i++; |
392 | pn = nd; | |
d1af65d1 DM |
393 | prev = vma->vm_start; |
394 | pend = vma->vm_end; | |
1da177e4 LT |
395 | } |
396 | j = 0; | |
5a0768f6 | 397 | for (nd = pn; nd; nd = rb_prev(nd)) |
1da177e4 | 398 | j++; |
5a0768f6 | 399 | if (i != j) { |
8542bdfc | 400 | pr_emerg("backwards %d, forwards %d\n", j, i); |
5a0768f6 | 401 | bug = 1; |
1da177e4 | 402 | } |
5a0768f6 | 403 | return bug ? -1 : i; |
1da177e4 LT |
404 | } |
405 | ||
d3737187 ML |
406 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
407 | { | |
408 | struct rb_node *nd; | |
409 | ||
410 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { | |
411 | struct vm_area_struct *vma; | |
412 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); | |
413 | BUG_ON(vma != ignore && | |
414 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma)); | |
1da177e4 | 415 | } |
1da177e4 LT |
416 | } |
417 | ||
eafd4dc4 | 418 | static void validate_mm(struct mm_struct *mm) |
1da177e4 LT |
419 | { |
420 | int bug = 0; | |
421 | int i = 0; | |
5a0768f6 | 422 | unsigned long highest_address = 0; |
ed8ea815 | 423 | struct vm_area_struct *vma = mm->mmap; |
ff26f70f | 424 | |
ed8ea815 ML |
425 | while (vma) { |
426 | struct anon_vma_chain *avc; | |
ff26f70f | 427 | |
63c3b902 | 428 | vma_lock_anon_vma(vma); |
ed8ea815 ML |
429 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
430 | anon_vma_interval_tree_verify(avc); | |
63c3b902 | 431 | vma_unlock_anon_vma(vma); |
5a0768f6 | 432 | highest_address = vma->vm_end; |
ed8ea815 | 433 | vma = vma->vm_next; |
1da177e4 LT |
434 | i++; |
435 | } | |
5a0768f6 | 436 | if (i != mm->map_count) { |
8542bdfc | 437 | pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); |
5a0768f6 ML |
438 | bug = 1; |
439 | } | |
440 | if (highest_address != mm->highest_vm_end) { | |
8542bdfc | 441 | pr_emerg("mm->highest_vm_end %lx, found %lx\n", |
ff26f70f | 442 | mm->highest_vm_end, highest_address); |
5a0768f6 ML |
443 | bug = 1; |
444 | } | |
1da177e4 | 445 | i = browse_rb(&mm->mm_rb); |
5a0768f6 | 446 | if (i != mm->map_count) { |
ff26f70f AM |
447 | if (i != -1) |
448 | pr_emerg("map_count %d rb %d\n", mm->map_count, i); | |
5a0768f6 ML |
449 | bug = 1; |
450 | } | |
46a350ef | 451 | BUG_ON(bug); |
1da177e4 LT |
452 | } |
453 | #else | |
d3737187 | 454 | #define validate_mm_rb(root, ignore) do { } while (0) |
1da177e4 LT |
455 | #define validate_mm(mm) do { } while (0) |
456 | #endif | |
457 | ||
d3737187 ML |
458 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
459 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) | |
460 | ||
461 | /* | |
462 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or | |
463 | * vma->vm_prev->vm_end values changed, without modifying the vma's position | |
464 | * in the rbtree. | |
465 | */ | |
466 | static void vma_gap_update(struct vm_area_struct *vma) | |
467 | { | |
468 | /* | |
469 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback | |
470 | * function that does exacltly what we want. | |
471 | */ | |
472 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); | |
473 | } | |
474 | ||
475 | static inline void vma_rb_insert(struct vm_area_struct *vma, | |
476 | struct rb_root *root) | |
477 | { | |
478 | /* All rb_subtree_gap values must be consistent prior to insertion */ | |
479 | validate_mm_rb(root, NULL); | |
480 | ||
481 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
482 | } | |
483 | ||
484 | static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) | |
485 | { | |
486 | /* | |
487 | * All rb_subtree_gap values must be consistent prior to erase, | |
488 | * with the possible exception of the vma being erased. | |
489 | */ | |
490 | validate_mm_rb(root, vma); | |
491 | ||
492 | /* | |
493 | * Note rb_erase_augmented is a fairly large inline function, | |
494 | * so make sure we instantiate it only once with our desired | |
495 | * augmented rbtree callbacks. | |
496 | */ | |
497 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); | |
498 | } | |
499 | ||
bf181b9f ML |
500 | /* |
501 | * vma has some anon_vma assigned, and is already inserted on that | |
502 | * anon_vma's interval trees. | |
503 | * | |
504 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the | |
505 | * vma must be removed from the anon_vma's interval trees using | |
506 | * anon_vma_interval_tree_pre_update_vma(). | |
507 | * | |
508 | * After the update, the vma will be reinserted using | |
509 | * anon_vma_interval_tree_post_update_vma(). | |
510 | * | |
511 | * The entire update must be protected by exclusive mmap_sem and by | |
512 | * the root anon_vma's mutex. | |
513 | */ | |
514 | static inline void | |
515 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) | |
516 | { | |
517 | struct anon_vma_chain *avc; | |
518 | ||
519 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
520 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); | |
521 | } | |
522 | ||
523 | static inline void | |
524 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) | |
525 | { | |
526 | struct anon_vma_chain *avc; | |
527 | ||
528 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) | |
529 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); | |
530 | } | |
531 | ||
6597d783 HD |
532 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
533 | unsigned long end, struct vm_area_struct **pprev, | |
534 | struct rb_node ***rb_link, struct rb_node **rb_parent) | |
1da177e4 | 535 | { |
6597d783 | 536 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
1da177e4 LT |
537 | |
538 | __rb_link = &mm->mm_rb.rb_node; | |
539 | rb_prev = __rb_parent = NULL; | |
1da177e4 LT |
540 | |
541 | while (*__rb_link) { | |
542 | struct vm_area_struct *vma_tmp; | |
543 | ||
544 | __rb_parent = *__rb_link; | |
545 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); | |
546 | ||
547 | if (vma_tmp->vm_end > addr) { | |
6597d783 HD |
548 | /* Fail if an existing vma overlaps the area */ |
549 | if (vma_tmp->vm_start < end) | |
550 | return -ENOMEM; | |
1da177e4 LT |
551 | __rb_link = &__rb_parent->rb_left; |
552 | } else { | |
553 | rb_prev = __rb_parent; | |
554 | __rb_link = &__rb_parent->rb_right; | |
555 | } | |
556 | } | |
557 | ||
558 | *pprev = NULL; | |
559 | if (rb_prev) | |
560 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
561 | *rb_link = __rb_link; | |
562 | *rb_parent = __rb_parent; | |
6597d783 | 563 | return 0; |
1da177e4 LT |
564 | } |
565 | ||
e8420a8e CH |
566 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
567 | unsigned long addr, unsigned long end) | |
568 | { | |
569 | unsigned long nr_pages = 0; | |
570 | struct vm_area_struct *vma; | |
571 | ||
572 | /* Find first overlaping mapping */ | |
573 | vma = find_vma_intersection(mm, addr, end); | |
574 | if (!vma) | |
575 | return 0; | |
576 | ||
577 | nr_pages = (min(end, vma->vm_end) - | |
578 | max(addr, vma->vm_start)) >> PAGE_SHIFT; | |
579 | ||
580 | /* Iterate over the rest of the overlaps */ | |
581 | for (vma = vma->vm_next; vma; vma = vma->vm_next) { | |
582 | unsigned long overlap_len; | |
583 | ||
584 | if (vma->vm_start > end) | |
585 | break; | |
586 | ||
587 | overlap_len = min(end, vma->vm_end) - vma->vm_start; | |
588 | nr_pages += overlap_len >> PAGE_SHIFT; | |
589 | } | |
590 | ||
591 | return nr_pages; | |
592 | } | |
593 | ||
1da177e4 LT |
594 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
595 | struct rb_node **rb_link, struct rb_node *rb_parent) | |
596 | { | |
d3737187 ML |
597 | /* Update tracking information for the gap following the new vma. */ |
598 | if (vma->vm_next) | |
599 | vma_gap_update(vma->vm_next); | |
600 | else | |
601 | mm->highest_vm_end = vma->vm_end; | |
602 | ||
603 | /* | |
604 | * vma->vm_prev wasn't known when we followed the rbtree to find the | |
605 | * correct insertion point for that vma. As a result, we could not | |
606 | * update the vma vm_rb parents rb_subtree_gap values on the way down. | |
607 | * So, we first insert the vma with a zero rb_subtree_gap value | |
608 | * (to be consistent with what we did on the way down), and then | |
609 | * immediately update the gap to the correct value. Finally we | |
610 | * rebalance the rbtree after all augmented values have been set. | |
611 | */ | |
1da177e4 | 612 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
d3737187 ML |
613 | vma->rb_subtree_gap = 0; |
614 | vma_gap_update(vma); | |
615 | vma_rb_insert(vma, &mm->mm_rb); | |
1da177e4 LT |
616 | } |
617 | ||
cb8f488c | 618 | static void __vma_link_file(struct vm_area_struct *vma) |
1da177e4 | 619 | { |
48aae425 | 620 | struct file *file; |
1da177e4 LT |
621 | |
622 | file = vma->vm_file; | |
623 | if (file) { | |
624 | struct address_space *mapping = file->f_mapping; | |
625 | ||
626 | if (vma->vm_flags & VM_DENYWRITE) | |
496ad9aa | 627 | atomic_dec(&file_inode(file)->i_writecount); |
1da177e4 | 628 | if (vma->vm_flags & VM_SHARED) |
4bb5f5d9 | 629 | atomic_inc(&mapping->i_mmap_writable); |
1da177e4 LT |
630 | |
631 | flush_dcache_mmap_lock(mapping); | |
632 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) | |
633 | vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); | |
634 | else | |
6b2dbba8 | 635 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 LT |
636 | flush_dcache_mmap_unlock(mapping); |
637 | } | |
638 | } | |
639 | ||
640 | static void | |
641 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
642 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
643 | struct rb_node *rb_parent) | |
644 | { | |
645 | __vma_link_list(mm, vma, prev, rb_parent); | |
646 | __vma_link_rb(mm, vma, rb_link, rb_parent); | |
1da177e4 LT |
647 | } |
648 | ||
649 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, | |
650 | struct vm_area_struct *prev, struct rb_node **rb_link, | |
651 | struct rb_node *rb_parent) | |
652 | { | |
653 | struct address_space *mapping = NULL; | |
654 | ||
64ac4940 | 655 | if (vma->vm_file) { |
1da177e4 | 656 | mapping = vma->vm_file->f_mapping; |
3d48ae45 | 657 | mutex_lock(&mapping->i_mmap_mutex); |
64ac4940 | 658 | } |
1da177e4 LT |
659 | |
660 | __vma_link(mm, vma, prev, rb_link, rb_parent); | |
661 | __vma_link_file(vma); | |
662 | ||
1da177e4 | 663 | if (mapping) |
3d48ae45 | 664 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
665 | |
666 | mm->map_count++; | |
667 | validate_mm(mm); | |
668 | } | |
669 | ||
670 | /* | |
88f6b4c3 | 671 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
6b2dbba8 | 672 | * mm's list and rbtree. It has already been inserted into the interval tree. |
1da177e4 | 673 | */ |
48aae425 | 674 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 675 | { |
6597d783 | 676 | struct vm_area_struct *prev; |
48aae425 | 677 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 678 | |
6597d783 HD |
679 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
680 | &prev, &rb_link, &rb_parent)) | |
681 | BUG(); | |
1da177e4 LT |
682 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
683 | mm->map_count++; | |
684 | } | |
685 | ||
686 | static inline void | |
687 | __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, | |
688 | struct vm_area_struct *prev) | |
689 | { | |
d3737187 | 690 | struct vm_area_struct *next; |
297c5eee | 691 | |
d3737187 ML |
692 | vma_rb_erase(vma, &mm->mm_rb); |
693 | prev->vm_next = next = vma->vm_next; | |
297c5eee LT |
694 | if (next) |
695 | next->vm_prev = prev; | |
615d6e87 DB |
696 | |
697 | /* Kill the cache */ | |
698 | vmacache_invalidate(mm); | |
1da177e4 LT |
699 | } |
700 | ||
701 | /* | |
702 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that | |
703 | * is already present in an i_mmap tree without adjusting the tree. | |
704 | * The following helper function should be used when such adjustments | |
705 | * are necessary. The "insert" vma (if any) is to be inserted | |
706 | * before we drop the necessary locks. | |
707 | */ | |
5beb4930 | 708 | int vma_adjust(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
709 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) |
710 | { | |
711 | struct mm_struct *mm = vma->vm_mm; | |
712 | struct vm_area_struct *next = vma->vm_next; | |
713 | struct vm_area_struct *importer = NULL; | |
714 | struct address_space *mapping = NULL; | |
6b2dbba8 | 715 | struct rb_root *root = NULL; |
012f1800 | 716 | struct anon_vma *anon_vma = NULL; |
1da177e4 | 717 | struct file *file = vma->vm_file; |
d3737187 | 718 | bool start_changed = false, end_changed = false; |
1da177e4 LT |
719 | long adjust_next = 0; |
720 | int remove_next = 0; | |
721 | ||
722 | if (next && !insert) { | |
287d97ac LT |
723 | struct vm_area_struct *exporter = NULL; |
724 | ||
1da177e4 LT |
725 | if (end >= next->vm_end) { |
726 | /* | |
727 | * vma expands, overlapping all the next, and | |
728 | * perhaps the one after too (mprotect case 6). | |
729 | */ | |
730 | again: remove_next = 1 + (end > next->vm_end); | |
731 | end = next->vm_end; | |
287d97ac | 732 | exporter = next; |
1da177e4 LT |
733 | importer = vma; |
734 | } else if (end > next->vm_start) { | |
735 | /* | |
736 | * vma expands, overlapping part of the next: | |
737 | * mprotect case 5 shifting the boundary up. | |
738 | */ | |
739 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | |
287d97ac | 740 | exporter = next; |
1da177e4 LT |
741 | importer = vma; |
742 | } else if (end < vma->vm_end) { | |
743 | /* | |
744 | * vma shrinks, and !insert tells it's not | |
745 | * split_vma inserting another: so it must be | |
746 | * mprotect case 4 shifting the boundary down. | |
747 | */ | |
cc71aba3 | 748 | adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); |
287d97ac | 749 | exporter = vma; |
1da177e4 LT |
750 | importer = next; |
751 | } | |
1da177e4 | 752 | |
5beb4930 RR |
753 | /* |
754 | * Easily overlooked: when mprotect shifts the boundary, | |
755 | * make sure the expanding vma has anon_vma set if the | |
756 | * shrinking vma had, to cover any anon pages imported. | |
757 | */ | |
287d97ac LT |
758 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
759 | if (anon_vma_clone(importer, exporter)) | |
5beb4930 | 760 | return -ENOMEM; |
287d97ac | 761 | importer->anon_vma = exporter->anon_vma; |
5beb4930 RR |
762 | } |
763 | } | |
764 | ||
1da177e4 LT |
765 | if (file) { |
766 | mapping = file->f_mapping; | |
682968e0 | 767 | if (!(vma->vm_flags & VM_NONLINEAR)) { |
1da177e4 | 768 | root = &mapping->i_mmap; |
cbc91f71 | 769 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); |
682968e0 SD |
770 | |
771 | if (adjust_next) | |
cbc91f71 SD |
772 | uprobe_munmap(next, next->vm_start, |
773 | next->vm_end); | |
682968e0 SD |
774 | } |
775 | ||
3d48ae45 | 776 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 | 777 | if (insert) { |
1da177e4 | 778 | /* |
6b2dbba8 | 779 | * Put into interval tree now, so instantiated pages |
1da177e4 LT |
780 | * are visible to arm/parisc __flush_dcache_page |
781 | * throughout; but we cannot insert into address | |
782 | * space until vma start or end is updated. | |
783 | */ | |
784 | __vma_link_file(insert); | |
785 | } | |
786 | } | |
787 | ||
94fcc585 AA |
788 | vma_adjust_trans_huge(vma, start, end, adjust_next); |
789 | ||
bf181b9f ML |
790 | anon_vma = vma->anon_vma; |
791 | if (!anon_vma && adjust_next) | |
792 | anon_vma = next->anon_vma; | |
793 | if (anon_vma) { | |
81d1b09c SL |
794 | VM_BUG_ON_VMA(adjust_next && next->anon_vma && |
795 | anon_vma != next->anon_vma, next); | |
4fc3f1d6 | 796 | anon_vma_lock_write(anon_vma); |
bf181b9f ML |
797 | anon_vma_interval_tree_pre_update_vma(vma); |
798 | if (adjust_next) | |
799 | anon_vma_interval_tree_pre_update_vma(next); | |
800 | } | |
012f1800 | 801 | |
1da177e4 LT |
802 | if (root) { |
803 | flush_dcache_mmap_lock(mapping); | |
6b2dbba8 | 804 | vma_interval_tree_remove(vma, root); |
1da177e4 | 805 | if (adjust_next) |
6b2dbba8 | 806 | vma_interval_tree_remove(next, root); |
1da177e4 LT |
807 | } |
808 | ||
d3737187 ML |
809 | if (start != vma->vm_start) { |
810 | vma->vm_start = start; | |
811 | start_changed = true; | |
812 | } | |
813 | if (end != vma->vm_end) { | |
814 | vma->vm_end = end; | |
815 | end_changed = true; | |
816 | } | |
1da177e4 LT |
817 | vma->vm_pgoff = pgoff; |
818 | if (adjust_next) { | |
819 | next->vm_start += adjust_next << PAGE_SHIFT; | |
820 | next->vm_pgoff += adjust_next; | |
821 | } | |
822 | ||
823 | if (root) { | |
824 | if (adjust_next) | |
6b2dbba8 ML |
825 | vma_interval_tree_insert(next, root); |
826 | vma_interval_tree_insert(vma, root); | |
1da177e4 LT |
827 | flush_dcache_mmap_unlock(mapping); |
828 | } | |
829 | ||
830 | if (remove_next) { | |
831 | /* | |
832 | * vma_merge has merged next into vma, and needs | |
833 | * us to remove next before dropping the locks. | |
834 | */ | |
835 | __vma_unlink(mm, next, vma); | |
836 | if (file) | |
837 | __remove_shared_vm_struct(next, file, mapping); | |
1da177e4 LT |
838 | } else if (insert) { |
839 | /* | |
840 | * split_vma has split insert from vma, and needs | |
841 | * us to insert it before dropping the locks | |
842 | * (it may either follow vma or precede it). | |
843 | */ | |
844 | __insert_vm_struct(mm, insert); | |
d3737187 ML |
845 | } else { |
846 | if (start_changed) | |
847 | vma_gap_update(vma); | |
848 | if (end_changed) { | |
849 | if (!next) | |
850 | mm->highest_vm_end = end; | |
851 | else if (!adjust_next) | |
852 | vma_gap_update(next); | |
853 | } | |
1da177e4 LT |
854 | } |
855 | ||
bf181b9f ML |
856 | if (anon_vma) { |
857 | anon_vma_interval_tree_post_update_vma(vma); | |
858 | if (adjust_next) | |
859 | anon_vma_interval_tree_post_update_vma(next); | |
08b52706 | 860 | anon_vma_unlock_write(anon_vma); |
bf181b9f | 861 | } |
1da177e4 | 862 | if (mapping) |
3d48ae45 | 863 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 | 864 | |
2b144498 | 865 | if (root) { |
7b2d81d4 | 866 | uprobe_mmap(vma); |
2b144498 SD |
867 | |
868 | if (adjust_next) | |
7b2d81d4 | 869 | uprobe_mmap(next); |
2b144498 SD |
870 | } |
871 | ||
1da177e4 | 872 | if (remove_next) { |
925d1c40 | 873 | if (file) { |
cbc91f71 | 874 | uprobe_munmap(next, next->vm_start, next->vm_end); |
1da177e4 | 875 | fput(file); |
925d1c40 | 876 | } |
5beb4930 RR |
877 | if (next->anon_vma) |
878 | anon_vma_merge(vma, next); | |
1da177e4 | 879 | mm->map_count--; |
3964acd0 | 880 | mpol_put(vma_policy(next)); |
1da177e4 LT |
881 | kmem_cache_free(vm_area_cachep, next); |
882 | /* | |
883 | * In mprotect's case 6 (see comments on vma_merge), | |
884 | * we must remove another next too. It would clutter | |
885 | * up the code too much to do both in one go. | |
886 | */ | |
d3737187 ML |
887 | next = vma->vm_next; |
888 | if (remove_next == 2) | |
1da177e4 | 889 | goto again; |
d3737187 ML |
890 | else if (next) |
891 | vma_gap_update(next); | |
892 | else | |
893 | mm->highest_vm_end = end; | |
1da177e4 | 894 | } |
2b144498 | 895 | if (insert && file) |
7b2d81d4 | 896 | uprobe_mmap(insert); |
1da177e4 LT |
897 | |
898 | validate_mm(mm); | |
5beb4930 RR |
899 | |
900 | return 0; | |
1da177e4 LT |
901 | } |
902 | ||
903 | /* | |
904 | * If the vma has a ->close operation then the driver probably needs to release | |
905 | * per-vma resources, so we don't attempt to merge those. | |
906 | */ | |
1da177e4 LT |
907 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
908 | struct file *file, unsigned long vm_flags) | |
909 | { | |
34228d47 CG |
910 | /* |
911 | * VM_SOFTDIRTY should not prevent from VMA merging, if we | |
912 | * match the flags but dirty bit -- the caller should mark | |
913 | * merged VMA as dirty. If dirty bit won't be excluded from | |
914 | * comparison, we increase pressue on the memory system forcing | |
915 | * the kernel to generate new VMAs when old one could be | |
916 | * extended instead. | |
917 | */ | |
918 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) | |
1da177e4 LT |
919 | return 0; |
920 | if (vma->vm_file != file) | |
921 | return 0; | |
922 | if (vma->vm_ops && vma->vm_ops->close) | |
923 | return 0; | |
924 | return 1; | |
925 | } | |
926 | ||
927 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, | |
965f55de SL |
928 | struct anon_vma *anon_vma2, |
929 | struct vm_area_struct *vma) | |
1da177e4 | 930 | { |
965f55de SL |
931 | /* |
932 | * The list_is_singular() test is to avoid merging VMA cloned from | |
933 | * parents. This can improve scalability caused by anon_vma lock. | |
934 | */ | |
935 | if ((!anon_vma1 || !anon_vma2) && (!vma || | |
936 | list_is_singular(&vma->anon_vma_chain))) | |
937 | return 1; | |
938 | return anon_vma1 == anon_vma2; | |
1da177e4 LT |
939 | } |
940 | ||
941 | /* | |
942 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
943 | * in front of (at a lower virtual address and file offset than) the vma. | |
944 | * | |
945 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
946 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
947 | * | |
948 | * We don't check here for the merged mmap wrapping around the end of pagecache | |
949 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which | |
950 | * wrap, nor mmaps which cover the final page at index -1UL. | |
951 | */ | |
952 | static int | |
953 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, | |
954 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
955 | { | |
956 | if (is_mergeable_vma(vma, file, vm_flags) && | |
965f55de | 957 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 LT |
958 | if (vma->vm_pgoff == vm_pgoff) |
959 | return 1; | |
960 | } | |
961 | return 0; | |
962 | } | |
963 | ||
964 | /* | |
965 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) | |
966 | * beyond (at a higher virtual address and file offset than) the vma. | |
967 | * | |
968 | * We cannot merge two vmas if they have differently assigned (non-NULL) | |
969 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. | |
970 | */ | |
971 | static int | |
972 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, | |
973 | struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) | |
974 | { | |
975 | if (is_mergeable_vma(vma, file, vm_flags) && | |
965f55de | 976 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
1da177e4 | 977 | pgoff_t vm_pglen; |
d6e93217 | 978 | vm_pglen = vma_pages(vma); |
1da177e4 LT |
979 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
980 | return 1; | |
981 | } | |
982 | return 0; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out | |
987 | * whether that can be merged with its predecessor or its successor. | |
988 | * Or both (it neatly fills a hole). | |
989 | * | |
990 | * In most cases - when called for mmap, brk or mremap - [addr,end) is | |
991 | * certain not to be mapped by the time vma_merge is called; but when | |
992 | * called for mprotect, it is certain to be already mapped (either at | |
993 | * an offset within prev, or at the start of next), and the flags of | |
994 | * this area are about to be changed to vm_flags - and the no-change | |
995 | * case has already been eliminated. | |
996 | * | |
997 | * The following mprotect cases have to be considered, where AAAA is | |
998 | * the area passed down from mprotect_fixup, never extending beyond one | |
999 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: | |
1000 | * | |
1001 | * AAAA AAAA AAAA AAAA | |
1002 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX | |
1003 | * cannot merge might become might become might become | |
1004 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or | |
1005 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or | |
1006 | * mremap move: PPPPNNNNNNNN 8 | |
1007 | * AAAA | |
1008 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN | |
1009 | * might become case 1 below case 2 below case 3 below | |
1010 | * | |
1011 | * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: | |
1012 | * mprotect_fixup updates vm_flags & vm_page_prot on successful return. | |
1013 | */ | |
1014 | struct vm_area_struct *vma_merge(struct mm_struct *mm, | |
1015 | struct vm_area_struct *prev, unsigned long addr, | |
1016 | unsigned long end, unsigned long vm_flags, | |
cc71aba3 | 1017 | struct anon_vma *anon_vma, struct file *file, |
1da177e4 LT |
1018 | pgoff_t pgoff, struct mempolicy *policy) |
1019 | { | |
1020 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; | |
1021 | struct vm_area_struct *area, *next; | |
5beb4930 | 1022 | int err; |
1da177e4 LT |
1023 | |
1024 | /* | |
1025 | * We later require that vma->vm_flags == vm_flags, | |
1026 | * so this tests vma->vm_flags & VM_SPECIAL, too. | |
1027 | */ | |
1028 | if (vm_flags & VM_SPECIAL) | |
1029 | return NULL; | |
1030 | ||
1031 | if (prev) | |
1032 | next = prev->vm_next; | |
1033 | else | |
1034 | next = mm->mmap; | |
1035 | area = next; | |
1036 | if (next && next->vm_end == end) /* cases 6, 7, 8 */ | |
1037 | next = next->vm_next; | |
1038 | ||
1039 | /* | |
1040 | * Can it merge with the predecessor? | |
1041 | */ | |
1042 | if (prev && prev->vm_end == addr && | |
cc71aba3 | 1043 | mpol_equal(vma_policy(prev), policy) && |
1da177e4 LT |
1044 | can_vma_merge_after(prev, vm_flags, |
1045 | anon_vma, file, pgoff)) { | |
1046 | /* | |
1047 | * OK, it can. Can we now merge in the successor as well? | |
1048 | */ | |
1049 | if (next && end == next->vm_start && | |
1050 | mpol_equal(policy, vma_policy(next)) && | |
1051 | can_vma_merge_before(next, vm_flags, | |
1052 | anon_vma, file, pgoff+pglen) && | |
1053 | is_mergeable_anon_vma(prev->anon_vma, | |
965f55de | 1054 | next->anon_vma, NULL)) { |
1da177e4 | 1055 | /* cases 1, 6 */ |
5beb4930 | 1056 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
1057 | next->vm_end, prev->vm_pgoff, NULL); |
1058 | } else /* cases 2, 5, 7 */ | |
5beb4930 | 1059 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 | 1060 | end, prev->vm_pgoff, NULL); |
5beb4930 RR |
1061 | if (err) |
1062 | return NULL; | |
b15d00b6 | 1063 | khugepaged_enter_vma_merge(prev); |
1da177e4 LT |
1064 | return prev; |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * Can this new request be merged in front of next? | |
1069 | */ | |
1070 | if (next && end == next->vm_start && | |
cc71aba3 | 1071 | mpol_equal(policy, vma_policy(next)) && |
1da177e4 LT |
1072 | can_vma_merge_before(next, vm_flags, |
1073 | anon_vma, file, pgoff+pglen)) { | |
1074 | if (prev && addr < prev->vm_end) /* case 4 */ | |
5beb4930 | 1075 | err = vma_adjust(prev, prev->vm_start, |
1da177e4 LT |
1076 | addr, prev->vm_pgoff, NULL); |
1077 | else /* cases 3, 8 */ | |
5beb4930 | 1078 | err = vma_adjust(area, addr, next->vm_end, |
1da177e4 | 1079 | next->vm_pgoff - pglen, NULL); |
5beb4930 RR |
1080 | if (err) |
1081 | return NULL; | |
b15d00b6 | 1082 | khugepaged_enter_vma_merge(area); |
1da177e4 LT |
1083 | return area; |
1084 | } | |
1085 | ||
1086 | return NULL; | |
1087 | } | |
1088 | ||
d0e9fe17 LT |
1089 | /* |
1090 | * Rough compatbility check to quickly see if it's even worth looking | |
1091 | * at sharing an anon_vma. | |
1092 | * | |
1093 | * They need to have the same vm_file, and the flags can only differ | |
1094 | * in things that mprotect may change. | |
1095 | * | |
1096 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | |
1097 | * we can merge the two vma's. For example, we refuse to merge a vma if | |
1098 | * there is a vm_ops->close() function, because that indicates that the | |
1099 | * driver is doing some kind of reference counting. But that doesn't | |
1100 | * really matter for the anon_vma sharing case. | |
1101 | */ | |
1102 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | |
1103 | { | |
1104 | return a->vm_end == b->vm_start && | |
1105 | mpol_equal(vma_policy(a), vma_policy(b)) && | |
1106 | a->vm_file == b->vm_file && | |
34228d47 | 1107 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && |
d0e9fe17 LT |
1108 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
1109 | } | |
1110 | ||
1111 | /* | |
1112 | * Do some basic sanity checking to see if we can re-use the anon_vma | |
1113 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | |
1114 | * the same as 'old', the other will be the new one that is trying | |
1115 | * to share the anon_vma. | |
1116 | * | |
1117 | * NOTE! This runs with mm_sem held for reading, so it is possible that | |
1118 | * the anon_vma of 'old' is concurrently in the process of being set up | |
1119 | * by another page fault trying to merge _that_. But that's ok: if it | |
1120 | * is being set up, that automatically means that it will be a singleton | |
1121 | * acceptable for merging, so we can do all of this optimistically. But | |
1122 | * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. | |
1123 | * | |
1124 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | |
1125 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | |
1126 | * is to return an anon_vma that is "complex" due to having gone through | |
1127 | * a fork). | |
1128 | * | |
1129 | * We also make sure that the two vma's are compatible (adjacent, | |
1130 | * and with the same memory policies). That's all stable, even with just | |
1131 | * a read lock on the mm_sem. | |
1132 | */ | |
1133 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | |
1134 | { | |
1135 | if (anon_vma_compatible(a, b)) { | |
1136 | struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); | |
1137 | ||
1138 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | |
1139 | return anon_vma; | |
1140 | } | |
1141 | return NULL; | |
1142 | } | |
1143 | ||
1da177e4 LT |
1144 | /* |
1145 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | |
1146 | * neighbouring vmas for a suitable anon_vma, before it goes off | |
1147 | * to allocate a new anon_vma. It checks because a repetitive | |
1148 | * sequence of mprotects and faults may otherwise lead to distinct | |
1149 | * anon_vmas being allocated, preventing vma merge in subsequent | |
1150 | * mprotect. | |
1151 | */ | |
1152 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | |
1153 | { | |
d0e9fe17 | 1154 | struct anon_vma *anon_vma; |
1da177e4 | 1155 | struct vm_area_struct *near; |
1da177e4 LT |
1156 | |
1157 | near = vma->vm_next; | |
1158 | if (!near) | |
1159 | goto try_prev; | |
1160 | ||
d0e9fe17 LT |
1161 | anon_vma = reusable_anon_vma(near, vma, near); |
1162 | if (anon_vma) | |
1163 | return anon_vma; | |
1da177e4 | 1164 | try_prev: |
9be34c9d | 1165 | near = vma->vm_prev; |
1da177e4 LT |
1166 | if (!near) |
1167 | goto none; | |
1168 | ||
d0e9fe17 LT |
1169 | anon_vma = reusable_anon_vma(near, near, vma); |
1170 | if (anon_vma) | |
1171 | return anon_vma; | |
1da177e4 LT |
1172 | none: |
1173 | /* | |
1174 | * There's no absolute need to look only at touching neighbours: | |
1175 | * we could search further afield for "compatible" anon_vmas. | |
1176 | * But it would probably just be a waste of time searching, | |
1177 | * or lead to too many vmas hanging off the same anon_vma. | |
1178 | * We're trying to allow mprotect remerging later on, | |
1179 | * not trying to minimize memory used for anon_vmas. | |
1180 | */ | |
1181 | return NULL; | |
1182 | } | |
1183 | ||
1184 | #ifdef CONFIG_PROC_FS | |
ab50b8ed | 1185 | void vm_stat_account(struct mm_struct *mm, unsigned long flags, |
1da177e4 LT |
1186 | struct file *file, long pages) |
1187 | { | |
1188 | const unsigned long stack_flags | |
1189 | = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); | |
1190 | ||
44de9d0c HS |
1191 | mm->total_vm += pages; |
1192 | ||
1da177e4 LT |
1193 | if (file) { |
1194 | mm->shared_vm += pages; | |
1195 | if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) | |
1196 | mm->exec_vm += pages; | |
1197 | } else if (flags & stack_flags) | |
1198 | mm->stack_vm += pages; | |
1da177e4 LT |
1199 | } |
1200 | #endif /* CONFIG_PROC_FS */ | |
1201 | ||
40401530 AV |
1202 | /* |
1203 | * If a hint addr is less than mmap_min_addr change hint to be as | |
1204 | * low as possible but still greater than mmap_min_addr | |
1205 | */ | |
1206 | static inline unsigned long round_hint_to_min(unsigned long hint) | |
1207 | { | |
1208 | hint &= PAGE_MASK; | |
1209 | if (((void *)hint != NULL) && | |
1210 | (hint < mmap_min_addr)) | |
1211 | return PAGE_ALIGN(mmap_min_addr); | |
1212 | return hint; | |
1213 | } | |
1214 | ||
363ee17f DB |
1215 | static inline int mlock_future_check(struct mm_struct *mm, |
1216 | unsigned long flags, | |
1217 | unsigned long len) | |
1218 | { | |
1219 | unsigned long locked, lock_limit; | |
1220 | ||
1221 | /* mlock MCL_FUTURE? */ | |
1222 | if (flags & VM_LOCKED) { | |
1223 | locked = len >> PAGE_SHIFT; | |
1224 | locked += mm->locked_vm; | |
1225 | lock_limit = rlimit(RLIMIT_MEMLOCK); | |
1226 | lock_limit >>= PAGE_SHIFT; | |
1227 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
1228 | return -EAGAIN; | |
1229 | } | |
1230 | return 0; | |
1231 | } | |
1232 | ||
1da177e4 | 1233 | /* |
27f5de79 | 1234 | * The caller must hold down_write(¤t->mm->mmap_sem). |
1da177e4 LT |
1235 | */ |
1236 | ||
e3fc629d | 1237 | unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, |
1da177e4 | 1238 | unsigned long len, unsigned long prot, |
bebeb3d6 | 1239 | unsigned long flags, unsigned long pgoff, |
41badc15 | 1240 | unsigned long *populate) |
1da177e4 | 1241 | { |
cc71aba3 | 1242 | struct mm_struct *mm = current->mm; |
ca16d140 | 1243 | vm_flags_t vm_flags; |
1da177e4 | 1244 | |
41badc15 | 1245 | *populate = 0; |
bebeb3d6 | 1246 | |
1da177e4 LT |
1247 | /* |
1248 | * Does the application expect PROT_READ to imply PROT_EXEC? | |
1249 | * | |
1250 | * (the exception is when the underlying filesystem is noexec | |
1251 | * mounted, in which case we dont add PROT_EXEC.) | |
1252 | */ | |
1253 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) | |
d3ac7f89 | 1254 | if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) |
1da177e4 LT |
1255 | prot |= PROT_EXEC; |
1256 | ||
1257 | if (!len) | |
1258 | return -EINVAL; | |
1259 | ||
7cd94146 EP |
1260 | if (!(flags & MAP_FIXED)) |
1261 | addr = round_hint_to_min(addr); | |
1262 | ||
1da177e4 LT |
1263 | /* Careful about overflows.. */ |
1264 | len = PAGE_ALIGN(len); | |
9206de95 | 1265 | if (!len) |
1da177e4 LT |
1266 | return -ENOMEM; |
1267 | ||
1268 | /* offset overflow? */ | |
1269 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | |
cc71aba3 | 1270 | return -EOVERFLOW; |
1da177e4 LT |
1271 | |
1272 | /* Too many mappings? */ | |
1273 | if (mm->map_count > sysctl_max_map_count) | |
1274 | return -ENOMEM; | |
1275 | ||
1276 | /* Obtain the address to map to. we verify (or select) it and ensure | |
1277 | * that it represents a valid section of the address space. | |
1278 | */ | |
1279 | addr = get_unmapped_area(file, addr, len, pgoff, flags); | |
1280 | if (addr & ~PAGE_MASK) | |
1281 | return addr; | |
1282 | ||
1283 | /* Do simple checking here so the lower-level routines won't have | |
1284 | * to. we assume access permissions have been handled by the open | |
1285 | * of the memory object, so we don't do any here. | |
1286 | */ | |
1287 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | | |
1288 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; | |
1289 | ||
cdf7b341 | 1290 | if (flags & MAP_LOCKED) |
1da177e4 LT |
1291 | if (!can_do_mlock()) |
1292 | return -EPERM; | |
ba470de4 | 1293 | |
363ee17f DB |
1294 | if (mlock_future_check(mm, vm_flags, len)) |
1295 | return -EAGAIN; | |
1da177e4 | 1296 | |
1da177e4 | 1297 | if (file) { |
077bf22b ON |
1298 | struct inode *inode = file_inode(file); |
1299 | ||
1da177e4 LT |
1300 | switch (flags & MAP_TYPE) { |
1301 | case MAP_SHARED: | |
1302 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) | |
1303 | return -EACCES; | |
1304 | ||
1305 | /* | |
1306 | * Make sure we don't allow writing to an append-only | |
1307 | * file.. | |
1308 | */ | |
1309 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) | |
1310 | return -EACCES; | |
1311 | ||
1312 | /* | |
1313 | * Make sure there are no mandatory locks on the file. | |
1314 | */ | |
d7a06983 | 1315 | if (locks_verify_locked(file)) |
1da177e4 LT |
1316 | return -EAGAIN; |
1317 | ||
1318 | vm_flags |= VM_SHARED | VM_MAYSHARE; | |
1319 | if (!(file->f_mode & FMODE_WRITE)) | |
1320 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); | |
1321 | ||
1322 | /* fall through */ | |
1323 | case MAP_PRIVATE: | |
1324 | if (!(file->f_mode & FMODE_READ)) | |
1325 | return -EACCES; | |
d3ac7f89 | 1326 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
80c5606c LT |
1327 | if (vm_flags & VM_EXEC) |
1328 | return -EPERM; | |
1329 | vm_flags &= ~VM_MAYEXEC; | |
1330 | } | |
80c5606c | 1331 | |
72c2d531 | 1332 | if (!file->f_op->mmap) |
80c5606c | 1333 | return -ENODEV; |
b2c56e4f ON |
1334 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1335 | return -EINVAL; | |
1da177e4 LT |
1336 | break; |
1337 | ||
1338 | default: | |
1339 | return -EINVAL; | |
1340 | } | |
1341 | } else { | |
1342 | switch (flags & MAP_TYPE) { | |
1343 | case MAP_SHARED: | |
b2c56e4f ON |
1344 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
1345 | return -EINVAL; | |
ce363942 TH |
1346 | /* |
1347 | * Ignore pgoff. | |
1348 | */ | |
1349 | pgoff = 0; | |
1da177e4 LT |
1350 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
1351 | break; | |
1352 | case MAP_PRIVATE: | |
1353 | /* | |
1354 | * Set pgoff according to addr for anon_vma. | |
1355 | */ | |
1356 | pgoff = addr >> PAGE_SHIFT; | |
1357 | break; | |
1358 | default: | |
1359 | return -EINVAL; | |
1360 | } | |
1361 | } | |
1362 | ||
c22c0d63 ML |
1363 | /* |
1364 | * Set 'VM_NORESERVE' if we should not account for the | |
1365 | * memory use of this mapping. | |
1366 | */ | |
1367 | if (flags & MAP_NORESERVE) { | |
1368 | /* We honor MAP_NORESERVE if allowed to overcommit */ | |
1369 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) | |
1370 | vm_flags |= VM_NORESERVE; | |
1371 | ||
1372 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ | |
1373 | if (file && is_file_hugepages(file)) | |
1374 | vm_flags |= VM_NORESERVE; | |
1375 | } | |
1376 | ||
1377 | addr = mmap_region(file, addr, len, vm_flags, pgoff); | |
09a9f1d2 ML |
1378 | if (!IS_ERR_VALUE(addr) && |
1379 | ((vm_flags & VM_LOCKED) || | |
1380 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) | |
41badc15 | 1381 | *populate = len; |
bebeb3d6 | 1382 | return addr; |
0165ab44 | 1383 | } |
6be5ceb0 | 1384 | |
66f0dc48 HD |
1385 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1386 | unsigned long, prot, unsigned long, flags, | |
1387 | unsigned long, fd, unsigned long, pgoff) | |
1388 | { | |
1389 | struct file *file = NULL; | |
1390 | unsigned long retval = -EBADF; | |
1391 | ||
1392 | if (!(flags & MAP_ANONYMOUS)) { | |
120a795d | 1393 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1394 | file = fget(fd); |
1395 | if (!file) | |
1396 | goto out; | |
af73e4d9 NH |
1397 | if (is_file_hugepages(file)) |
1398 | len = ALIGN(len, huge_page_size(hstate_file(file))); | |
493af578 JE |
1399 | retval = -EINVAL; |
1400 | if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) | |
1401 | goto out_fput; | |
66f0dc48 HD |
1402 | } else if (flags & MAP_HUGETLB) { |
1403 | struct user_struct *user = NULL; | |
c103a4dc | 1404 | struct hstate *hs; |
af73e4d9 | 1405 | |
c103a4dc | 1406 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); |
091d0d55 LZ |
1407 | if (!hs) |
1408 | return -EINVAL; | |
1409 | ||
1410 | len = ALIGN(len, huge_page_size(hs)); | |
66f0dc48 HD |
1411 | /* |
1412 | * VM_NORESERVE is used because the reservations will be | |
1413 | * taken when vm_ops->mmap() is called | |
1414 | * A dummy user value is used because we are not locking | |
1415 | * memory so no accounting is necessary | |
1416 | */ | |
af73e4d9 | 1417 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
42d7395f AK |
1418 | VM_NORESERVE, |
1419 | &user, HUGETLB_ANONHUGE_INODE, | |
1420 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); | |
66f0dc48 HD |
1421 | if (IS_ERR(file)) |
1422 | return PTR_ERR(file); | |
1423 | } | |
1424 | ||
1425 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1426 | ||
eb36c587 | 1427 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
493af578 | 1428 | out_fput: |
66f0dc48 HD |
1429 | if (file) |
1430 | fput(file); | |
1431 | out: | |
1432 | return retval; | |
1433 | } | |
1434 | ||
a4679373 CH |
1435 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1436 | struct mmap_arg_struct { | |
1437 | unsigned long addr; | |
1438 | unsigned long len; | |
1439 | unsigned long prot; | |
1440 | unsigned long flags; | |
1441 | unsigned long fd; | |
1442 | unsigned long offset; | |
1443 | }; | |
1444 | ||
1445 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1446 | { | |
1447 | struct mmap_arg_struct a; | |
1448 | ||
1449 | if (copy_from_user(&a, arg, sizeof(a))) | |
1450 | return -EFAULT; | |
1451 | if (a.offset & ~PAGE_MASK) | |
1452 | return -EINVAL; | |
1453 | ||
1454 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1455 | a.offset >> PAGE_SHIFT); | |
1456 | } | |
1457 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1458 | ||
4e950f6f AD |
1459 | /* |
1460 | * Some shared mappigns will want the pages marked read-only | |
1461 | * to track write events. If so, we'll downgrade vm_page_prot | |
1462 | * to the private version (using protection_map[] without the | |
1463 | * VM_SHARED bit). | |
1464 | */ | |
1465 | int vma_wants_writenotify(struct vm_area_struct *vma) | |
1466 | { | |
ca16d140 | 1467 | vm_flags_t vm_flags = vma->vm_flags; |
4e950f6f AD |
1468 | |
1469 | /* If it was private or non-writable, the write bit is already clear */ | |
1470 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) | |
1471 | return 0; | |
1472 | ||
1473 | /* The backer wishes to know when pages are first written to? */ | |
1474 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) | |
1475 | return 1; | |
1476 | ||
1477 | /* The open routine did something to the protections already? */ | |
1478 | if (pgprot_val(vma->vm_page_prot) != | |
3ed75eb8 | 1479 | pgprot_val(vm_get_page_prot(vm_flags))) |
4e950f6f AD |
1480 | return 0; |
1481 | ||
1482 | /* Specialty mapping? */ | |
4b6e1e37 | 1483 | if (vm_flags & VM_PFNMAP) |
4e950f6f AD |
1484 | return 0; |
1485 | ||
1486 | /* Can the mapping track the dirty pages? */ | |
1487 | return vma->vm_file && vma->vm_file->f_mapping && | |
1488 | mapping_cap_account_dirty(vma->vm_file->f_mapping); | |
1489 | } | |
1490 | ||
fc8744ad LT |
1491 | /* |
1492 | * We account for memory if it's a private writeable mapping, | |
5a6fe125 | 1493 | * not hugepages and VM_NORESERVE wasn't set. |
fc8744ad | 1494 | */ |
ca16d140 | 1495 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
fc8744ad | 1496 | { |
5a6fe125 MG |
1497 | /* |
1498 | * hugetlb has its own accounting separate from the core VM | |
1499 | * VM_HUGETLB may not be set yet so we cannot check for that flag. | |
1500 | */ | |
1501 | if (file && is_file_hugepages(file)) | |
1502 | return 0; | |
1503 | ||
fc8744ad LT |
1504 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
1505 | } | |
1506 | ||
0165ab44 | 1507 | unsigned long mmap_region(struct file *file, unsigned long addr, |
c22c0d63 | 1508 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) |
0165ab44 MS |
1509 | { |
1510 | struct mm_struct *mm = current->mm; | |
1511 | struct vm_area_struct *vma, *prev; | |
0165ab44 MS |
1512 | int error; |
1513 | struct rb_node **rb_link, *rb_parent; | |
1514 | unsigned long charged = 0; | |
0165ab44 | 1515 | |
e8420a8e CH |
1516 | /* Check against address space limit. */ |
1517 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) { | |
1518 | unsigned long nr_pages; | |
1519 | ||
1520 | /* | |
1521 | * MAP_FIXED may remove pages of mappings that intersects with | |
1522 | * requested mapping. Account for the pages it would unmap. | |
1523 | */ | |
1524 | if (!(vm_flags & MAP_FIXED)) | |
1525 | return -ENOMEM; | |
1526 | ||
1527 | nr_pages = count_vma_pages_range(mm, addr, addr + len); | |
1528 | ||
1529 | if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages)) | |
1530 | return -ENOMEM; | |
1531 | } | |
1532 | ||
1da177e4 LT |
1533 | /* Clear old maps */ |
1534 | error = -ENOMEM; | |
1535 | munmap_back: | |
6597d783 | 1536 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { |
1da177e4 LT |
1537 | if (do_munmap(mm, addr, len)) |
1538 | return -ENOMEM; | |
1539 | goto munmap_back; | |
1540 | } | |
1541 | ||
fc8744ad LT |
1542 | /* |
1543 | * Private writable mapping: check memory availability | |
1544 | */ | |
5a6fe125 | 1545 | if (accountable_mapping(file, vm_flags)) { |
fc8744ad | 1546 | charged = len >> PAGE_SHIFT; |
191c5424 | 1547 | if (security_vm_enough_memory_mm(mm, charged)) |
fc8744ad LT |
1548 | return -ENOMEM; |
1549 | vm_flags |= VM_ACCOUNT; | |
1da177e4 LT |
1550 | } |
1551 | ||
1552 | /* | |
de33c8db | 1553 | * Can we just expand an old mapping? |
1da177e4 | 1554 | */ |
de33c8db LT |
1555 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); |
1556 | if (vma) | |
1557 | goto out; | |
1da177e4 LT |
1558 | |
1559 | /* | |
1560 | * Determine the object being mapped and call the appropriate | |
1561 | * specific mapper. the address has already been validated, but | |
1562 | * not unmapped, but the maps are removed from the list. | |
1563 | */ | |
c5e3b83e | 1564 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
1565 | if (!vma) { |
1566 | error = -ENOMEM; | |
1567 | goto unacct_error; | |
1568 | } | |
1da177e4 LT |
1569 | |
1570 | vma->vm_mm = mm; | |
1571 | vma->vm_start = addr; | |
1572 | vma->vm_end = addr + len; | |
1573 | vma->vm_flags = vm_flags; | |
3ed75eb8 | 1574 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
1da177e4 | 1575 | vma->vm_pgoff = pgoff; |
5beb4930 | 1576 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
1577 | |
1578 | if (file) { | |
1da177e4 LT |
1579 | if (vm_flags & VM_DENYWRITE) { |
1580 | error = deny_write_access(file); | |
1581 | if (error) | |
1582 | goto free_vma; | |
1da177e4 | 1583 | } |
4bb5f5d9 DH |
1584 | if (vm_flags & VM_SHARED) { |
1585 | error = mapping_map_writable(file->f_mapping); | |
1586 | if (error) | |
1587 | goto allow_write_and_free_vma; | |
1588 | } | |
1589 | ||
1590 | /* ->mmap() can change vma->vm_file, but must guarantee that | |
1591 | * vma_link() below can deny write-access if VM_DENYWRITE is set | |
1592 | * and map writably if VM_SHARED is set. This usually means the | |
1593 | * new file must not have been exposed to user-space, yet. | |
1594 | */ | |
cb0942b8 | 1595 | vma->vm_file = get_file(file); |
1da177e4 LT |
1596 | error = file->f_op->mmap(file, vma); |
1597 | if (error) | |
1598 | goto unmap_and_free_vma; | |
f8dbf0a7 HS |
1599 | |
1600 | /* Can addr have changed?? | |
1601 | * | |
1602 | * Answer: Yes, several device drivers can do it in their | |
1603 | * f_op->mmap method. -DaveM | |
2897b4d2 JK |
1604 | * Bug: If addr is changed, prev, rb_link, rb_parent should |
1605 | * be updated for vma_link() | |
f8dbf0a7 | 1606 | */ |
2897b4d2 JK |
1607 | WARN_ON_ONCE(addr != vma->vm_start); |
1608 | ||
f8dbf0a7 | 1609 | addr = vma->vm_start; |
f8dbf0a7 | 1610 | vm_flags = vma->vm_flags; |
1da177e4 LT |
1611 | } else if (vm_flags & VM_SHARED) { |
1612 | error = shmem_zero_setup(vma); | |
1613 | if (error) | |
1614 | goto free_vma; | |
1615 | } | |
1616 | ||
c9d0bf24 MD |
1617 | if (vma_wants_writenotify(vma)) { |
1618 | pgprot_t pprot = vma->vm_page_prot; | |
1619 | ||
1620 | /* Can vma->vm_page_prot have changed?? | |
1621 | * | |
1622 | * Answer: Yes, drivers may have changed it in their | |
1623 | * f_op->mmap method. | |
1624 | * | |
1625 | * Ensures that vmas marked as uncached stay that way. | |
1626 | */ | |
1ddd439e | 1627 | vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); |
c9d0bf24 MD |
1628 | if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot))) |
1629 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
1630 | } | |
d08b3851 | 1631 | |
de33c8db | 1632 | vma_link(mm, vma, prev, rb_link, rb_parent); |
4d3d5b41 | 1633 | /* Once vma denies write, undo our temporary denial count */ |
4bb5f5d9 DH |
1634 | if (file) { |
1635 | if (vm_flags & VM_SHARED) | |
1636 | mapping_unmap_writable(file->f_mapping); | |
1637 | if (vm_flags & VM_DENYWRITE) | |
1638 | allow_write_access(file); | |
1639 | } | |
e8686772 | 1640 | file = vma->vm_file; |
4d3d5b41 | 1641 | out: |
cdd6c482 | 1642 | perf_event_mmap(vma); |
0a4a9391 | 1643 | |
ab50b8ed | 1644 | vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); |
1da177e4 | 1645 | if (vm_flags & VM_LOCKED) { |
bebeb3d6 ML |
1646 | if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || |
1647 | vma == get_gate_vma(current->mm))) | |
06f9d8c2 | 1648 | mm->locked_vm += (len >> PAGE_SHIFT); |
bebeb3d6 ML |
1649 | else |
1650 | vma->vm_flags &= ~VM_LOCKED; | |
1651 | } | |
2b144498 | 1652 | |
c7a3a88c ON |
1653 | if (file) |
1654 | uprobe_mmap(vma); | |
2b144498 | 1655 | |
d9104d1c CG |
1656 | /* |
1657 | * New (or expanded) vma always get soft dirty status. | |
1658 | * Otherwise user-space soft-dirty page tracker won't | |
1659 | * be able to distinguish situation when vma area unmapped, | |
1660 | * then new mapped in-place (which must be aimed as | |
1661 | * a completely new data area). | |
1662 | */ | |
1663 | vma->vm_flags |= VM_SOFTDIRTY; | |
1664 | ||
1da177e4 LT |
1665 | return addr; |
1666 | ||
1667 | unmap_and_free_vma: | |
1da177e4 LT |
1668 | vma->vm_file = NULL; |
1669 | fput(file); | |
1670 | ||
1671 | /* Undo any partial mapping done by a device driver. */ | |
e0da382c HD |
1672 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
1673 | charged = 0; | |
4bb5f5d9 DH |
1674 | if (vm_flags & VM_SHARED) |
1675 | mapping_unmap_writable(file->f_mapping); | |
1676 | allow_write_and_free_vma: | |
1677 | if (vm_flags & VM_DENYWRITE) | |
1678 | allow_write_access(file); | |
1da177e4 LT |
1679 | free_vma: |
1680 | kmem_cache_free(vm_area_cachep, vma); | |
1681 | unacct_error: | |
1682 | if (charged) | |
1683 | vm_unacct_memory(charged); | |
1684 | return error; | |
1685 | } | |
1686 | ||
db4fbfb9 ML |
1687 | unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
1688 | { | |
1689 | /* | |
1690 | * We implement the search by looking for an rbtree node that | |
1691 | * immediately follows a suitable gap. That is, | |
1692 | * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; | |
1693 | * - gap_end = vma->vm_start >= info->low_limit + length; | |
1694 | * - gap_end - gap_start >= length | |
1695 | */ | |
1696 | ||
1697 | struct mm_struct *mm = current->mm; | |
1698 | struct vm_area_struct *vma; | |
1699 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1700 | ||
1701 | /* Adjust search length to account for worst case alignment overhead */ | |
1702 | length = info->length + info->align_mask; | |
1703 | if (length < info->length) | |
1704 | return -ENOMEM; | |
1705 | ||
1706 | /* Adjust search limits by the desired length */ | |
1707 | if (info->high_limit < length) | |
1708 | return -ENOMEM; | |
1709 | high_limit = info->high_limit - length; | |
1710 | ||
1711 | if (info->low_limit > high_limit) | |
1712 | return -ENOMEM; | |
1713 | low_limit = info->low_limit + length; | |
1714 | ||
1715 | /* Check if rbtree root looks promising */ | |
1716 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1717 | goto check_highest; | |
1718 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1719 | if (vma->rb_subtree_gap < length) | |
1720 | goto check_highest; | |
1721 | ||
1722 | while (true) { | |
1723 | /* Visit left subtree if it looks promising */ | |
1724 | gap_end = vma->vm_start; | |
1725 | if (gap_end >= low_limit && vma->vm_rb.rb_left) { | |
1726 | struct vm_area_struct *left = | |
1727 | rb_entry(vma->vm_rb.rb_left, | |
1728 | struct vm_area_struct, vm_rb); | |
1729 | if (left->rb_subtree_gap >= length) { | |
1730 | vma = left; | |
1731 | continue; | |
1732 | } | |
1733 | } | |
1734 | ||
1735 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1736 | check_current: | |
1737 | /* Check if current node has a suitable gap */ | |
1738 | if (gap_start > high_limit) | |
1739 | return -ENOMEM; | |
1740 | if (gap_end >= low_limit && gap_end - gap_start >= length) | |
1741 | goto found; | |
1742 | ||
1743 | /* Visit right subtree if it looks promising */ | |
1744 | if (vma->vm_rb.rb_right) { | |
1745 | struct vm_area_struct *right = | |
1746 | rb_entry(vma->vm_rb.rb_right, | |
1747 | struct vm_area_struct, vm_rb); | |
1748 | if (right->rb_subtree_gap >= length) { | |
1749 | vma = right; | |
1750 | continue; | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | /* Go back up the rbtree to find next candidate node */ | |
1755 | while (true) { | |
1756 | struct rb_node *prev = &vma->vm_rb; | |
1757 | if (!rb_parent(prev)) | |
1758 | goto check_highest; | |
1759 | vma = rb_entry(rb_parent(prev), | |
1760 | struct vm_area_struct, vm_rb); | |
1761 | if (prev == vma->vm_rb.rb_left) { | |
1762 | gap_start = vma->vm_prev->vm_end; | |
1763 | gap_end = vma->vm_start; | |
1764 | goto check_current; | |
1765 | } | |
1766 | } | |
1767 | } | |
1768 | ||
1769 | check_highest: | |
1770 | /* Check highest gap, which does not precede any rbtree node */ | |
1771 | gap_start = mm->highest_vm_end; | |
1772 | gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ | |
1773 | if (gap_start > high_limit) | |
1774 | return -ENOMEM; | |
1775 | ||
1776 | found: | |
1777 | /* We found a suitable gap. Clip it with the original low_limit. */ | |
1778 | if (gap_start < info->low_limit) | |
1779 | gap_start = info->low_limit; | |
1780 | ||
1781 | /* Adjust gap address to the desired alignment */ | |
1782 | gap_start += (info->align_offset - gap_start) & info->align_mask; | |
1783 | ||
1784 | VM_BUG_ON(gap_start + info->length > info->high_limit); | |
1785 | VM_BUG_ON(gap_start + info->length > gap_end); | |
1786 | return gap_start; | |
1787 | } | |
1788 | ||
1789 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) | |
1790 | { | |
1791 | struct mm_struct *mm = current->mm; | |
1792 | struct vm_area_struct *vma; | |
1793 | unsigned long length, low_limit, high_limit, gap_start, gap_end; | |
1794 | ||
1795 | /* Adjust search length to account for worst case alignment overhead */ | |
1796 | length = info->length + info->align_mask; | |
1797 | if (length < info->length) | |
1798 | return -ENOMEM; | |
1799 | ||
1800 | /* | |
1801 | * Adjust search limits by the desired length. | |
1802 | * See implementation comment at top of unmapped_area(). | |
1803 | */ | |
1804 | gap_end = info->high_limit; | |
1805 | if (gap_end < length) | |
1806 | return -ENOMEM; | |
1807 | high_limit = gap_end - length; | |
1808 | ||
1809 | if (info->low_limit > high_limit) | |
1810 | return -ENOMEM; | |
1811 | low_limit = info->low_limit + length; | |
1812 | ||
1813 | /* Check highest gap, which does not precede any rbtree node */ | |
1814 | gap_start = mm->highest_vm_end; | |
1815 | if (gap_start <= high_limit) | |
1816 | goto found_highest; | |
1817 | ||
1818 | /* Check if rbtree root looks promising */ | |
1819 | if (RB_EMPTY_ROOT(&mm->mm_rb)) | |
1820 | return -ENOMEM; | |
1821 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); | |
1822 | if (vma->rb_subtree_gap < length) | |
1823 | return -ENOMEM; | |
1824 | ||
1825 | while (true) { | |
1826 | /* Visit right subtree if it looks promising */ | |
1827 | gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; | |
1828 | if (gap_start <= high_limit && vma->vm_rb.rb_right) { | |
1829 | struct vm_area_struct *right = | |
1830 | rb_entry(vma->vm_rb.rb_right, | |
1831 | struct vm_area_struct, vm_rb); | |
1832 | if (right->rb_subtree_gap >= length) { | |
1833 | vma = right; | |
1834 | continue; | |
1835 | } | |
1836 | } | |
1837 | ||
1838 | check_current: | |
1839 | /* Check if current node has a suitable gap */ | |
1840 | gap_end = vma->vm_start; | |
1841 | if (gap_end < low_limit) | |
1842 | return -ENOMEM; | |
1843 | if (gap_start <= high_limit && gap_end - gap_start >= length) | |
1844 | goto found; | |
1845 | ||
1846 | /* Visit left subtree if it looks promising */ | |
1847 | if (vma->vm_rb.rb_left) { | |
1848 | struct vm_area_struct *left = | |
1849 | rb_entry(vma->vm_rb.rb_left, | |
1850 | struct vm_area_struct, vm_rb); | |
1851 | if (left->rb_subtree_gap >= length) { | |
1852 | vma = left; | |
1853 | continue; | |
1854 | } | |
1855 | } | |
1856 | ||
1857 | /* Go back up the rbtree to find next candidate node */ | |
1858 | while (true) { | |
1859 | struct rb_node *prev = &vma->vm_rb; | |
1860 | if (!rb_parent(prev)) | |
1861 | return -ENOMEM; | |
1862 | vma = rb_entry(rb_parent(prev), | |
1863 | struct vm_area_struct, vm_rb); | |
1864 | if (prev == vma->vm_rb.rb_right) { | |
1865 | gap_start = vma->vm_prev ? | |
1866 | vma->vm_prev->vm_end : 0; | |
1867 | goto check_current; | |
1868 | } | |
1869 | } | |
1870 | } | |
1871 | ||
1872 | found: | |
1873 | /* We found a suitable gap. Clip it with the original high_limit. */ | |
1874 | if (gap_end > info->high_limit) | |
1875 | gap_end = info->high_limit; | |
1876 | ||
1877 | found_highest: | |
1878 | /* Compute highest gap address at the desired alignment */ | |
1879 | gap_end -= info->length; | |
1880 | gap_end -= (gap_end - info->align_offset) & info->align_mask; | |
1881 | ||
1882 | VM_BUG_ON(gap_end < info->low_limit); | |
1883 | VM_BUG_ON(gap_end < gap_start); | |
1884 | return gap_end; | |
1885 | } | |
1886 | ||
1da177e4 LT |
1887 | /* Get an address range which is currently unmapped. |
1888 | * For shmat() with addr=0. | |
1889 | * | |
1890 | * Ugly calling convention alert: | |
1891 | * Return value with the low bits set means error value, | |
1892 | * ie | |
1893 | * if (ret & ~PAGE_MASK) | |
1894 | * error = ret; | |
1895 | * | |
1896 | * This function "knows" that -ENOMEM has the bits set. | |
1897 | */ | |
1898 | #ifndef HAVE_ARCH_UNMAPPED_AREA | |
1899 | unsigned long | |
1900 | arch_get_unmapped_area(struct file *filp, unsigned long addr, | |
1901 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1902 | { | |
1903 | struct mm_struct *mm = current->mm; | |
1904 | struct vm_area_struct *vma; | |
db4fbfb9 | 1905 | struct vm_unmapped_area_info info; |
1da177e4 | 1906 | |
2afc745f | 1907 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1908 | return -ENOMEM; |
1909 | ||
06abdfb4 BH |
1910 | if (flags & MAP_FIXED) |
1911 | return addr; | |
1912 | ||
1da177e4 LT |
1913 | if (addr) { |
1914 | addr = PAGE_ALIGN(addr); | |
1915 | vma = find_vma(mm, addr); | |
2afc745f | 1916 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1917 | (!vma || addr + len <= vma->vm_start)) |
1918 | return addr; | |
1919 | } | |
1da177e4 | 1920 | |
db4fbfb9 ML |
1921 | info.flags = 0; |
1922 | info.length = len; | |
4e99b021 | 1923 | info.low_limit = mm->mmap_base; |
db4fbfb9 ML |
1924 | info.high_limit = TASK_SIZE; |
1925 | info.align_mask = 0; | |
1926 | return vm_unmapped_area(&info); | |
1da177e4 | 1927 | } |
cc71aba3 | 1928 | #endif |
1da177e4 | 1929 | |
1da177e4 LT |
1930 | /* |
1931 | * This mmap-allocator allocates new areas top-down from below the | |
1932 | * stack's low limit (the base): | |
1933 | */ | |
1934 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN | |
1935 | unsigned long | |
1936 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, | |
1937 | const unsigned long len, const unsigned long pgoff, | |
1938 | const unsigned long flags) | |
1939 | { | |
1940 | struct vm_area_struct *vma; | |
1941 | struct mm_struct *mm = current->mm; | |
db4fbfb9 ML |
1942 | unsigned long addr = addr0; |
1943 | struct vm_unmapped_area_info info; | |
1da177e4 LT |
1944 | |
1945 | /* requested length too big for entire address space */ | |
2afc745f | 1946 | if (len > TASK_SIZE - mmap_min_addr) |
1da177e4 LT |
1947 | return -ENOMEM; |
1948 | ||
06abdfb4 BH |
1949 | if (flags & MAP_FIXED) |
1950 | return addr; | |
1951 | ||
1da177e4 LT |
1952 | /* requesting a specific address */ |
1953 | if (addr) { | |
1954 | addr = PAGE_ALIGN(addr); | |
1955 | vma = find_vma(mm, addr); | |
2afc745f | 1956 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
1da177e4 LT |
1957 | (!vma || addr + len <= vma->vm_start)) |
1958 | return addr; | |
1959 | } | |
1960 | ||
db4fbfb9 ML |
1961 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
1962 | info.length = len; | |
2afc745f | 1963 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); |
db4fbfb9 ML |
1964 | info.high_limit = mm->mmap_base; |
1965 | info.align_mask = 0; | |
1966 | addr = vm_unmapped_area(&info); | |
b716ad95 | 1967 | |
1da177e4 LT |
1968 | /* |
1969 | * A failed mmap() very likely causes application failure, | |
1970 | * so fall back to the bottom-up function here. This scenario | |
1971 | * can happen with large stack limits and large mmap() | |
1972 | * allocations. | |
1973 | */ | |
db4fbfb9 ML |
1974 | if (addr & ~PAGE_MASK) { |
1975 | VM_BUG_ON(addr != -ENOMEM); | |
1976 | info.flags = 0; | |
1977 | info.low_limit = TASK_UNMAPPED_BASE; | |
1978 | info.high_limit = TASK_SIZE; | |
1979 | addr = vm_unmapped_area(&info); | |
1980 | } | |
1da177e4 LT |
1981 | |
1982 | return addr; | |
1983 | } | |
1984 | #endif | |
1985 | ||
1da177e4 LT |
1986 | unsigned long |
1987 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, | |
1988 | unsigned long pgoff, unsigned long flags) | |
1989 | { | |
06abdfb4 BH |
1990 | unsigned long (*get_area)(struct file *, unsigned long, |
1991 | unsigned long, unsigned long, unsigned long); | |
1992 | ||
9206de95 AV |
1993 | unsigned long error = arch_mmap_check(addr, len, flags); |
1994 | if (error) | |
1995 | return error; | |
1996 | ||
1997 | /* Careful about overflows.. */ | |
1998 | if (len > TASK_SIZE) | |
1999 | return -ENOMEM; | |
2000 | ||
06abdfb4 | 2001 | get_area = current->mm->get_unmapped_area; |
72c2d531 | 2002 | if (file && file->f_op->get_unmapped_area) |
06abdfb4 BH |
2003 | get_area = file->f_op->get_unmapped_area; |
2004 | addr = get_area(file, addr, len, pgoff, flags); | |
2005 | if (IS_ERR_VALUE(addr)) | |
2006 | return addr; | |
1da177e4 | 2007 | |
07ab67c8 LT |
2008 | if (addr > TASK_SIZE - len) |
2009 | return -ENOMEM; | |
2010 | if (addr & ~PAGE_MASK) | |
2011 | return -EINVAL; | |
06abdfb4 | 2012 | |
9ac4ed4b AV |
2013 | addr = arch_rebalance_pgtables(addr, len); |
2014 | error = security_mmap_addr(addr); | |
2015 | return error ? error : addr; | |
1da177e4 LT |
2016 | } |
2017 | ||
2018 | EXPORT_SYMBOL(get_unmapped_area); | |
2019 | ||
2020 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ | |
48aae425 | 2021 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2022 | { |
615d6e87 DB |
2023 | struct rb_node *rb_node; |
2024 | struct vm_area_struct *vma; | |
1da177e4 | 2025 | |
841e31e5 | 2026 | /* Check the cache first. */ |
615d6e87 DB |
2027 | vma = vmacache_find(mm, addr); |
2028 | if (likely(vma)) | |
2029 | return vma; | |
841e31e5 | 2030 | |
615d6e87 DB |
2031 | rb_node = mm->mm_rb.rb_node; |
2032 | vma = NULL; | |
841e31e5 | 2033 | |
615d6e87 DB |
2034 | while (rb_node) { |
2035 | struct vm_area_struct *tmp; | |
2036 | ||
2037 | tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
2038 | ||
2039 | if (tmp->vm_end > addr) { | |
2040 | vma = tmp; | |
2041 | if (tmp->vm_start <= addr) | |
2042 | break; | |
2043 | rb_node = rb_node->rb_left; | |
2044 | } else | |
2045 | rb_node = rb_node->rb_right; | |
1da177e4 | 2046 | } |
615d6e87 DB |
2047 | |
2048 | if (vma) | |
2049 | vmacache_update(addr, vma); | |
1da177e4 LT |
2050 | return vma; |
2051 | } | |
2052 | ||
2053 | EXPORT_SYMBOL(find_vma); | |
2054 | ||
6bd4837d KM |
2055 | /* |
2056 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. | |
6bd4837d | 2057 | */ |
1da177e4 LT |
2058 | struct vm_area_struct * |
2059 | find_vma_prev(struct mm_struct *mm, unsigned long addr, | |
2060 | struct vm_area_struct **pprev) | |
2061 | { | |
6bd4837d | 2062 | struct vm_area_struct *vma; |
1da177e4 | 2063 | |
6bd4837d | 2064 | vma = find_vma(mm, addr); |
83cd904d MP |
2065 | if (vma) { |
2066 | *pprev = vma->vm_prev; | |
2067 | } else { | |
2068 | struct rb_node *rb_node = mm->mm_rb.rb_node; | |
2069 | *pprev = NULL; | |
2070 | while (rb_node) { | |
2071 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); | |
2072 | rb_node = rb_node->rb_right; | |
2073 | } | |
2074 | } | |
6bd4837d | 2075 | return vma; |
1da177e4 LT |
2076 | } |
2077 | ||
2078 | /* | |
2079 | * Verify that the stack growth is acceptable and | |
2080 | * update accounting. This is shared with both the | |
2081 | * grow-up and grow-down cases. | |
2082 | */ | |
48aae425 | 2083 | static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) |
1da177e4 LT |
2084 | { |
2085 | struct mm_struct *mm = vma->vm_mm; | |
2086 | struct rlimit *rlim = current->signal->rlim; | |
0d59a01b | 2087 | unsigned long new_start; |
1da177e4 LT |
2088 | |
2089 | /* address space limit tests */ | |
119f657c | 2090 | if (!may_expand_vm(mm, grow)) |
1da177e4 LT |
2091 | return -ENOMEM; |
2092 | ||
2093 | /* Stack limit test */ | |
59e99e5b | 2094 | if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) |
1da177e4 LT |
2095 | return -ENOMEM; |
2096 | ||
2097 | /* mlock limit tests */ | |
2098 | if (vma->vm_flags & VM_LOCKED) { | |
2099 | unsigned long locked; | |
2100 | unsigned long limit; | |
2101 | locked = mm->locked_vm + grow; | |
59e99e5b JS |
2102 | limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); |
2103 | limit >>= PAGE_SHIFT; | |
1da177e4 LT |
2104 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
2105 | return -ENOMEM; | |
2106 | } | |
2107 | ||
0d59a01b AL |
2108 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
2109 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : | |
2110 | vma->vm_end - size; | |
2111 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) | |
2112 | return -EFAULT; | |
2113 | ||
1da177e4 LT |
2114 | /* |
2115 | * Overcommit.. This must be the final test, as it will | |
2116 | * update security statistics. | |
2117 | */ | |
05fa199d | 2118 | if (security_vm_enough_memory_mm(mm, grow)) |
1da177e4 LT |
2119 | return -ENOMEM; |
2120 | ||
2121 | /* Ok, everything looks good - let it rip */ | |
1da177e4 LT |
2122 | if (vma->vm_flags & VM_LOCKED) |
2123 | mm->locked_vm += grow; | |
ab50b8ed | 2124 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); |
1da177e4 LT |
2125 | return 0; |
2126 | } | |
2127 | ||
46dea3d0 | 2128 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
1da177e4 | 2129 | /* |
46dea3d0 HD |
2130 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
2131 | * vma is the last one with address > vma->vm_end. Have to extend vma. | |
1da177e4 | 2132 | */ |
46dea3d0 | 2133 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
1da177e4 LT |
2134 | { |
2135 | int error; | |
2136 | ||
2137 | if (!(vma->vm_flags & VM_GROWSUP)) | |
2138 | return -EFAULT; | |
2139 | ||
2140 | /* | |
2141 | * We must make sure the anon_vma is allocated | |
2142 | * so that the anon_vma locking is not a noop. | |
2143 | */ | |
2144 | if (unlikely(anon_vma_prepare(vma))) | |
2145 | return -ENOMEM; | |
bb4a340e | 2146 | vma_lock_anon_vma(vma); |
1da177e4 LT |
2147 | |
2148 | /* | |
2149 | * vma->vm_start/vm_end cannot change under us because the caller | |
2150 | * is required to hold the mmap_sem in read mode. We need the | |
2151 | * anon_vma lock to serialize against concurrent expand_stacks. | |
06b32f3a | 2152 | * Also guard against wrapping around to address 0. |
1da177e4 | 2153 | */ |
06b32f3a HD |
2154 | if (address < PAGE_ALIGN(address+4)) |
2155 | address = PAGE_ALIGN(address+4); | |
2156 | else { | |
bb4a340e | 2157 | vma_unlock_anon_vma(vma); |
06b32f3a HD |
2158 | return -ENOMEM; |
2159 | } | |
1da177e4 LT |
2160 | error = 0; |
2161 | ||
2162 | /* Somebody else might have raced and expanded it already */ | |
2163 | if (address > vma->vm_end) { | |
2164 | unsigned long size, grow; | |
2165 | ||
2166 | size = address - vma->vm_start; | |
2167 | grow = (address - vma->vm_end) >> PAGE_SHIFT; | |
2168 | ||
42c36f63 HD |
2169 | error = -ENOMEM; |
2170 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { | |
2171 | error = acct_stack_growth(vma, size, grow); | |
2172 | if (!error) { | |
4128997b ML |
2173 | /* |
2174 | * vma_gap_update() doesn't support concurrent | |
2175 | * updates, but we only hold a shared mmap_sem | |
2176 | * lock here, so we need to protect against | |
2177 | * concurrent vma expansions. | |
2178 | * vma_lock_anon_vma() doesn't help here, as | |
2179 | * we don't guarantee that all growable vmas | |
2180 | * in a mm share the same root anon vma. | |
2181 | * So, we reuse mm->page_table_lock to guard | |
2182 | * against concurrent vma expansions. | |
2183 | */ | |
2184 | spin_lock(&vma->vm_mm->page_table_lock); | |
bf181b9f | 2185 | anon_vma_interval_tree_pre_update_vma(vma); |
42c36f63 | 2186 | vma->vm_end = address; |
bf181b9f | 2187 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 ML |
2188 | if (vma->vm_next) |
2189 | vma_gap_update(vma->vm_next); | |
2190 | else | |
2191 | vma->vm_mm->highest_vm_end = address; | |
4128997b ML |
2192 | spin_unlock(&vma->vm_mm->page_table_lock); |
2193 | ||
42c36f63 HD |
2194 | perf_event_mmap(vma); |
2195 | } | |
3af9e859 | 2196 | } |
1da177e4 | 2197 | } |
bb4a340e | 2198 | vma_unlock_anon_vma(vma); |
b15d00b6 | 2199 | khugepaged_enter_vma_merge(vma); |
ed8ea815 | 2200 | validate_mm(vma->vm_mm); |
1da177e4 LT |
2201 | return error; |
2202 | } | |
46dea3d0 HD |
2203 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
2204 | ||
1da177e4 LT |
2205 | /* |
2206 | * vma is the first one with address < vma->vm_start. Have to extend vma. | |
2207 | */ | |
d05f3169 | 2208 | int expand_downwards(struct vm_area_struct *vma, |
b6a2fea3 | 2209 | unsigned long address) |
1da177e4 LT |
2210 | { |
2211 | int error; | |
2212 | ||
2213 | /* | |
2214 | * We must make sure the anon_vma is allocated | |
2215 | * so that the anon_vma locking is not a noop. | |
2216 | */ | |
2217 | if (unlikely(anon_vma_prepare(vma))) | |
2218 | return -ENOMEM; | |
8869477a EP |
2219 | |
2220 | address &= PAGE_MASK; | |
e5467859 | 2221 | error = security_mmap_addr(address); |
8869477a EP |
2222 | if (error) |
2223 | return error; | |
2224 | ||
bb4a340e | 2225 | vma_lock_anon_vma(vma); |
1da177e4 LT |
2226 | |
2227 | /* | |
2228 | * vma->vm_start/vm_end cannot change under us because the caller | |
2229 | * is required to hold the mmap_sem in read mode. We need the | |
2230 | * anon_vma lock to serialize against concurrent expand_stacks. | |
2231 | */ | |
1da177e4 LT |
2232 | |
2233 | /* Somebody else might have raced and expanded it already */ | |
2234 | if (address < vma->vm_start) { | |
2235 | unsigned long size, grow; | |
2236 | ||
2237 | size = vma->vm_end - address; | |
2238 | grow = (vma->vm_start - address) >> PAGE_SHIFT; | |
2239 | ||
a626ca6a LT |
2240 | error = -ENOMEM; |
2241 | if (grow <= vma->vm_pgoff) { | |
2242 | error = acct_stack_growth(vma, size, grow); | |
2243 | if (!error) { | |
4128997b ML |
2244 | /* |
2245 | * vma_gap_update() doesn't support concurrent | |
2246 | * updates, but we only hold a shared mmap_sem | |
2247 | * lock here, so we need to protect against | |
2248 | * concurrent vma expansions. | |
2249 | * vma_lock_anon_vma() doesn't help here, as | |
2250 | * we don't guarantee that all growable vmas | |
2251 | * in a mm share the same root anon vma. | |
2252 | * So, we reuse mm->page_table_lock to guard | |
2253 | * against concurrent vma expansions. | |
2254 | */ | |
2255 | spin_lock(&vma->vm_mm->page_table_lock); | |
bf181b9f | 2256 | anon_vma_interval_tree_pre_update_vma(vma); |
a626ca6a LT |
2257 | vma->vm_start = address; |
2258 | vma->vm_pgoff -= grow; | |
bf181b9f | 2259 | anon_vma_interval_tree_post_update_vma(vma); |
d3737187 | 2260 | vma_gap_update(vma); |
4128997b ML |
2261 | spin_unlock(&vma->vm_mm->page_table_lock); |
2262 | ||
a626ca6a LT |
2263 | perf_event_mmap(vma); |
2264 | } | |
1da177e4 LT |
2265 | } |
2266 | } | |
bb4a340e | 2267 | vma_unlock_anon_vma(vma); |
b15d00b6 | 2268 | khugepaged_enter_vma_merge(vma); |
ed8ea815 | 2269 | validate_mm(vma->vm_mm); |
1da177e4 LT |
2270 | return error; |
2271 | } | |
2272 | ||
09884964 LT |
2273 | /* |
2274 | * Note how expand_stack() refuses to expand the stack all the way to | |
2275 | * abut the next virtual mapping, *unless* that mapping itself is also | |
2276 | * a stack mapping. We want to leave room for a guard page, after all | |
2277 | * (the guard page itself is not added here, that is done by the | |
2278 | * actual page faulting logic) | |
2279 | * | |
2280 | * This matches the behavior of the guard page logic (see mm/memory.c: | |
2281 | * check_stack_guard_page()), which only allows the guard page to be | |
2282 | * removed under these circumstances. | |
2283 | */ | |
b6a2fea3 OW |
2284 | #ifdef CONFIG_STACK_GROWSUP |
2285 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2286 | { | |
09884964 LT |
2287 | struct vm_area_struct *next; |
2288 | ||
2289 | address &= PAGE_MASK; | |
2290 | next = vma->vm_next; | |
2291 | if (next && next->vm_start == address + PAGE_SIZE) { | |
2292 | if (!(next->vm_flags & VM_GROWSUP)) | |
2293 | return -ENOMEM; | |
2294 | } | |
b6a2fea3 OW |
2295 | return expand_upwards(vma, address); |
2296 | } | |
2297 | ||
2298 | struct vm_area_struct * | |
2299 | find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
2300 | { | |
2301 | struct vm_area_struct *vma, *prev; | |
2302 | ||
2303 | addr &= PAGE_MASK; | |
2304 | vma = find_vma_prev(mm, addr, &prev); | |
2305 | if (vma && (vma->vm_start <= addr)) | |
2306 | return vma; | |
1c127185 | 2307 | if (!prev || expand_stack(prev, addr)) |
b6a2fea3 | 2308 | return NULL; |
cea10a19 ML |
2309 | if (prev->vm_flags & VM_LOCKED) |
2310 | __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL); | |
b6a2fea3 OW |
2311 | return prev; |
2312 | } | |
2313 | #else | |
2314 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
2315 | { | |
09884964 LT |
2316 | struct vm_area_struct *prev; |
2317 | ||
2318 | address &= PAGE_MASK; | |
2319 | prev = vma->vm_prev; | |
2320 | if (prev && prev->vm_end == address) { | |
2321 | if (!(prev->vm_flags & VM_GROWSDOWN)) | |
2322 | return -ENOMEM; | |
2323 | } | |
b6a2fea3 OW |
2324 | return expand_downwards(vma, address); |
2325 | } | |
2326 | ||
1da177e4 | 2327 | struct vm_area_struct * |
cc71aba3 | 2328 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 2329 | { |
cc71aba3 | 2330 | struct vm_area_struct *vma; |
1da177e4 LT |
2331 | unsigned long start; |
2332 | ||
2333 | addr &= PAGE_MASK; | |
cc71aba3 | 2334 | vma = find_vma(mm, addr); |
1da177e4 LT |
2335 | if (!vma) |
2336 | return NULL; | |
2337 | if (vma->vm_start <= addr) | |
2338 | return vma; | |
2339 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
2340 | return NULL; | |
2341 | start = vma->vm_start; | |
2342 | if (expand_stack(vma, addr)) | |
2343 | return NULL; | |
cea10a19 ML |
2344 | if (vma->vm_flags & VM_LOCKED) |
2345 | __mlock_vma_pages_range(vma, addr, start, NULL); | |
1da177e4 LT |
2346 | return vma; |
2347 | } | |
2348 | #endif | |
2349 | ||
1da177e4 | 2350 | /* |
2c0b3814 | 2351 | * Ok - we have the memory areas we should free on the vma list, |
1da177e4 | 2352 | * so release them, and do the vma updates. |
2c0b3814 HD |
2353 | * |
2354 | * Called with the mm semaphore held. | |
1da177e4 | 2355 | */ |
2c0b3814 | 2356 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2357 | { |
4f74d2c8 LT |
2358 | unsigned long nr_accounted = 0; |
2359 | ||
365e9c87 HD |
2360 | /* Update high watermark before we lower total_vm */ |
2361 | update_hiwater_vm(mm); | |
1da177e4 | 2362 | do { |
2c0b3814 HD |
2363 | long nrpages = vma_pages(vma); |
2364 | ||
4f74d2c8 LT |
2365 | if (vma->vm_flags & VM_ACCOUNT) |
2366 | nr_accounted += nrpages; | |
2c0b3814 | 2367 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); |
a8fb5618 | 2368 | vma = remove_vma(vma); |
146425a3 | 2369 | } while (vma); |
4f74d2c8 | 2370 | vm_unacct_memory(nr_accounted); |
1da177e4 LT |
2371 | validate_mm(mm); |
2372 | } | |
2373 | ||
2374 | /* | |
2375 | * Get rid of page table information in the indicated region. | |
2376 | * | |
f10df686 | 2377 | * Called with the mm semaphore held. |
1da177e4 LT |
2378 | */ |
2379 | static void unmap_region(struct mm_struct *mm, | |
e0da382c HD |
2380 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
2381 | unsigned long start, unsigned long end) | |
1da177e4 | 2382 | { |
cc71aba3 | 2383 | struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; |
d16dfc55 | 2384 | struct mmu_gather tlb; |
1da177e4 LT |
2385 | |
2386 | lru_add_drain(); | |
2b047252 | 2387 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 2388 | update_hiwater_rss(mm); |
4f74d2c8 | 2389 | unmap_vmas(&tlb, vma, start, end); |
d16dfc55 | 2390 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
6ee8630e | 2391 | next ? next->vm_start : USER_PGTABLES_CEILING); |
d16dfc55 | 2392 | tlb_finish_mmu(&tlb, start, end); |
1da177e4 LT |
2393 | } |
2394 | ||
2395 | /* | |
2396 | * Create a list of vma's touched by the unmap, removing them from the mm's | |
2397 | * vma list as we go.. | |
2398 | */ | |
2399 | static void | |
2400 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, | |
2401 | struct vm_area_struct *prev, unsigned long end) | |
2402 | { | |
2403 | struct vm_area_struct **insertion_point; | |
2404 | struct vm_area_struct *tail_vma = NULL; | |
2405 | ||
2406 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); | |
297c5eee | 2407 | vma->vm_prev = NULL; |
1da177e4 | 2408 | do { |
d3737187 | 2409 | vma_rb_erase(vma, &mm->mm_rb); |
1da177e4 LT |
2410 | mm->map_count--; |
2411 | tail_vma = vma; | |
2412 | vma = vma->vm_next; | |
2413 | } while (vma && vma->vm_start < end); | |
2414 | *insertion_point = vma; | |
d3737187 | 2415 | if (vma) { |
297c5eee | 2416 | vma->vm_prev = prev; |
d3737187 ML |
2417 | vma_gap_update(vma); |
2418 | } else | |
2419 | mm->highest_vm_end = prev ? prev->vm_end : 0; | |
1da177e4 | 2420 | tail_vma->vm_next = NULL; |
615d6e87 DB |
2421 | |
2422 | /* Kill the cache */ | |
2423 | vmacache_invalidate(mm); | |
1da177e4 LT |
2424 | } |
2425 | ||
2426 | /* | |
659ace58 KM |
2427 | * __split_vma() bypasses sysctl_max_map_count checking. We use this on the |
2428 | * munmap path where it doesn't make sense to fail. | |
1da177e4 | 2429 | */ |
cc71aba3 | 2430 | static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
2431 | unsigned long addr, int new_below) |
2432 | { | |
1da177e4 | 2433 | struct vm_area_struct *new; |
5beb4930 | 2434 | int err = -ENOMEM; |
1da177e4 | 2435 | |
a5516438 AK |
2436 | if (is_vm_hugetlb_page(vma) && (addr & |
2437 | ~(huge_page_mask(hstate_vma(vma))))) | |
1da177e4 LT |
2438 | return -EINVAL; |
2439 | ||
e94b1766 | 2440 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 | 2441 | if (!new) |
5beb4930 | 2442 | goto out_err; |
1da177e4 LT |
2443 | |
2444 | /* most fields are the same, copy all, and then fixup */ | |
2445 | *new = *vma; | |
2446 | ||
5beb4930 RR |
2447 | INIT_LIST_HEAD(&new->anon_vma_chain); |
2448 | ||
1da177e4 LT |
2449 | if (new_below) |
2450 | new->vm_end = addr; | |
2451 | else { | |
2452 | new->vm_start = addr; | |
2453 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); | |
2454 | } | |
2455 | ||
ef0855d3 ON |
2456 | err = vma_dup_policy(vma, new); |
2457 | if (err) | |
5beb4930 | 2458 | goto out_free_vma; |
1da177e4 | 2459 | |
5beb4930 RR |
2460 | if (anon_vma_clone(new, vma)) |
2461 | goto out_free_mpol; | |
2462 | ||
e9714acf | 2463 | if (new->vm_file) |
1da177e4 LT |
2464 | get_file(new->vm_file); |
2465 | ||
2466 | if (new->vm_ops && new->vm_ops->open) | |
2467 | new->vm_ops->open(new); | |
2468 | ||
2469 | if (new_below) | |
5beb4930 | 2470 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
1da177e4 LT |
2471 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
2472 | else | |
5beb4930 | 2473 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
1da177e4 | 2474 | |
5beb4930 RR |
2475 | /* Success. */ |
2476 | if (!err) | |
2477 | return 0; | |
2478 | ||
2479 | /* Clean everything up if vma_adjust failed. */ | |
58927533 RR |
2480 | if (new->vm_ops && new->vm_ops->close) |
2481 | new->vm_ops->close(new); | |
e9714acf | 2482 | if (new->vm_file) |
5beb4930 | 2483 | fput(new->vm_file); |
2aeadc30 | 2484 | unlink_anon_vmas(new); |
5beb4930 | 2485 | out_free_mpol: |
ef0855d3 | 2486 | mpol_put(vma_policy(new)); |
5beb4930 RR |
2487 | out_free_vma: |
2488 | kmem_cache_free(vm_area_cachep, new); | |
2489 | out_err: | |
2490 | return err; | |
1da177e4 LT |
2491 | } |
2492 | ||
659ace58 KM |
2493 | /* |
2494 | * Split a vma into two pieces at address 'addr', a new vma is allocated | |
2495 | * either for the first part or the tail. | |
2496 | */ | |
2497 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, | |
2498 | unsigned long addr, int new_below) | |
2499 | { | |
2500 | if (mm->map_count >= sysctl_max_map_count) | |
2501 | return -ENOMEM; | |
2502 | ||
2503 | return __split_vma(mm, vma, addr, new_below); | |
2504 | } | |
2505 | ||
1da177e4 LT |
2506 | /* Munmap is split into 2 main parts -- this part which finds |
2507 | * what needs doing, and the areas themselves, which do the | |
2508 | * work. This now handles partial unmappings. | |
2509 | * Jeremy Fitzhardinge <jeremy@goop.org> | |
2510 | */ | |
2511 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
2512 | { | |
2513 | unsigned long end; | |
146425a3 | 2514 | struct vm_area_struct *vma, *prev, *last; |
1da177e4 LT |
2515 | |
2516 | if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) | |
2517 | return -EINVAL; | |
2518 | ||
cc71aba3 | 2519 | len = PAGE_ALIGN(len); |
2520 | if (len == 0) | |
1da177e4 LT |
2521 | return -EINVAL; |
2522 | ||
2523 | /* Find the first overlapping VMA */ | |
9be34c9d | 2524 | vma = find_vma(mm, start); |
146425a3 | 2525 | if (!vma) |
1da177e4 | 2526 | return 0; |
9be34c9d | 2527 | prev = vma->vm_prev; |
146425a3 | 2528 | /* we have start < vma->vm_end */ |
1da177e4 LT |
2529 | |
2530 | /* if it doesn't overlap, we have nothing.. */ | |
2531 | end = start + len; | |
146425a3 | 2532 | if (vma->vm_start >= end) |
1da177e4 LT |
2533 | return 0; |
2534 | ||
2535 | /* | |
2536 | * If we need to split any vma, do it now to save pain later. | |
2537 | * | |
2538 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially | |
2539 | * unmapped vm_area_struct will remain in use: so lower split_vma | |
2540 | * places tmp vma above, and higher split_vma places tmp vma below. | |
2541 | */ | |
146425a3 | 2542 | if (start > vma->vm_start) { |
659ace58 KM |
2543 | int error; |
2544 | ||
2545 | /* | |
2546 | * Make sure that map_count on return from munmap() will | |
2547 | * not exceed its limit; but let map_count go just above | |
2548 | * its limit temporarily, to help free resources as expected. | |
2549 | */ | |
2550 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) | |
2551 | return -ENOMEM; | |
2552 | ||
2553 | error = __split_vma(mm, vma, start, 0); | |
1da177e4 LT |
2554 | if (error) |
2555 | return error; | |
146425a3 | 2556 | prev = vma; |
1da177e4 LT |
2557 | } |
2558 | ||
2559 | /* Does it split the last one? */ | |
2560 | last = find_vma(mm, end); | |
2561 | if (last && end > last->vm_start) { | |
659ace58 | 2562 | int error = __split_vma(mm, last, end, 1); |
1da177e4 LT |
2563 | if (error) |
2564 | return error; | |
2565 | } | |
cc71aba3 | 2566 | vma = prev ? prev->vm_next : mm->mmap; |
1da177e4 | 2567 | |
ba470de4 RR |
2568 | /* |
2569 | * unlock any mlock()ed ranges before detaching vmas | |
2570 | */ | |
2571 | if (mm->locked_vm) { | |
2572 | struct vm_area_struct *tmp = vma; | |
2573 | while (tmp && tmp->vm_start < end) { | |
2574 | if (tmp->vm_flags & VM_LOCKED) { | |
2575 | mm->locked_vm -= vma_pages(tmp); | |
2576 | munlock_vma_pages_all(tmp); | |
2577 | } | |
2578 | tmp = tmp->vm_next; | |
2579 | } | |
2580 | } | |
2581 | ||
1da177e4 LT |
2582 | /* |
2583 | * Remove the vma's, and unmap the actual pages | |
2584 | */ | |
146425a3 HD |
2585 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
2586 | unmap_region(mm, vma, prev, start, end); | |
1da177e4 LT |
2587 | |
2588 | /* Fix up all other VM information */ | |
2c0b3814 | 2589 | remove_vma_list(mm, vma); |
1da177e4 LT |
2590 | |
2591 | return 0; | |
2592 | } | |
1da177e4 | 2593 | |
bfce281c | 2594 | int vm_munmap(unsigned long start, size_t len) |
1da177e4 LT |
2595 | { |
2596 | int ret; | |
bfce281c | 2597 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
2598 | |
2599 | down_write(&mm->mmap_sem); | |
a46ef99d | 2600 | ret = do_munmap(mm, start, len); |
1da177e4 LT |
2601 | up_write(&mm->mmap_sem); |
2602 | return ret; | |
2603 | } | |
a46ef99d LT |
2604 | EXPORT_SYMBOL(vm_munmap); |
2605 | ||
2606 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
2607 | { | |
2608 | profile_munmap(addr); | |
bfce281c | 2609 | return vm_munmap(addr, len); |
a46ef99d | 2610 | } |
1da177e4 LT |
2611 | |
2612 | static inline void verify_mm_writelocked(struct mm_struct *mm) | |
2613 | { | |
a241ec65 | 2614 | #ifdef CONFIG_DEBUG_VM |
1da177e4 LT |
2615 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
2616 | WARN_ON(1); | |
2617 | up_read(&mm->mmap_sem); | |
2618 | } | |
2619 | #endif | |
2620 | } | |
2621 | ||
2622 | /* | |
2623 | * this is really a simplified "do_mmap". it only handles | |
2624 | * anonymous maps. eventually we may be able to do some | |
2625 | * brk-specific accounting here. | |
2626 | */ | |
e4eb1ff6 | 2627 | static unsigned long do_brk(unsigned long addr, unsigned long len) |
1da177e4 | 2628 | { |
cc71aba3 | 2629 | struct mm_struct *mm = current->mm; |
2630 | struct vm_area_struct *vma, *prev; | |
1da177e4 | 2631 | unsigned long flags; |
cc71aba3 | 2632 | struct rb_node **rb_link, *rb_parent; |
1da177e4 | 2633 | pgoff_t pgoff = addr >> PAGE_SHIFT; |
3a459756 | 2634 | int error; |
1da177e4 LT |
2635 | |
2636 | len = PAGE_ALIGN(len); | |
2637 | if (!len) | |
2638 | return addr; | |
2639 | ||
3a459756 KK |
2640 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; |
2641 | ||
2c6a1016 AV |
2642 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
2643 | if (error & ~PAGE_MASK) | |
3a459756 KK |
2644 | return error; |
2645 | ||
363ee17f DB |
2646 | error = mlock_future_check(mm, mm->def_flags, len); |
2647 | if (error) | |
2648 | return error; | |
1da177e4 LT |
2649 | |
2650 | /* | |
2651 | * mm->mmap_sem is required to protect against another thread | |
2652 | * changing the mappings in case we sleep. | |
2653 | */ | |
2654 | verify_mm_writelocked(mm); | |
2655 | ||
2656 | /* | |
2657 | * Clear old maps. this also does some error checking for us | |
2658 | */ | |
2659 | munmap_back: | |
6597d783 | 2660 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { |
1da177e4 LT |
2661 | if (do_munmap(mm, addr, len)) |
2662 | return -ENOMEM; | |
2663 | goto munmap_back; | |
2664 | } | |
2665 | ||
2666 | /* Check against address space limits *after* clearing old maps... */ | |
119f657c | 2667 | if (!may_expand_vm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2668 | return -ENOMEM; |
2669 | ||
2670 | if (mm->map_count > sysctl_max_map_count) | |
2671 | return -ENOMEM; | |
2672 | ||
191c5424 | 2673 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
1da177e4 LT |
2674 | return -ENOMEM; |
2675 | ||
1da177e4 | 2676 | /* Can we just expand an old private anonymous mapping? */ |
ba470de4 RR |
2677 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
2678 | NULL, NULL, pgoff, NULL); | |
2679 | if (vma) | |
1da177e4 LT |
2680 | goto out; |
2681 | ||
2682 | /* | |
2683 | * create a vma struct for an anonymous mapping | |
2684 | */ | |
c5e3b83e | 2685 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2686 | if (!vma) { |
2687 | vm_unacct_memory(len >> PAGE_SHIFT); | |
2688 | return -ENOMEM; | |
2689 | } | |
1da177e4 | 2690 | |
5beb4930 | 2691 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
1da177e4 LT |
2692 | vma->vm_mm = mm; |
2693 | vma->vm_start = addr; | |
2694 | vma->vm_end = addr + len; | |
2695 | vma->vm_pgoff = pgoff; | |
2696 | vma->vm_flags = flags; | |
3ed75eb8 | 2697 | vma->vm_page_prot = vm_get_page_prot(flags); |
1da177e4 LT |
2698 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2699 | out: | |
3af9e859 | 2700 | perf_event_mmap(vma); |
1da177e4 | 2701 | mm->total_vm += len >> PAGE_SHIFT; |
128557ff ML |
2702 | if (flags & VM_LOCKED) |
2703 | mm->locked_vm += (len >> PAGE_SHIFT); | |
d9104d1c | 2704 | vma->vm_flags |= VM_SOFTDIRTY; |
1da177e4 LT |
2705 | return addr; |
2706 | } | |
2707 | ||
e4eb1ff6 LT |
2708 | unsigned long vm_brk(unsigned long addr, unsigned long len) |
2709 | { | |
2710 | struct mm_struct *mm = current->mm; | |
2711 | unsigned long ret; | |
128557ff | 2712 | bool populate; |
e4eb1ff6 LT |
2713 | |
2714 | down_write(&mm->mmap_sem); | |
2715 | ret = do_brk(addr, len); | |
128557ff | 2716 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
e4eb1ff6 | 2717 | up_write(&mm->mmap_sem); |
128557ff ML |
2718 | if (populate) |
2719 | mm_populate(addr, len); | |
e4eb1ff6 LT |
2720 | return ret; |
2721 | } | |
2722 | EXPORT_SYMBOL(vm_brk); | |
1da177e4 LT |
2723 | |
2724 | /* Release all mmaps. */ | |
2725 | void exit_mmap(struct mm_struct *mm) | |
2726 | { | |
d16dfc55 | 2727 | struct mmu_gather tlb; |
ba470de4 | 2728 | struct vm_area_struct *vma; |
1da177e4 LT |
2729 | unsigned long nr_accounted = 0; |
2730 | ||
d6dd61c8 | 2731 | /* mm's last user has gone, and its about to be pulled down */ |
cddb8a5c | 2732 | mmu_notifier_release(mm); |
d6dd61c8 | 2733 | |
ba470de4 RR |
2734 | if (mm->locked_vm) { |
2735 | vma = mm->mmap; | |
2736 | while (vma) { | |
2737 | if (vma->vm_flags & VM_LOCKED) | |
2738 | munlock_vma_pages_all(vma); | |
2739 | vma = vma->vm_next; | |
2740 | } | |
2741 | } | |
9480c53e JF |
2742 | |
2743 | arch_exit_mmap(mm); | |
2744 | ||
ba470de4 | 2745 | vma = mm->mmap; |
9480c53e JF |
2746 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
2747 | return; | |
2748 | ||
1da177e4 | 2749 | lru_add_drain(); |
1da177e4 | 2750 | flush_cache_mm(mm); |
2b047252 | 2751 | tlb_gather_mmu(&tlb, mm, 0, -1); |
901608d9 | 2752 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
e0da382c | 2753 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
4f74d2c8 | 2754 | unmap_vmas(&tlb, vma, 0, -1); |
9ba69294 | 2755 | |
6ee8630e | 2756 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); |
853f5e26 | 2757 | tlb_finish_mmu(&tlb, 0, -1); |
1da177e4 | 2758 | |
1da177e4 | 2759 | /* |
8f4f8c16 HD |
2760 | * Walk the list again, actually closing and freeing it, |
2761 | * with preemption enabled, without holding any MM locks. | |
1da177e4 | 2762 | */ |
4f74d2c8 LT |
2763 | while (vma) { |
2764 | if (vma->vm_flags & VM_ACCOUNT) | |
2765 | nr_accounted += vma_pages(vma); | |
a8fb5618 | 2766 | vma = remove_vma(vma); |
4f74d2c8 LT |
2767 | } |
2768 | vm_unacct_memory(nr_accounted); | |
e0da382c | 2769 | |
e1f56c89 KS |
2770 | WARN_ON(atomic_long_read(&mm->nr_ptes) > |
2771 | (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); | |
1da177e4 LT |
2772 | } |
2773 | ||
2774 | /* Insert vm structure into process list sorted by address | |
2775 | * and into the inode's i_mmap tree. If vm_file is non-NULL | |
3d48ae45 | 2776 | * then i_mmap_mutex is taken here. |
1da177e4 | 2777 | */ |
6597d783 | 2778 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 2779 | { |
6597d783 HD |
2780 | struct vm_area_struct *prev; |
2781 | struct rb_node **rb_link, *rb_parent; | |
1da177e4 LT |
2782 | |
2783 | /* | |
2784 | * The vm_pgoff of a purely anonymous vma should be irrelevant | |
2785 | * until its first write fault, when page's anon_vma and index | |
2786 | * are set. But now set the vm_pgoff it will almost certainly | |
2787 | * end up with (unless mremap moves it elsewhere before that | |
2788 | * first wfault), so /proc/pid/maps tells a consistent story. | |
2789 | * | |
2790 | * By setting it to reflect the virtual start address of the | |
2791 | * vma, merges and splits can happen in a seamless way, just | |
2792 | * using the existing file pgoff checks and manipulations. | |
2793 | * Similarly in do_mmap_pgoff and in do_brk. | |
2794 | */ | |
2795 | if (!vma->vm_file) { | |
2796 | BUG_ON(vma->anon_vma); | |
2797 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; | |
2798 | } | |
6597d783 HD |
2799 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
2800 | &prev, &rb_link, &rb_parent)) | |
1da177e4 | 2801 | return -ENOMEM; |
2fd4ef85 | 2802 | if ((vma->vm_flags & VM_ACCOUNT) && |
34b4e4aa | 2803 | security_vm_enough_memory_mm(mm, vma_pages(vma))) |
2fd4ef85 | 2804 | return -ENOMEM; |
2b144498 | 2805 | |
1da177e4 LT |
2806 | vma_link(mm, vma, prev, rb_link, rb_parent); |
2807 | return 0; | |
2808 | } | |
2809 | ||
2810 | /* | |
2811 | * Copy the vma structure to a new location in the same mm, | |
2812 | * prior to moving page table entries, to effect an mremap move. | |
2813 | */ | |
2814 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, | |
38a76013 ML |
2815 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
2816 | bool *need_rmap_locks) | |
1da177e4 LT |
2817 | { |
2818 | struct vm_area_struct *vma = *vmap; | |
2819 | unsigned long vma_start = vma->vm_start; | |
2820 | struct mm_struct *mm = vma->vm_mm; | |
2821 | struct vm_area_struct *new_vma, *prev; | |
2822 | struct rb_node **rb_link, *rb_parent; | |
948f017b | 2823 | bool faulted_in_anon_vma = true; |
1da177e4 LT |
2824 | |
2825 | /* | |
2826 | * If anonymous vma has not yet been faulted, update new pgoff | |
2827 | * to match new location, to increase its chance of merging. | |
2828 | */ | |
948f017b | 2829 | if (unlikely(!vma->vm_file && !vma->anon_vma)) { |
1da177e4 | 2830 | pgoff = addr >> PAGE_SHIFT; |
948f017b AA |
2831 | faulted_in_anon_vma = false; |
2832 | } | |
1da177e4 | 2833 | |
6597d783 HD |
2834 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
2835 | return NULL; /* should never get here */ | |
1da177e4 LT |
2836 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
2837 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); | |
2838 | if (new_vma) { | |
2839 | /* | |
2840 | * Source vma may have been merged into new_vma | |
2841 | */ | |
948f017b AA |
2842 | if (unlikely(vma_start >= new_vma->vm_start && |
2843 | vma_start < new_vma->vm_end)) { | |
2844 | /* | |
2845 | * The only way we can get a vma_merge with | |
2846 | * self during an mremap is if the vma hasn't | |
2847 | * been faulted in yet and we were allowed to | |
2848 | * reset the dst vma->vm_pgoff to the | |
2849 | * destination address of the mremap to allow | |
2850 | * the merge to happen. mremap must change the | |
2851 | * vm_pgoff linearity between src and dst vmas | |
2852 | * (in turn preventing a vma_merge) to be | |
2853 | * safe. It is only safe to keep the vm_pgoff | |
2854 | * linear if there are no pages mapped yet. | |
2855 | */ | |
81d1b09c | 2856 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
38a76013 | 2857 | *vmap = vma = new_vma; |
108d6642 | 2858 | } |
38a76013 | 2859 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
1da177e4 | 2860 | } else { |
e94b1766 | 2861 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1da177e4 LT |
2862 | if (new_vma) { |
2863 | *new_vma = *vma; | |
523d4e20 ML |
2864 | new_vma->vm_start = addr; |
2865 | new_vma->vm_end = addr + len; | |
2866 | new_vma->vm_pgoff = pgoff; | |
ef0855d3 | 2867 | if (vma_dup_policy(vma, new_vma)) |
5beb4930 RR |
2868 | goto out_free_vma; |
2869 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); | |
2870 | if (anon_vma_clone(new_vma, vma)) | |
2871 | goto out_free_mempol; | |
e9714acf | 2872 | if (new_vma->vm_file) |
1da177e4 LT |
2873 | get_file(new_vma->vm_file); |
2874 | if (new_vma->vm_ops && new_vma->vm_ops->open) | |
2875 | new_vma->vm_ops->open(new_vma); | |
2876 | vma_link(mm, new_vma, prev, rb_link, rb_parent); | |
38a76013 | 2877 | *need_rmap_locks = false; |
1da177e4 LT |
2878 | } |
2879 | } | |
2880 | return new_vma; | |
5beb4930 RR |
2881 | |
2882 | out_free_mempol: | |
ef0855d3 | 2883 | mpol_put(vma_policy(new_vma)); |
5beb4930 RR |
2884 | out_free_vma: |
2885 | kmem_cache_free(vm_area_cachep, new_vma); | |
2886 | return NULL; | |
1da177e4 | 2887 | } |
119f657c | 2888 | |
2889 | /* | |
2890 | * Return true if the calling process may expand its vm space by the passed | |
2891 | * number of pages | |
2892 | */ | |
2893 | int may_expand_vm(struct mm_struct *mm, unsigned long npages) | |
2894 | { | |
2895 | unsigned long cur = mm->total_vm; /* pages */ | |
2896 | unsigned long lim; | |
2897 | ||
59e99e5b | 2898 | lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; |
119f657c | 2899 | |
2900 | if (cur + npages > lim) | |
2901 | return 0; | |
2902 | return 1; | |
2903 | } | |
fa5dc22f | 2904 | |
a62c34bd AL |
2905 | static int special_mapping_fault(struct vm_area_struct *vma, |
2906 | struct vm_fault *vmf); | |
2907 | ||
2908 | /* | |
2909 | * Having a close hook prevents vma merging regardless of flags. | |
2910 | */ | |
2911 | static void special_mapping_close(struct vm_area_struct *vma) | |
2912 | { | |
2913 | } | |
2914 | ||
2915 | static const char *special_mapping_name(struct vm_area_struct *vma) | |
2916 | { | |
2917 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; | |
2918 | } | |
2919 | ||
2920 | static const struct vm_operations_struct special_mapping_vmops = { | |
2921 | .close = special_mapping_close, | |
2922 | .fault = special_mapping_fault, | |
2923 | .name = special_mapping_name, | |
2924 | }; | |
2925 | ||
2926 | static const struct vm_operations_struct legacy_special_mapping_vmops = { | |
2927 | .close = special_mapping_close, | |
2928 | .fault = special_mapping_fault, | |
2929 | }; | |
fa5dc22f | 2930 | |
b1d0e4f5 NP |
2931 | static int special_mapping_fault(struct vm_area_struct *vma, |
2932 | struct vm_fault *vmf) | |
fa5dc22f | 2933 | { |
b1d0e4f5 | 2934 | pgoff_t pgoff; |
fa5dc22f RM |
2935 | struct page **pages; |
2936 | ||
b1d0e4f5 NP |
2937 | /* |
2938 | * special mappings have no vm_file, and in that case, the mm | |
2939 | * uses vm_pgoff internally. So we have to subtract it from here. | |
2940 | * We are allowed to do this because we are the mm; do not copy | |
2941 | * this code into drivers! | |
2942 | */ | |
2943 | pgoff = vmf->pgoff - vma->vm_pgoff; | |
fa5dc22f | 2944 | |
a62c34bd AL |
2945 | if (vma->vm_ops == &legacy_special_mapping_vmops) |
2946 | pages = vma->vm_private_data; | |
2947 | else | |
2948 | pages = ((struct vm_special_mapping *)vma->vm_private_data)-> | |
2949 | pages; | |
2950 | ||
2951 | for (; pgoff && *pages; ++pages) | |
b1d0e4f5 | 2952 | pgoff--; |
fa5dc22f RM |
2953 | |
2954 | if (*pages) { | |
2955 | struct page *page = *pages; | |
2956 | get_page(page); | |
b1d0e4f5 NP |
2957 | vmf->page = page; |
2958 | return 0; | |
fa5dc22f RM |
2959 | } |
2960 | ||
b1d0e4f5 | 2961 | return VM_FAULT_SIGBUS; |
fa5dc22f RM |
2962 | } |
2963 | ||
a62c34bd AL |
2964 | static struct vm_area_struct *__install_special_mapping( |
2965 | struct mm_struct *mm, | |
2966 | unsigned long addr, unsigned long len, | |
2967 | unsigned long vm_flags, const struct vm_operations_struct *ops, | |
2968 | void *priv) | |
fa5dc22f | 2969 | { |
462e635e | 2970 | int ret; |
fa5dc22f RM |
2971 | struct vm_area_struct *vma; |
2972 | ||
2973 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
2974 | if (unlikely(vma == NULL)) | |
3935ed6a | 2975 | return ERR_PTR(-ENOMEM); |
fa5dc22f | 2976 | |
5beb4930 | 2977 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
fa5dc22f RM |
2978 | vma->vm_mm = mm; |
2979 | vma->vm_start = addr; | |
2980 | vma->vm_end = addr + len; | |
2981 | ||
d9104d1c | 2982 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; |
3ed75eb8 | 2983 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
fa5dc22f | 2984 | |
a62c34bd AL |
2985 | vma->vm_ops = ops; |
2986 | vma->vm_private_data = priv; | |
fa5dc22f | 2987 | |
462e635e TO |
2988 | ret = insert_vm_struct(mm, vma); |
2989 | if (ret) | |
2990 | goto out; | |
fa5dc22f RM |
2991 | |
2992 | mm->total_vm += len >> PAGE_SHIFT; | |
2993 | ||
cdd6c482 | 2994 | perf_event_mmap(vma); |
089dd79d | 2995 | |
3935ed6a | 2996 | return vma; |
462e635e TO |
2997 | |
2998 | out: | |
2999 | kmem_cache_free(vm_area_cachep, vma); | |
3935ed6a SS |
3000 | return ERR_PTR(ret); |
3001 | } | |
3002 | ||
a62c34bd AL |
3003 | /* |
3004 | * Called with mm->mmap_sem held for writing. | |
3005 | * Insert a new vma covering the given region, with the given flags. | |
3006 | * Its pages are supplied by the given array of struct page *. | |
3007 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. | |
3008 | * The region past the last page supplied will always produce SIGBUS. | |
3009 | * The array pointer and the pages it points to are assumed to stay alive | |
3010 | * for as long as this mapping might exist. | |
3011 | */ | |
3012 | struct vm_area_struct *_install_special_mapping( | |
3013 | struct mm_struct *mm, | |
3014 | unsigned long addr, unsigned long len, | |
3015 | unsigned long vm_flags, const struct vm_special_mapping *spec) | |
3016 | { | |
3017 | return __install_special_mapping(mm, addr, len, vm_flags, | |
3018 | &special_mapping_vmops, (void *)spec); | |
3019 | } | |
3020 | ||
3935ed6a SS |
3021 | int install_special_mapping(struct mm_struct *mm, |
3022 | unsigned long addr, unsigned long len, | |
3023 | unsigned long vm_flags, struct page **pages) | |
3024 | { | |
a62c34bd AL |
3025 | struct vm_area_struct *vma = __install_special_mapping( |
3026 | mm, addr, len, vm_flags, &legacy_special_mapping_vmops, | |
3027 | (void *)pages); | |
3935ed6a | 3028 | |
14bd5b45 | 3029 | return PTR_ERR_OR_ZERO(vma); |
fa5dc22f | 3030 | } |
7906d00c AA |
3031 | |
3032 | static DEFINE_MUTEX(mm_all_locks_mutex); | |
3033 | ||
454ed842 | 3034 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
7906d00c | 3035 | { |
bf181b9f | 3036 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3037 | /* |
3038 | * The LSB of head.next can't change from under us | |
3039 | * because we hold the mm_all_locks_mutex. | |
3040 | */ | |
572043c9 | 3041 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); |
7906d00c AA |
3042 | /* |
3043 | * We can safely modify head.next after taking the | |
5a505085 | 3044 | * anon_vma->root->rwsem. If some other vma in this mm shares |
7906d00c AA |
3045 | * the same anon_vma we won't take it again. |
3046 | * | |
3047 | * No need of atomic instructions here, head.next | |
3048 | * can't change from under us thanks to the | |
5a505085 | 3049 | * anon_vma->root->rwsem. |
7906d00c AA |
3050 | */ |
3051 | if (__test_and_set_bit(0, (unsigned long *) | |
bf181b9f | 3052 | &anon_vma->root->rb_root.rb_node)) |
7906d00c AA |
3053 | BUG(); |
3054 | } | |
3055 | } | |
3056 | ||
454ed842 | 3057 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
7906d00c AA |
3058 | { |
3059 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3060 | /* | |
3061 | * AS_MM_ALL_LOCKS can't change from under us because | |
3062 | * we hold the mm_all_locks_mutex. | |
3063 | * | |
3064 | * Operations on ->flags have to be atomic because | |
3065 | * even if AS_MM_ALL_LOCKS is stable thanks to the | |
3066 | * mm_all_locks_mutex, there may be other cpus | |
3067 | * changing other bitflags in parallel to us. | |
3068 | */ | |
3069 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) | |
3070 | BUG(); | |
3d48ae45 | 3071 | mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem); |
7906d00c AA |
3072 | } |
3073 | } | |
3074 | ||
3075 | /* | |
3076 | * This operation locks against the VM for all pte/vma/mm related | |
3077 | * operations that could ever happen on a certain mm. This includes | |
3078 | * vmtruncate, try_to_unmap, and all page faults. | |
3079 | * | |
3080 | * The caller must take the mmap_sem in write mode before calling | |
3081 | * mm_take_all_locks(). The caller isn't allowed to release the | |
3082 | * mmap_sem until mm_drop_all_locks() returns. | |
3083 | * | |
3084 | * mmap_sem in write mode is required in order to block all operations | |
3085 | * that could modify pagetables and free pages without need of | |
3086 | * altering the vma layout (for example populate_range() with | |
3087 | * nonlinear vmas). It's also needed in write mode to avoid new | |
3088 | * anon_vmas to be associated with existing vmas. | |
3089 | * | |
3090 | * A single task can't take more than one mm_take_all_locks() in a row | |
3091 | * or it would deadlock. | |
3092 | * | |
bf181b9f | 3093 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
7906d00c AA |
3094 | * mapping->flags avoid to take the same lock twice, if more than one |
3095 | * vma in this mm is backed by the same anon_vma or address_space. | |
3096 | * | |
3097 | * We can take all the locks in random order because the VM code | |
631b0cfd | 3098 | * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never |
7906d00c AA |
3099 | * takes more than one of them in a row. Secondly we're protected |
3100 | * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. | |
3101 | * | |
3102 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations | |
3103 | * that may have to take thousand of locks. | |
3104 | * | |
3105 | * mm_take_all_locks() can fail if it's interrupted by signals. | |
3106 | */ | |
3107 | int mm_take_all_locks(struct mm_struct *mm) | |
3108 | { | |
3109 | struct vm_area_struct *vma; | |
5beb4930 | 3110 | struct anon_vma_chain *avc; |
7906d00c AA |
3111 | |
3112 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3113 | ||
3114 | mutex_lock(&mm_all_locks_mutex); | |
3115 | ||
3116 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3117 | if (signal_pending(current)) | |
3118 | goto out_unlock; | |
7906d00c | 3119 | if (vma->vm_file && vma->vm_file->f_mapping) |
454ed842 | 3120 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
7906d00c | 3121 | } |
7cd5a02f PZ |
3122 | |
3123 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3124 | if (signal_pending(current)) | |
3125 | goto out_unlock; | |
3126 | if (vma->anon_vma) | |
5beb4930 RR |
3127 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3128 | vm_lock_anon_vma(mm, avc->anon_vma); | |
7906d00c | 3129 | } |
7cd5a02f | 3130 | |
584cff54 | 3131 | return 0; |
7906d00c AA |
3132 | |
3133 | out_unlock: | |
584cff54 KC |
3134 | mm_drop_all_locks(mm); |
3135 | return -EINTR; | |
7906d00c AA |
3136 | } |
3137 | ||
3138 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |
3139 | { | |
bf181b9f | 3140 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
7906d00c AA |
3141 | /* |
3142 | * The LSB of head.next can't change to 0 from under | |
3143 | * us because we hold the mm_all_locks_mutex. | |
3144 | * | |
3145 | * We must however clear the bitflag before unlocking | |
bf181b9f | 3146 | * the vma so the users using the anon_vma->rb_root will |
7906d00c AA |
3147 | * never see our bitflag. |
3148 | * | |
3149 | * No need of atomic instructions here, head.next | |
3150 | * can't change from under us until we release the | |
5a505085 | 3151 | * anon_vma->root->rwsem. |
7906d00c AA |
3152 | */ |
3153 | if (!__test_and_clear_bit(0, (unsigned long *) | |
bf181b9f | 3154 | &anon_vma->root->rb_root.rb_node)) |
7906d00c | 3155 | BUG(); |
08b52706 | 3156 | anon_vma_unlock_write(anon_vma); |
7906d00c AA |
3157 | } |
3158 | } | |
3159 | ||
3160 | static void vm_unlock_mapping(struct address_space *mapping) | |
3161 | { | |
3162 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { | |
3163 | /* | |
3164 | * AS_MM_ALL_LOCKS can't change to 0 from under us | |
3165 | * because we hold the mm_all_locks_mutex. | |
3166 | */ | |
3d48ae45 | 3167 | mutex_unlock(&mapping->i_mmap_mutex); |
7906d00c AA |
3168 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
3169 | &mapping->flags)) | |
3170 | BUG(); | |
3171 | } | |
3172 | } | |
3173 | ||
3174 | /* | |
3175 | * The mmap_sem cannot be released by the caller until | |
3176 | * mm_drop_all_locks() returns. | |
3177 | */ | |
3178 | void mm_drop_all_locks(struct mm_struct *mm) | |
3179 | { | |
3180 | struct vm_area_struct *vma; | |
5beb4930 | 3181 | struct anon_vma_chain *avc; |
7906d00c AA |
3182 | |
3183 | BUG_ON(down_read_trylock(&mm->mmap_sem)); | |
3184 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); | |
3185 | ||
3186 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3187 | if (vma->anon_vma) | |
5beb4930 RR |
3188 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
3189 | vm_unlock_anon_vma(avc->anon_vma); | |
7906d00c AA |
3190 | if (vma->vm_file && vma->vm_file->f_mapping) |
3191 | vm_unlock_mapping(vma->vm_file->f_mapping); | |
3192 | } | |
3193 | ||
3194 | mutex_unlock(&mm_all_locks_mutex); | |
3195 | } | |
8feae131 DH |
3196 | |
3197 | /* | |
3198 | * initialise the VMA slab | |
3199 | */ | |
3200 | void __init mmap_init(void) | |
3201 | { | |
00a62ce9 KM |
3202 | int ret; |
3203 | ||
3204 | ret = percpu_counter_init(&vm_committed_as, 0); | |
3205 | VM_BUG_ON(ret); | |
8feae131 | 3206 | } |
c9b1d098 AS |
3207 | |
3208 | /* | |
3209 | * Initialise sysctl_user_reserve_kbytes. | |
3210 | * | |
3211 | * This is intended to prevent a user from starting a single memory hogging | |
3212 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
3213 | * mode. | |
3214 | * | |
3215 | * The default value is min(3% of free memory, 128MB) | |
3216 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
3217 | */ | |
1640879a | 3218 | static int init_user_reserve(void) |
c9b1d098 AS |
3219 | { |
3220 | unsigned long free_kbytes; | |
3221 | ||
3222 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3223 | ||
3224 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
3225 | return 0; | |
3226 | } | |
a64fb3cd | 3227 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
3228 | |
3229 | /* | |
3230 | * Initialise sysctl_admin_reserve_kbytes. | |
3231 | * | |
3232 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
3233 | * to log in and kill a memory hogging process. | |
3234 | * | |
3235 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
3236 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
3237 | * only reserve 3% of free pages by default. | |
3238 | */ | |
1640879a | 3239 | static int init_admin_reserve(void) |
4eeab4f5 AS |
3240 | { |
3241 | unsigned long free_kbytes; | |
3242 | ||
3243 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3244 | ||
3245 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
3246 | return 0; | |
3247 | } | |
a64fb3cd | 3248 | subsys_initcall(init_admin_reserve); |
1640879a AS |
3249 | |
3250 | /* | |
3251 | * Reinititalise user and admin reserves if memory is added or removed. | |
3252 | * | |
3253 | * The default user reserve max is 128MB, and the default max for the | |
3254 | * admin reserve is 8MB. These are usually, but not always, enough to | |
3255 | * enable recovery from a memory hogging process using login/sshd, a shell, | |
3256 | * and tools like top. It may make sense to increase or even disable the | |
3257 | * reserve depending on the existence of swap or variations in the recovery | |
3258 | * tools. So, the admin may have changed them. | |
3259 | * | |
3260 | * If memory is added and the reserves have been eliminated or increased above | |
3261 | * the default max, then we'll trust the admin. | |
3262 | * | |
3263 | * If memory is removed and there isn't enough free memory, then we | |
3264 | * need to reset the reserves. | |
3265 | * | |
3266 | * Otherwise keep the reserve set by the admin. | |
3267 | */ | |
3268 | static int reserve_mem_notifier(struct notifier_block *nb, | |
3269 | unsigned long action, void *data) | |
3270 | { | |
3271 | unsigned long tmp, free_kbytes; | |
3272 | ||
3273 | switch (action) { | |
3274 | case MEM_ONLINE: | |
3275 | /* Default max is 128MB. Leave alone if modified by operator. */ | |
3276 | tmp = sysctl_user_reserve_kbytes; | |
3277 | if (0 < tmp && tmp < (1UL << 17)) | |
3278 | init_user_reserve(); | |
3279 | ||
3280 | /* Default max is 8MB. Leave alone if modified by operator. */ | |
3281 | tmp = sysctl_admin_reserve_kbytes; | |
3282 | if (0 < tmp && tmp < (1UL << 13)) | |
3283 | init_admin_reserve(); | |
3284 | ||
3285 | break; | |
3286 | case MEM_OFFLINE: | |
3287 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
3288 | ||
3289 | if (sysctl_user_reserve_kbytes > free_kbytes) { | |
3290 | init_user_reserve(); | |
3291 | pr_info("vm.user_reserve_kbytes reset to %lu\n", | |
3292 | sysctl_user_reserve_kbytes); | |
3293 | } | |
3294 | ||
3295 | if (sysctl_admin_reserve_kbytes > free_kbytes) { | |
3296 | init_admin_reserve(); | |
3297 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", | |
3298 | sysctl_admin_reserve_kbytes); | |
3299 | } | |
3300 | break; | |
3301 | default: | |
3302 | break; | |
3303 | } | |
3304 | return NOTIFY_OK; | |
3305 | } | |
3306 | ||
3307 | static struct notifier_block reserve_mem_nb = { | |
3308 | .notifier_call = reserve_mem_notifier, | |
3309 | }; | |
3310 | ||
3311 | static int __meminit init_reserve_notifier(void) | |
3312 | { | |
3313 | if (register_hotmemory_notifier(&reserve_mem_nb)) | |
b1de0d13 | 3314 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
1640879a AS |
3315 | |
3316 | return 0; | |
3317 | } | |
a64fb3cd | 3318 | subsys_initcall(init_reserve_notifier); |