]> git.ipfire.org Git - thirdparty/linux.git/blame - mm/mmap.c
mailmap: update Matti Vaittinen's email address
[thirdparty/linux.git] / mm / mmap.c
CommitLineData
457c8996 1// SPDX-License-Identifier: GPL-2.0-only
1da177e4
LT
2/*
3 * mm/mmap.c
4 *
5 * Written by obz.
6 *
046c6884 7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
1da177e4
LT
8 */
9
b1de0d13
MH
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
e8420a8e 12#include <linux/kernel.h>
1da177e4 13#include <linux/slab.h>
4af3c9cc 14#include <linux/backing-dev.h>
1da177e4 15#include <linux/mm.h>
17fca131 16#include <linux/mm_inline.h>
1da177e4
LT
17#include <linux/shm.h>
18#include <linux/mman.h>
19#include <linux/pagemap.h>
20#include <linux/swap.h>
21#include <linux/syscalls.h>
c59ede7b 22#include <linux/capability.h>
1da177e4
LT
23#include <linux/init.h>
24#include <linux/file.h>
25#include <linux/fs.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/hugetlb.h>
c01d5b30 29#include <linux/shmem_fs.h>
1da177e4 30#include <linux/profile.h>
b95f1b31 31#include <linux/export.h>
1da177e4
LT
32#include <linux/mount.h>
33#include <linux/mempolicy.h>
34#include <linux/rmap.h>
cddb8a5c 35#include <linux/mmu_notifier.h>
82f71ae4 36#include <linux/mmdebug.h>
cdd6c482 37#include <linux/perf_event.h>
120a795d 38#include <linux/audit.h>
b15d00b6 39#include <linux/khugepaged.h>
2b144498 40#include <linux/uprobes.h>
1640879a
AS
41#include <linux/notifier.h>
42#include <linux/memory.h>
b1de0d13 43#include <linux/printk.h>
19a809af 44#include <linux/userfaultfd_k.h>
d977d56c 45#include <linux/moduleparam.h>
62b5f7d0 46#include <linux/pkeys.h>
21292580 47#include <linux/oom.h>
04f5866e 48#include <linux/sched/mm.h>
1da177e4 49
7c0f6ba6 50#include <linux/uaccess.h>
1da177e4
LT
51#include <asm/cacheflush.h>
52#include <asm/tlb.h>
d6dd61c8 53#include <asm/mmu_context.h>
1da177e4 54
df529cab
JK
55#define CREATE_TRACE_POINTS
56#include <trace/events/mmap.h>
57
42b77728
JB
58#include "internal.h"
59
3a459756
KK
60#ifndef arch_mmap_check
61#define arch_mmap_check(addr, len, flags) (0)
62#endif
63
d07e2259
DC
64#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
65const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
66const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
67int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
68#endif
69#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
70const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
71const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
72int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
73#endif
74
f4fcd558 75static bool ignore_rlimit_data;
d977d56c 76core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
d07e2259 77
763ecb03 78static void unmap_region(struct mm_struct *mm, struct maple_tree *mt,
e0da382c 79 struct vm_area_struct *vma, struct vm_area_struct *prev,
763ecb03
LH
80 struct vm_area_struct *next, unsigned long start,
81 unsigned long end);
e0da382c 82
64e45507
PF
83static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
84{
85 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
86}
87
88/* Update vma->vm_page_prot to reflect vma->vm_flags. */
89void vma_set_page_prot(struct vm_area_struct *vma)
90{
91 unsigned long vm_flags = vma->vm_flags;
6d2329f8 92 pgprot_t vm_page_prot;
64e45507 93
6d2329f8
AA
94 vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
95 if (vma_wants_writenotify(vma, vm_page_prot)) {
64e45507 96 vm_flags &= ~VM_SHARED;
6d2329f8 97 vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
64e45507 98 }
c1e8d7c6 99 /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */
6d2329f8 100 WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
64e45507
PF
101}
102
1da177e4 103/*
c8c06efa 104 * Requires inode->i_mapping->i_mmap_rwsem
1da177e4
LT
105 */
106static void __remove_shared_vm_struct(struct vm_area_struct *vma,
107 struct file *file, struct address_space *mapping)
108{
1da177e4 109 if (vma->vm_flags & VM_SHARED)
4bb5f5d9 110 mapping_unmap_writable(mapping);
1da177e4
LT
111
112 flush_dcache_mmap_lock(mapping);
27ba0644 113 vma_interval_tree_remove(vma, &mapping->i_mmap);
1da177e4
LT
114 flush_dcache_mmap_unlock(mapping);
115}
116
117/*
6b2dbba8 118 * Unlink a file-based vm structure from its interval tree, to hide
a8fb5618 119 * vma from rmap and vmtruncate before freeing its page tables.
1da177e4 120 */
a8fb5618 121void unlink_file_vma(struct vm_area_struct *vma)
1da177e4
LT
122{
123 struct file *file = vma->vm_file;
124
1da177e4
LT
125 if (file) {
126 struct address_space *mapping = file->f_mapping;
83cde9e8 127 i_mmap_lock_write(mapping);
1da177e4 128 __remove_shared_vm_struct(vma, file, mapping);
83cde9e8 129 i_mmap_unlock_write(mapping);
1da177e4 130 }
a8fb5618
HD
131}
132
133/*
763ecb03 134 * Close a vm structure and free it.
a8fb5618 135 */
763ecb03 136static void remove_vma(struct vm_area_struct *vma)
a8fb5618 137{
a8fb5618 138 might_sleep();
1da177e4
LT
139 if (vma->vm_ops && vma->vm_ops->close)
140 vma->vm_ops->close(vma);
e9714acf 141 if (vma->vm_file)
a8fb5618 142 fput(vma->vm_file);
f0be3d32 143 mpol_put(vma_policy(vma));
3928d4f5 144 vm_area_free(vma);
1da177e4
LT
145}
146
2e7ce7d3
LH
147/*
148 * check_brk_limits() - Use platform specific check of range & verify mlock
149 * limits.
150 * @addr: The address to check
151 * @len: The size of increase.
152 *
153 * Return: 0 on success.
154 */
155static int check_brk_limits(unsigned long addr, unsigned long len)
156{
157 unsigned long mapped_addr;
158
159 mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
160 if (IS_ERR_VALUE(mapped_addr))
161 return mapped_addr;
162
163 return mlock_future_check(current->mm, current->mm->def_flags, len);
164}
165static int do_brk_munmap(struct ma_state *mas, struct vm_area_struct *vma,
166 unsigned long newbrk, unsigned long oldbrk,
167 struct list_head *uf);
168static int do_brk_flags(struct ma_state *mas, struct vm_area_struct *brkvma,
763ecb03 169 unsigned long addr, unsigned long request, unsigned long flags);
6a6160a7 170SYSCALL_DEFINE1(brk, unsigned long, brk)
1da177e4 171{
9bc8039e 172 unsigned long newbrk, oldbrk, origbrk;
1da177e4 173 struct mm_struct *mm = current->mm;
2e7ce7d3 174 struct vm_area_struct *brkvma, *next = NULL;
a5b4592c 175 unsigned long min_brk;
128557ff 176 bool populate;
9bc8039e 177 bool downgraded = false;
897ab3e0 178 LIST_HEAD(uf);
2e7ce7d3 179 MA_STATE(mas, &mm->mm_mt, 0, 0);
1da177e4 180
d8ed45c5 181 if (mmap_write_lock_killable(mm))
dc0ef0df 182 return -EINTR;
1da177e4 183
9bc8039e
YS
184 origbrk = mm->brk;
185
a5b4592c 186#ifdef CONFIG_COMPAT_BRK
5520e894
JK
187 /*
188 * CONFIG_COMPAT_BRK can still be overridden by setting
189 * randomize_va_space to 2, which will still cause mm->start_brk
190 * to be arbitrarily shifted
191 */
4471a675 192 if (current->brk_randomized)
5520e894
JK
193 min_brk = mm->start_brk;
194 else
195 min_brk = mm->end_data;
a5b4592c
JK
196#else
197 min_brk = mm->start_brk;
198#endif
199 if (brk < min_brk)
1da177e4 200 goto out;
1e624196
RG
201
202 /*
203 * Check against rlimit here. If this check is done later after the test
204 * of oldbrk with newbrk then it can escape the test and let the data
205 * segment grow beyond its set limit the in case where the limit is
206 * not page aligned -Ram Gupta
207 */
8764b338
CG
208 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
209 mm->end_data, mm->start_data))
1e624196
RG
210 goto out;
211
1da177e4
LT
212 newbrk = PAGE_ALIGN(brk);
213 oldbrk = PAGE_ALIGN(mm->brk);
9bc8039e
YS
214 if (oldbrk == newbrk) {
215 mm->brk = brk;
216 goto success;
217 }
1da177e4 218
9bc8039e
YS
219 /*
220 * Always allow shrinking brk.
2e7ce7d3 221 * do_brk_munmap() may downgrade mmap_lock to read.
9bc8039e 222 */
1da177e4 223 if (brk <= mm->brk) {
9bc8039e
YS
224 int ret;
225
2e7ce7d3
LH
226 /* Search one past newbrk */
227 mas_set(&mas, newbrk);
228 brkvma = mas_find(&mas, oldbrk);
229 BUG_ON(brkvma == NULL);
230 if (brkvma->vm_start >= oldbrk)
231 goto out; /* mapping intersects with an existing non-brk vma. */
9bc8039e 232 /*
2e7ce7d3
LH
233 * mm->brk must be protected by write mmap_lock.
234 * do_brk_munmap() may downgrade the lock, so update it
235 * before calling do_brk_munmap().
9bc8039e
YS
236 */
237 mm->brk = brk;
2e7ce7d3
LH
238 ret = do_brk_munmap(&mas, brkvma, newbrk, oldbrk, &uf);
239 if (ret == 1) {
9bc8039e 240 downgraded = true;
2e7ce7d3
LH
241 goto success;
242 } else if (!ret)
243 goto success;
244
245 mm->brk = origbrk;
246 goto out;
1da177e4
LT
247 }
248
2e7ce7d3
LH
249 if (check_brk_limits(oldbrk, newbrk - oldbrk))
250 goto out;
251
252 /*
253 * Only check if the next VMA is within the stack_guard_gap of the
254 * expansion area
255 */
256 mas_set(&mas, oldbrk);
257 next = mas_find(&mas, newbrk - 1 + PAGE_SIZE + stack_guard_gap);
1be7107f 258 if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
1da177e4
LT
259 goto out;
260
2e7ce7d3 261 brkvma = mas_prev(&mas, mm->start_brk);
1da177e4 262 /* Ok, looks good - let it rip. */
2e7ce7d3 263 if (do_brk_flags(&mas, brkvma, oldbrk, newbrk - oldbrk, 0) < 0)
1da177e4 264 goto out;
2e7ce7d3 265
1da177e4 266 mm->brk = brk;
9bc8039e
YS
267
268success:
128557ff 269 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
9bc8039e 270 if (downgraded)
d8ed45c5 271 mmap_read_unlock(mm);
9bc8039e 272 else
d8ed45c5 273 mmap_write_unlock(mm);
897ab3e0 274 userfaultfd_unmap_complete(mm, &uf);
128557ff
ML
275 if (populate)
276 mm_populate(oldbrk, newbrk - oldbrk);
277 return brk;
278
1da177e4 279out:
d8ed45c5 280 mmap_write_unlock(mm);
b7204006 281 return origbrk;
1da177e4
LT
282}
283
d4af56c5
LH
284#if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
285extern void mt_validate(struct maple_tree *mt);
286extern void mt_dump(const struct maple_tree *mt);
1da177e4 287
d4af56c5
LH
288/* Validate the maple tree */
289static void validate_mm_mt(struct mm_struct *mm)
290{
291 struct maple_tree *mt = &mm->mm_mt;
763ecb03 292 struct vm_area_struct *vma_mt;
d4af56c5
LH
293
294 MA_STATE(mas, mt, 0, 0);
295
296 mt_validate(&mm->mm_mt);
297 mas_for_each(&mas, vma_mt, ULONG_MAX) {
763ecb03
LH
298 if ((vma_mt->vm_start != mas.index) ||
299 (vma_mt->vm_end - 1 != mas.last)) {
d4af56c5
LH
300 pr_emerg("issue in %s\n", current->comm);
301 dump_stack();
d4af56c5 302 dump_vma(vma_mt);
d4af56c5
LH
303 pr_emerg("mt piv: %p %lu - %lu\n", vma_mt,
304 mas.index, mas.last);
305 pr_emerg("mt vma: %p %lu - %lu\n", vma_mt,
306 vma_mt->vm_start, vma_mt->vm_end);
d4af56c5
LH
307
308 mt_dump(mas.tree);
309 if (vma_mt->vm_end != mas.last + 1) {
310 pr_err("vma: %p vma_mt %lu-%lu\tmt %lu-%lu\n",
311 mm, vma_mt->vm_start, vma_mt->vm_end,
312 mas.index, mas.last);
313 mt_dump(mas.tree);
314 }
315 VM_BUG_ON_MM(vma_mt->vm_end != mas.last + 1, mm);
316 if (vma_mt->vm_start != mas.index) {
317 pr_err("vma: %p vma_mt %p %lu - %lu doesn't match\n",
318 mm, vma_mt, vma_mt->vm_start, vma_mt->vm_end);
319 mt_dump(mas.tree);
320 }
321 VM_BUG_ON_MM(vma_mt->vm_start != mas.index, mm);
322 }
d4af56c5 323 }
d4af56c5 324}
1da177e4 325
eafd4dc4 326static void validate_mm(struct mm_struct *mm)
1da177e4
LT
327{
328 int bug = 0;
329 int i = 0;
763ecb03
LH
330 struct vm_area_struct *vma;
331 MA_STATE(mas, &mm->mm_mt, 0, 0);
ff26f70f 332
524e00b3
LH
333 validate_mm_mt(mm);
334
763ecb03 335 mas_for_each(&mas, vma, ULONG_MAX) {
524e00b3 336#ifdef CONFIG_DEBUG_VM_RB
12352d3c 337 struct anon_vma *anon_vma = vma->anon_vma;
ed8ea815 338 struct anon_vma_chain *avc;
ff26f70f 339
12352d3c
KK
340 if (anon_vma) {
341 anon_vma_lock_read(anon_vma);
342 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
343 anon_vma_interval_tree_verify(avc);
344 anon_vma_unlock_read(anon_vma);
345 }
524e00b3 346#endif
1da177e4
LT
347 i++;
348 }
5a0768f6 349 if (i != mm->map_count) {
763ecb03 350 pr_emerg("map_count %d mas_for_each %d\n", mm->map_count, i);
5a0768f6
ML
351 bug = 1;
352 }
96dad67f 353 VM_BUG_ON_MM(bug, mm);
1da177e4 354}
524e00b3
LH
355
356#else /* !CONFIG_DEBUG_VM_MAPLE_TREE */
d4af56c5 357#define validate_mm_mt(root) do { } while (0)
1da177e4 358#define validate_mm(mm) do { } while (0)
524e00b3 359#endif /* CONFIG_DEBUG_VM_MAPLE_TREE */
8f26e0b1 360
bf181b9f
ML
361/*
362 * vma has some anon_vma assigned, and is already inserted on that
363 * anon_vma's interval trees.
364 *
365 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
366 * vma must be removed from the anon_vma's interval trees using
367 * anon_vma_interval_tree_pre_update_vma().
368 *
369 * After the update, the vma will be reinserted using
370 * anon_vma_interval_tree_post_update_vma().
371 *
c1e8d7c6 372 * The entire update must be protected by exclusive mmap_lock and by
bf181b9f
ML
373 * the root anon_vma's mutex.
374 */
375static inline void
376anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
377{
378 struct anon_vma_chain *avc;
379
380 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
381 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
382}
383
384static inline void
385anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
386{
387 struct anon_vma_chain *avc;
388
389 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
390 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
391}
392
e8420a8e
CH
393static unsigned long count_vma_pages_range(struct mm_struct *mm,
394 unsigned long addr, unsigned long end)
395{
2e3af1db 396 VMA_ITERATOR(vmi, mm, addr);
e8420a8e 397 struct vm_area_struct *vma;
2e3af1db 398 unsigned long nr_pages = 0;
e8420a8e 399
2e3af1db
MWO
400 for_each_vma_range(vmi, vma, end) {
401 unsigned long vm_start = max(addr, vma->vm_start);
402 unsigned long vm_end = min(end, vma->vm_end);
e8420a8e 403
2e3af1db 404 nr_pages += PHYS_PFN(vm_end - vm_start);
e8420a8e
CH
405 }
406
407 return nr_pages;
408}
409
c154124f
LH
410static void __vma_link_file(struct vm_area_struct *vma,
411 struct address_space *mapping)
1da177e4 412{
c154124f
LH
413 if (vma->vm_flags & VM_SHARED)
414 mapping_allow_writable(mapping);
1da177e4 415
c154124f
LH
416 flush_dcache_mmap_lock(mapping);
417 vma_interval_tree_insert(vma, &mapping->i_mmap);
418 flush_dcache_mmap_unlock(mapping);
1da177e4
LT
419}
420
d4af56c5
LH
421/*
422 * vma_mas_store() - Store a VMA in the maple tree.
423 * @vma: The vm_area_struct
424 * @mas: The maple state
425 *
426 * Efficient way to store a VMA in the maple tree when the @mas has already
427 * walked to the correct location.
428 *
429 * Note: the end address is inclusive in the maple tree.
430 */
431void vma_mas_store(struct vm_area_struct *vma, struct ma_state *mas)
432{
433 trace_vma_store(mas->tree, vma);
434 mas_set_range(mas, vma->vm_start, vma->vm_end - 1);
435 mas_store_prealloc(mas, vma);
436}
437
438/*
439 * vma_mas_remove() - Remove a VMA from the maple tree.
440 * @vma: The vm_area_struct
441 * @mas: The maple state
442 *
443 * Efficient way to remove a VMA from the maple tree when the @mas has already
444 * been established and points to the correct location.
445 * Note: the end address is inclusive in the maple tree.
446 */
447void vma_mas_remove(struct vm_area_struct *vma, struct ma_state *mas)
448{
449 trace_vma_mas_szero(mas->tree, vma->vm_start, vma->vm_end - 1);
450 mas->index = vma->vm_start;
451 mas->last = vma->vm_end - 1;
452 mas_store_prealloc(mas, NULL);
453}
454
455/*
456 * vma_mas_szero() - Set a given range to zero. Used when modifying a
457 * vm_area_struct start or end.
458 *
4a423440 459 * @mas: The maple tree ma_state
d4af56c5
LH
460 * @start: The start address to zero
461 * @end: The end address to zero.
462 */
463static inline void vma_mas_szero(struct ma_state *mas, unsigned long start,
464 unsigned long end)
465{
466 trace_vma_mas_szero(mas->tree, start, end - 1);
467 mas_set_range(mas, start, end - 1);
468 mas_store_prealloc(mas, NULL);
469}
470
763ecb03 471static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 472{
d4af56c5 473 MA_STATE(mas, &mm->mm_mt, 0, 0);
1da177e4
LT
474 struct address_space *mapping = NULL;
475
d4af56c5
LH
476 if (mas_preallocate(&mas, vma, GFP_KERNEL))
477 return -ENOMEM;
478
64ac4940 479 if (vma->vm_file) {
1da177e4 480 mapping = vma->vm_file->f_mapping;
83cde9e8 481 i_mmap_lock_write(mapping);
64ac4940 482 }
1da177e4 483
d4af56c5 484 vma_mas_store(vma, &mas);
1da177e4 485
c154124f
LH
486 if (mapping) {
487 __vma_link_file(vma, mapping);
83cde9e8 488 i_mmap_unlock_write(mapping);
c154124f 489 }
1da177e4
LT
490
491 mm->map_count++;
492 validate_mm(mm);
d4af56c5 493 return 0;
1da177e4
LT
494}
495
4dd1b841
LH
496/*
497 * vma_expand - Expand an existing VMA
498 *
499 * @mas: The maple state
500 * @vma: The vma to expand
501 * @start: The start of the vma
502 * @end: The exclusive end of the vma
503 * @pgoff: The page offset of vma
504 * @next: The current of next vma.
505 *
506 * Expand @vma to @start and @end. Can expand off the start and end. Will
507 * expand over @next if it's different from @vma and @end == @next->vm_end.
508 * Checking if the @vma can expand and merge with @next needs to be handled by
509 * the caller.
510 *
511 * Returns: 0 on success
512 */
513inline int vma_expand(struct ma_state *mas, struct vm_area_struct *vma,
514 unsigned long start, unsigned long end, pgoff_t pgoff,
515 struct vm_area_struct *next)
516{
517 struct mm_struct *mm = vma->vm_mm;
518 struct address_space *mapping = NULL;
519 struct rb_root_cached *root = NULL;
520 struct anon_vma *anon_vma = vma->anon_vma;
521 struct file *file = vma->vm_file;
522 bool remove_next = false;
523
524 if (next && (vma != next) && (end == next->vm_end)) {
525 remove_next = true;
526 if (next->anon_vma && !vma->anon_vma) {
527 int error;
528
529 anon_vma = next->anon_vma;
530 vma->anon_vma = anon_vma;
531 error = anon_vma_clone(vma, next);
532 if (error)
533 return error;
534 }
535 }
536
537 /* Not merging but overwriting any part of next is not handled. */
538 VM_BUG_ON(next && !remove_next && next != vma && end > next->vm_start);
539 /* Only handles expanding */
540 VM_BUG_ON(vma->vm_start < start || vma->vm_end > end);
541
542 if (mas_preallocate(mas, vma, GFP_KERNEL))
543 goto nomem;
544
545 vma_adjust_trans_huge(vma, start, end, 0);
546
547 if (file) {
548 mapping = file->f_mapping;
549 root = &mapping->i_mmap;
550 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
551 i_mmap_lock_write(mapping);
552 }
553
554 if (anon_vma) {
555 anon_vma_lock_write(anon_vma);
556 anon_vma_interval_tree_pre_update_vma(vma);
557 }
558
559 if (file) {
560 flush_dcache_mmap_lock(mapping);
561 vma_interval_tree_remove(vma, root);
562 }
563
564 vma->vm_start = start;
565 vma->vm_end = end;
566 vma->vm_pgoff = pgoff;
567 /* Note: mas must be pointing to the expanding VMA */
568 vma_mas_store(vma, mas);
569
570 if (file) {
571 vma_interval_tree_insert(vma, root);
572 flush_dcache_mmap_unlock(mapping);
573 }
574
575 /* Expanding over the next vma */
763ecb03
LH
576 if (remove_next && file) {
577 __remove_shared_vm_struct(next, file, mapping);
4dd1b841
LH
578 }
579
580 if (anon_vma) {
581 anon_vma_interval_tree_post_update_vma(vma);
582 anon_vma_unlock_write(anon_vma);
583 }
584
585 if (file) {
586 i_mmap_unlock_write(mapping);
587 uprobe_mmap(vma);
588 }
589
590 if (remove_next) {
591 if (file) {
592 uprobe_munmap(next, next->vm_start, next->vm_end);
593 fput(file);
594 }
595 if (next->anon_vma)
596 anon_vma_merge(vma, next);
597 mm->map_count--;
598 mpol_put(vma_policy(next));
599 vm_area_free(next);
600 }
601
602 validate_mm(mm);
603 return 0;
604
605nomem:
606 return -ENOMEM;
607}
608
1da177e4
LT
609/*
610 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
611 * is already present in an i_mmap tree without adjusting the tree.
612 * The following helper function should be used when such adjustments
613 * are necessary. The "insert" vma (if any) is to be inserted
614 * before we drop the necessary locks.
615 */
e86f15ee
AA
616int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
617 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
618 struct vm_area_struct *expand)
1da177e4
LT
619{
620 struct mm_struct *mm = vma->vm_mm;
1cd916d0
AM
621 struct vm_area_struct *next_next = NULL; /* uninit var warning */
622 struct vm_area_struct *next = find_vma(mm, vma->vm_end);
524e00b3 623 struct vm_area_struct *orig_vma = vma;
1da177e4 624 struct address_space *mapping = NULL;
f808c13f 625 struct rb_root_cached *root = NULL;
012f1800 626 struct anon_vma *anon_vma = NULL;
1da177e4 627 struct file *file = vma->vm_file;
524e00b3 628 bool vma_changed = false;
1da177e4
LT
629 long adjust_next = 0;
630 int remove_next = 0;
d4af56c5
LH
631 MA_STATE(mas, &mm->mm_mt, 0, 0);
632 struct vm_area_struct *exporter = NULL, *importer = NULL;
1da177e4 633
d4af56c5 634 if (next && !insert) {
1da177e4
LT
635 if (end >= next->vm_end) {
636 /*
637 * vma expands, overlapping all the next, and
638 * perhaps the one after too (mprotect case 6).
86d12e47 639 * The only other cases that gets here are
e86f15ee 640 * case 1, case 7 and case 8.
1da177e4 641 */
e86f15ee
AA
642 if (next == expand) {
643 /*
644 * The only case where we don't expand "vma"
645 * and we expand "next" instead is case 8.
646 */
647 VM_WARN_ON(end != next->vm_end);
648 /*
649 * remove_next == 3 means we're
650 * removing "vma" and that to do so we
651 * swapped "vma" and "next".
652 */
653 remove_next = 3;
654 VM_WARN_ON(file != next->vm_file);
655 swap(vma, next);
656 } else {
657 VM_WARN_ON(expand != vma);
658 /*
659 * case 1, 6, 7, remove_next == 2 is case 6,
660 * remove_next == 1 is case 1 or 7.
661 */
662 remove_next = 1 + (end > next->vm_end);
d4af56c5
LH
663 if (remove_next == 2)
664 next_next = find_vma(mm, next->vm_end);
665
e86f15ee 666 VM_WARN_ON(remove_next == 2 &&
763ecb03 667 end != next_next->vm_end);
e86f15ee
AA
668 }
669
287d97ac 670 exporter = next;
1da177e4 671 importer = vma;
734537c9
KS
672
673 /*
674 * If next doesn't have anon_vma, import from vma after
675 * next, if the vma overlaps with it.
676 */
97a42cd4 677 if (remove_next == 2 && !next->anon_vma)
763ecb03 678 exporter = next_next;
734537c9 679
1da177e4
LT
680 } else if (end > next->vm_start) {
681 /*
682 * vma expands, overlapping part of the next:
683 * mprotect case 5 shifting the boundary up.
684 */
f9d86a60 685 adjust_next = (end - next->vm_start);
287d97ac 686 exporter = next;
1da177e4 687 importer = vma;
e86f15ee 688 VM_WARN_ON(expand != importer);
1da177e4
LT
689 } else if (end < vma->vm_end) {
690 /*
691 * vma shrinks, and !insert tells it's not
692 * split_vma inserting another: so it must be
693 * mprotect case 4 shifting the boundary down.
694 */
f9d86a60 695 adjust_next = -(vma->vm_end - end);
287d97ac 696 exporter = vma;
1da177e4 697 importer = next;
e86f15ee 698 VM_WARN_ON(expand != importer);
1da177e4 699 }
1da177e4 700
5beb4930
RR
701 /*
702 * Easily overlooked: when mprotect shifts the boundary,
703 * make sure the expanding vma has anon_vma set if the
704 * shrinking vma had, to cover any anon pages imported.
705 */
287d97ac 706 if (exporter && exporter->anon_vma && !importer->anon_vma) {
c4ea95d7
DF
707 int error;
708
b800c91a 709 importer->anon_vma = exporter->anon_vma;
c4ea95d7 710 error = anon_vma_clone(importer, exporter);
3fe89b3e 711 if (error)
c4ea95d7 712 return error;
5beb4930
RR
713 }
714 }
37f9f559 715
d4af56c5
LH
716 if (mas_preallocate(&mas, vma, GFP_KERNEL))
717 return -ENOMEM;
718
719 vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
1da177e4
LT
720 if (file) {
721 mapping = file->f_mapping;
27ba0644
KS
722 root = &mapping->i_mmap;
723 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
682968e0 724
27ba0644
KS
725 if (adjust_next)
726 uprobe_munmap(next, next->vm_start, next->vm_end);
682968e0 727
83cde9e8 728 i_mmap_lock_write(mapping);
c154124f 729 if (insert && insert->vm_file) {
1da177e4 730 /*
6b2dbba8 731 * Put into interval tree now, so instantiated pages
1da177e4
LT
732 * are visible to arm/parisc __flush_dcache_page
733 * throughout; but we cannot insert into address
734 * space until vma start or end is updated.
735 */
c154124f 736 __vma_link_file(insert, insert->vm_file->f_mapping);
1da177e4
LT
737 }
738 }
739
bf181b9f
ML
740 anon_vma = vma->anon_vma;
741 if (!anon_vma && adjust_next)
742 anon_vma = next->anon_vma;
743 if (anon_vma) {
e86f15ee
AA
744 VM_WARN_ON(adjust_next && next->anon_vma &&
745 anon_vma != next->anon_vma);
4fc3f1d6 746 anon_vma_lock_write(anon_vma);
bf181b9f
ML
747 anon_vma_interval_tree_pre_update_vma(vma);
748 if (adjust_next)
749 anon_vma_interval_tree_pre_update_vma(next);
750 }
012f1800 751
0fc48a6e 752 if (file) {
1da177e4 753 flush_dcache_mmap_lock(mapping);
6b2dbba8 754 vma_interval_tree_remove(vma, root);
1da177e4 755 if (adjust_next)
6b2dbba8 756 vma_interval_tree_remove(next, root);
1da177e4
LT
757 }
758
d3737187 759 if (start != vma->vm_start) {
3b0e81a1
LH
760 if ((vma->vm_start < start) &&
761 (!insert || (insert->vm_end != start))) {
524e00b3 762 vma_mas_szero(&mas, vma->vm_start, start);
3b0e81a1
LH
763 VM_WARN_ON(insert && insert->vm_start > vma->vm_start);
764 } else {
765 vma_changed = true;
766 }
d3737187 767 vma->vm_start = start;
d3737187
ML
768 }
769 if (end != vma->vm_end) {
3b0e81a1
LH
770 if (vma->vm_end > end) {
771 if (!insert || (insert->vm_start != end)) {
772 vma_mas_szero(&mas, end, vma->vm_end);
763ecb03 773 mas_reset(&mas);
3b0e81a1
LH
774 VM_WARN_ON(insert &&
775 insert->vm_end < vma->vm_end);
3b0e81a1
LH
776 }
777 } else {
778 vma_changed = true;
779 }
d3737187 780 vma->vm_end = end;
d3737187 781 }
d4af56c5 782
524e00b3 783 if (vma_changed)
d4af56c5
LH
784 vma_mas_store(vma, &mas);
785
1da177e4
LT
786 vma->vm_pgoff = pgoff;
787 if (adjust_next) {
f9d86a60
WY
788 next->vm_start += adjust_next;
789 next->vm_pgoff += adjust_next >> PAGE_SHIFT;
d4af56c5 790 vma_mas_store(next, &mas);
1da177e4
LT
791 }
792
0fc48a6e 793 if (file) {
1da177e4 794 if (adjust_next)
6b2dbba8
ML
795 vma_interval_tree_insert(next, root);
796 vma_interval_tree_insert(vma, root);
1da177e4
LT
797 flush_dcache_mmap_unlock(mapping);
798 }
799
763ecb03
LH
800 if (remove_next && file) {
801 __remove_shared_vm_struct(next, file, mapping);
524e00b3 802 if (remove_next == 2)
763ecb03 803 __remove_shared_vm_struct(next_next, file, mapping);
1da177e4
LT
804 } else if (insert) {
805 /*
806 * split_vma has split insert from vma, and needs
807 * us to insert it before dropping the locks
808 * (it may either follow vma or precede it).
809 */
763ecb03
LH
810 mas_reset(&mas);
811 vma_mas_store(insert, &mas);
812 mm->map_count++;
1da177e4
LT
813 }
814
bf181b9f
ML
815 if (anon_vma) {
816 anon_vma_interval_tree_post_update_vma(vma);
817 if (adjust_next)
818 anon_vma_interval_tree_post_update_vma(next);
08b52706 819 anon_vma_unlock_write(anon_vma);
bf181b9f 820 }
1da177e4 821
0fc48a6e 822 if (file) {
808fbdbe 823 i_mmap_unlock_write(mapping);
7b2d81d4 824 uprobe_mmap(vma);
2b144498
SD
825
826 if (adjust_next)
7b2d81d4 827 uprobe_mmap(next);
2b144498
SD
828 }
829
1da177e4 830 if (remove_next) {
d4af56c5 831again:
925d1c40 832 if (file) {
cbc91f71 833 uprobe_munmap(next, next->vm_start, next->vm_end);
1da177e4 834 fput(file);
925d1c40 835 }
5beb4930
RR
836 if (next->anon_vma)
837 anon_vma_merge(vma, next);
1da177e4 838 mm->map_count--;
3964acd0 839 mpol_put(vma_policy(next));
524e00b3
LH
840 if (remove_next != 2)
841 BUG_ON(vma->vm_end < next->vm_end);
3928d4f5 842 vm_area_free(next);
524e00b3 843
1da177e4
LT
844 /*
845 * In mprotect's case 6 (see comments on vma_merge),
763ecb03 846 * we must remove next_next too.
1da177e4 847 */
734537c9
KS
848 if (remove_next == 2) {
849 remove_next = 1;
763ecb03 850 next = next_next;
1da177e4 851 goto again;
fb8c41e9 852 }
1da177e4 853 }
2b144498 854 if (insert && file)
7b2d81d4 855 uprobe_mmap(insert);
1da177e4 856
3b0e81a1 857 mas_destroy(&mas);
1da177e4 858 validate_mm(mm);
763ecb03 859
5beb4930 860 return 0;
1da177e4
LT
861}
862
863/*
864 * If the vma has a ->close operation then the driver probably needs to release
865 * per-vma resources, so we don't attempt to merge those.
866 */
1da177e4 867static inline int is_mergeable_vma(struct vm_area_struct *vma,
19a809af 868 struct file *file, unsigned long vm_flags,
9a10064f 869 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
5c26f6ac 870 struct anon_vma_name *anon_name)
1da177e4 871{
34228d47
CG
872 /*
873 * VM_SOFTDIRTY should not prevent from VMA merging, if we
874 * match the flags but dirty bit -- the caller should mark
875 * merged VMA as dirty. If dirty bit won't be excluded from
8bb4e7a2 876 * comparison, we increase pressure on the memory system forcing
34228d47
CG
877 * the kernel to generate new VMAs when old one could be
878 * extended instead.
879 */
880 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
1da177e4
LT
881 return 0;
882 if (vma->vm_file != file)
883 return 0;
884 if (vma->vm_ops && vma->vm_ops->close)
885 return 0;
19a809af
AA
886 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
887 return 0;
5c26f6ac 888 if (!anon_vma_name_eq(anon_vma_name(vma), anon_name))
9a10064f 889 return 0;
1da177e4
LT
890 return 1;
891}
892
893static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
965f55de
SL
894 struct anon_vma *anon_vma2,
895 struct vm_area_struct *vma)
1da177e4 896{
965f55de
SL
897 /*
898 * The list_is_singular() test is to avoid merging VMA cloned from
899 * parents. This can improve scalability caused by anon_vma lock.
900 */
901 if ((!anon_vma1 || !anon_vma2) && (!vma ||
902 list_is_singular(&vma->anon_vma_chain)))
903 return 1;
904 return anon_vma1 == anon_vma2;
1da177e4
LT
905}
906
907/*
908 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
909 * in front of (at a lower virtual address and file offset than) the vma.
910 *
911 * We cannot merge two vmas if they have differently assigned (non-NULL)
912 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
913 *
914 * We don't check here for the merged mmap wrapping around the end of pagecache
45e55300 915 * indices (16TB on ia32) because do_mmap() does not permit mmap's which
1da177e4
LT
916 * wrap, nor mmaps which cover the final page at index -1UL.
917 */
918static int
919can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
19a809af
AA
920 struct anon_vma *anon_vma, struct file *file,
921 pgoff_t vm_pgoff,
9a10064f 922 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
5c26f6ac 923 struct anon_vma_name *anon_name)
1da177e4 924{
9a10064f 925 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
965f55de 926 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1da177e4
LT
927 if (vma->vm_pgoff == vm_pgoff)
928 return 1;
929 }
930 return 0;
931}
932
933/*
934 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
935 * beyond (at a higher virtual address and file offset than) the vma.
936 *
937 * We cannot merge two vmas if they have differently assigned (non-NULL)
938 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
939 */
940static int
941can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
19a809af
AA
942 struct anon_vma *anon_vma, struct file *file,
943 pgoff_t vm_pgoff,
9a10064f 944 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
5c26f6ac 945 struct anon_vma_name *anon_name)
1da177e4 946{
9a10064f 947 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
965f55de 948 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1da177e4 949 pgoff_t vm_pglen;
d6e93217 950 vm_pglen = vma_pages(vma);
1da177e4
LT
951 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
952 return 1;
953 }
954 return 0;
955}
956
957/*
9a10064f
CC
958 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
959 * figure out whether that can be merged with its predecessor or its
960 * successor. Or both (it neatly fills a hole).
1da177e4
LT
961 *
962 * In most cases - when called for mmap, brk or mremap - [addr,end) is
963 * certain not to be mapped by the time vma_merge is called; but when
964 * called for mprotect, it is certain to be already mapped (either at
965 * an offset within prev, or at the start of next), and the flags of
966 * this area are about to be changed to vm_flags - and the no-change
967 * case has already been eliminated.
968 *
969 * The following mprotect cases have to be considered, where AAAA is
970 * the area passed down from mprotect_fixup, never extending beyond one
971 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
972 *
5d42ab29
WY
973 * AAAA AAAA AAAA
974 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN
975 * cannot merge might become might become
976 * PPNNNNNNNNNN PPPPPPPPPPNN
977 * mmap, brk or case 4 below case 5 below
978 * mremap move:
979 * AAAA AAAA
980 * PPPP NNNN PPPPNNNNXXXX
981 * might become might become
982 * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or
983 * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or
984 * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8
1da177e4 985 *
8bb4e7a2 986 * It is important for case 8 that the vma NNNN overlapping the
e86f15ee
AA
987 * region AAAA is never going to extended over XXXX. Instead XXXX must
988 * be extended in region AAAA and NNNN must be removed. This way in
989 * all cases where vma_merge succeeds, the moment vma_adjust drops the
990 * rmap_locks, the properties of the merged vma will be already
991 * correct for the whole merged range. Some of those properties like
992 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
993 * be correct for the whole merged range immediately after the
994 * rmap_locks are released. Otherwise if XXXX would be removed and
995 * NNNN would be extended over the XXXX range, remove_migration_ptes
996 * or other rmap walkers (if working on addresses beyond the "end"
997 * parameter) may establish ptes with the wrong permissions of NNNN
998 * instead of the right permissions of XXXX.
1da177e4
LT
999 */
1000struct vm_area_struct *vma_merge(struct mm_struct *mm,
1001 struct vm_area_struct *prev, unsigned long addr,
1002 unsigned long end, unsigned long vm_flags,
cc71aba3 1003 struct anon_vma *anon_vma, struct file *file,
19a809af 1004 pgoff_t pgoff, struct mempolicy *policy,
9a10064f 1005 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
5c26f6ac 1006 struct anon_vma_name *anon_name)
1da177e4
LT
1007{
1008 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
eef19944
JM
1009 struct vm_area_struct *mid, *next, *res;
1010 int err = -1;
1011 bool merge_prev = false;
1012 bool merge_next = false;
1da177e4
LT
1013
1014 /*
1015 * We later require that vma->vm_flags == vm_flags,
1016 * so this tests vma->vm_flags & VM_SPECIAL, too.
1017 */
1018 if (vm_flags & VM_SPECIAL)
1019 return NULL;
1020
763ecb03 1021 next = find_vma(mm, prev ? prev->vm_end : 0);
eef19944
JM
1022 mid = next;
1023 if (next && next->vm_end == end) /* cases 6, 7, 8 */
763ecb03 1024 next = find_vma(mm, next->vm_end);
1da177e4 1025
e86f15ee
AA
1026 /* verify some invariant that must be enforced by the caller */
1027 VM_WARN_ON(prev && addr <= prev->vm_start);
eef19944 1028 VM_WARN_ON(mid && end > mid->vm_end);
e86f15ee
AA
1029 VM_WARN_ON(addr >= end);
1030
eef19944 1031 /* Can we merge the predecessor? */
1da177e4 1032 if (prev && prev->vm_end == addr &&
cc71aba3 1033 mpol_equal(vma_policy(prev), policy) &&
1da177e4 1034 can_vma_merge_after(prev, vm_flags,
19a809af 1035 anon_vma, file, pgoff,
9a10064f 1036 vm_userfaultfd_ctx, anon_name)) {
eef19944 1037 merge_prev = true;
1da177e4 1038 }
eef19944 1039 /* Can we merge the successor? */
1da177e4 1040 if (next && end == next->vm_start &&
cc71aba3 1041 mpol_equal(policy, vma_policy(next)) &&
1da177e4 1042 can_vma_merge_before(next, vm_flags,
19a809af 1043 anon_vma, file, pgoff+pglen,
9a10064f 1044 vm_userfaultfd_ctx, anon_name)) {
eef19944
JM
1045 merge_next = true;
1046 }
1047 /* Can we merge both the predecessor and the successor? */
1048 if (merge_prev && merge_next &&
1049 is_mergeable_anon_vma(prev->anon_vma,
1050 next->anon_vma, NULL)) { /* cases 1, 6 */
1051 err = __vma_adjust(prev, prev->vm_start,
1052 next->vm_end, prev->vm_pgoff, NULL,
1053 prev);
1054 res = prev;
1055 } else if (merge_prev) { /* cases 2, 5, 7 */
1056 err = __vma_adjust(prev, prev->vm_start,
1057 end, prev->vm_pgoff, NULL, prev);
1058 res = prev;
1059 } else if (merge_next) {
1da177e4 1060 if (prev && addr < prev->vm_end) /* case 4 */
e86f15ee 1061 err = __vma_adjust(prev, prev->vm_start,
eef19944
JM
1062 addr, prev->vm_pgoff, NULL, next);
1063 else /* cases 3, 8 */
1064 err = __vma_adjust(mid, addr, next->vm_end,
1065 next->vm_pgoff - pglen, NULL, next);
1066 res = next;
1da177e4
LT
1067 }
1068
eef19944
JM
1069 /*
1070 * Cannot merge with predecessor or successor or error in __vma_adjust?
1071 */
1072 if (err)
1073 return NULL;
1074 khugepaged_enter_vma(res, vm_flags);
1075 return res;
1da177e4
LT
1076}
1077
d0e9fe17 1078/*
b4f315b4 1079 * Rough compatibility check to quickly see if it's even worth looking
d0e9fe17
LT
1080 * at sharing an anon_vma.
1081 *
1082 * They need to have the same vm_file, and the flags can only differ
1083 * in things that mprotect may change.
1084 *
1085 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1086 * we can merge the two vma's. For example, we refuse to merge a vma if
1087 * there is a vm_ops->close() function, because that indicates that the
1088 * driver is doing some kind of reference counting. But that doesn't
1089 * really matter for the anon_vma sharing case.
1090 */
1091static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1092{
1093 return a->vm_end == b->vm_start &&
1094 mpol_equal(vma_policy(a), vma_policy(b)) &&
1095 a->vm_file == b->vm_file &&
6cb4d9a2 1096 !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) &&
d0e9fe17
LT
1097 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1098}
1099
1100/*
1101 * Do some basic sanity checking to see if we can re-use the anon_vma
1102 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1103 * the same as 'old', the other will be the new one that is trying
1104 * to share the anon_vma.
1105 *
5b449489 1106 * NOTE! This runs with mmap_lock held for reading, so it is possible that
d0e9fe17
LT
1107 * the anon_vma of 'old' is concurrently in the process of being set up
1108 * by another page fault trying to merge _that_. But that's ok: if it
1109 * is being set up, that automatically means that it will be a singleton
1110 * acceptable for merging, so we can do all of this optimistically. But
4db0c3c2 1111 * we do that READ_ONCE() to make sure that we never re-load the pointer.
d0e9fe17
LT
1112 *
1113 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1114 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1115 * is to return an anon_vma that is "complex" due to having gone through
1116 * a fork).
1117 *
1118 * We also make sure that the two vma's are compatible (adjacent,
1119 * and with the same memory policies). That's all stable, even with just
5b449489 1120 * a read lock on the mmap_lock.
d0e9fe17
LT
1121 */
1122static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1123{
1124 if (anon_vma_compatible(a, b)) {
4db0c3c2 1125 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
d0e9fe17
LT
1126
1127 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1128 return anon_vma;
1129 }
1130 return NULL;
1131}
1132
1da177e4
LT
1133/*
1134 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1135 * neighbouring vmas for a suitable anon_vma, before it goes off
1136 * to allocate a new anon_vma. It checks because a repetitive
1137 * sequence of mprotects and faults may otherwise lead to distinct
1138 * anon_vmas being allocated, preventing vma merge in subsequent
1139 * mprotect.
1140 */
1141struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1142{
763ecb03 1143 MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end);
a67c8caa 1144 struct anon_vma *anon_vma = NULL;
763ecb03 1145 struct vm_area_struct *prev, *next;
a67c8caa
ML
1146
1147 /* Try next first. */
763ecb03
LH
1148 next = mas_walk(&mas);
1149 if (next) {
1150 anon_vma = reusable_anon_vma(next, vma, next);
a67c8caa
ML
1151 if (anon_vma)
1152 return anon_vma;
1153 }
1154
763ecb03
LH
1155 prev = mas_prev(&mas, 0);
1156 VM_BUG_ON_VMA(prev != vma, vma);
1157 prev = mas_prev(&mas, 0);
a67c8caa 1158 /* Try prev next. */
763ecb03
LH
1159 if (prev)
1160 anon_vma = reusable_anon_vma(prev, prev, vma);
a67c8caa 1161
1da177e4 1162 /*
a67c8caa
ML
1163 * We might reach here with anon_vma == NULL if we can't find
1164 * any reusable anon_vma.
1da177e4
LT
1165 * There's no absolute need to look only at touching neighbours:
1166 * we could search further afield for "compatible" anon_vmas.
1167 * But it would probably just be a waste of time searching,
1168 * or lead to too many vmas hanging off the same anon_vma.
1169 * We're trying to allow mprotect remerging later on,
1170 * not trying to minimize memory used for anon_vmas.
1171 */
a67c8caa 1172 return anon_vma;
1da177e4
LT
1173}
1174
40401530
AV
1175/*
1176 * If a hint addr is less than mmap_min_addr change hint to be as
1177 * low as possible but still greater than mmap_min_addr
1178 */
1179static inline unsigned long round_hint_to_min(unsigned long hint)
1180{
1181 hint &= PAGE_MASK;
1182 if (((void *)hint != NULL) &&
1183 (hint < mmap_min_addr))
1184 return PAGE_ALIGN(mmap_min_addr);
1185 return hint;
1186}
1187
6aeb2542
MR
1188int mlock_future_check(struct mm_struct *mm, unsigned long flags,
1189 unsigned long len)
363ee17f
DB
1190{
1191 unsigned long locked, lock_limit;
1192
1193 /* mlock MCL_FUTURE? */
1194 if (flags & VM_LOCKED) {
1195 locked = len >> PAGE_SHIFT;
1196 locked += mm->locked_vm;
1197 lock_limit = rlimit(RLIMIT_MEMLOCK);
1198 lock_limit >>= PAGE_SHIFT;
1199 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1200 return -EAGAIN;
1201 }
1202 return 0;
1203}
1204
be83bbf8
LT
1205static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1206{
1207 if (S_ISREG(inode->i_mode))
423913ad 1208 return MAX_LFS_FILESIZE;
be83bbf8
LT
1209
1210 if (S_ISBLK(inode->i_mode))
1211 return MAX_LFS_FILESIZE;
1212
76f34950
IK
1213 if (S_ISSOCK(inode->i_mode))
1214 return MAX_LFS_FILESIZE;
1215
be83bbf8
LT
1216 /* Special "we do even unsigned file positions" case */
1217 if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1218 return 0;
1219
1220 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1221 return ULONG_MAX;
1222}
1223
1224static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1225 unsigned long pgoff, unsigned long len)
1226{
1227 u64 maxsize = file_mmap_size_max(file, inode);
1228
1229 if (maxsize && len > maxsize)
1230 return false;
1231 maxsize -= len;
1232 if (pgoff > maxsize >> PAGE_SHIFT)
1233 return false;
1234 return true;
1235}
1236
1da177e4 1237/*
3e4e28c5 1238 * The caller must write-lock current->mm->mmap_lock.
1da177e4 1239 */
1fcfd8db 1240unsigned long do_mmap(struct file *file, unsigned long addr,
1da177e4 1241 unsigned long len, unsigned long prot,
45e55300
PC
1242 unsigned long flags, unsigned long pgoff,
1243 unsigned long *populate, struct list_head *uf)
1da177e4 1244{
cc71aba3 1245 struct mm_struct *mm = current->mm;
45e55300 1246 vm_flags_t vm_flags;
62b5f7d0 1247 int pkey = 0;
1da177e4 1248
524e00b3 1249 validate_mm(mm);
41badc15 1250 *populate = 0;
bebeb3d6 1251
e37609bb
PK
1252 if (!len)
1253 return -EINVAL;
1254
1da177e4
LT
1255 /*
1256 * Does the application expect PROT_READ to imply PROT_EXEC?
1257 *
1258 * (the exception is when the underlying filesystem is noexec
1259 * mounted, in which case we dont add PROT_EXEC.)
1260 */
1261 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
90f8572b 1262 if (!(file && path_noexec(&file->f_path)))
1da177e4
LT
1263 prot |= PROT_EXEC;
1264
a4ff8e86
MH
1265 /* force arch specific MAP_FIXED handling in get_unmapped_area */
1266 if (flags & MAP_FIXED_NOREPLACE)
1267 flags |= MAP_FIXED;
1268
7cd94146
EP
1269 if (!(flags & MAP_FIXED))
1270 addr = round_hint_to_min(addr);
1271
1da177e4
LT
1272 /* Careful about overflows.. */
1273 len = PAGE_ALIGN(len);
9206de95 1274 if (!len)
1da177e4
LT
1275 return -ENOMEM;
1276
1277 /* offset overflow? */
1278 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
cc71aba3 1279 return -EOVERFLOW;
1da177e4
LT
1280
1281 /* Too many mappings? */
1282 if (mm->map_count > sysctl_max_map_count)
1283 return -ENOMEM;
1284
1285 /* Obtain the address to map to. we verify (or select) it and ensure
1286 * that it represents a valid section of the address space.
1287 */
1288 addr = get_unmapped_area(file, addr, len, pgoff, flags);
ff68dac6 1289 if (IS_ERR_VALUE(addr))
1da177e4
LT
1290 return addr;
1291
a4ff8e86 1292 if (flags & MAP_FIXED_NOREPLACE) {
35e43c5f 1293 if (find_vma_intersection(mm, addr, addr + len))
a4ff8e86
MH
1294 return -EEXIST;
1295 }
1296
62b5f7d0
DH
1297 if (prot == PROT_EXEC) {
1298 pkey = execute_only_pkey(mm);
1299 if (pkey < 0)
1300 pkey = 0;
1301 }
1302
1da177e4
LT
1303 /* Do simple checking here so the lower-level routines won't have
1304 * to. we assume access permissions have been handled by the open
1305 * of the memory object, so we don't do any here.
1306 */
45e55300 1307 vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1da177e4
LT
1308 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1309
cdf7b341 1310 if (flags & MAP_LOCKED)
1da177e4
LT
1311 if (!can_do_mlock())
1312 return -EPERM;
ba470de4 1313
363ee17f
DB
1314 if (mlock_future_check(mm, vm_flags, len))
1315 return -EAGAIN;
1da177e4 1316
1da177e4 1317 if (file) {
077bf22b 1318 struct inode *inode = file_inode(file);
1c972597
DW
1319 unsigned long flags_mask;
1320
be83bbf8
LT
1321 if (!file_mmap_ok(file, inode, pgoff, len))
1322 return -EOVERFLOW;
1323
1c972597 1324 flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
077bf22b 1325
1da177e4
LT
1326 switch (flags & MAP_TYPE) {
1327 case MAP_SHARED:
1c972597
DW
1328 /*
1329 * Force use of MAP_SHARED_VALIDATE with non-legacy
1330 * flags. E.g. MAP_SYNC is dangerous to use with
1331 * MAP_SHARED as you don't know which consistency model
1332 * you will get. We silently ignore unsupported flags
1333 * with MAP_SHARED to preserve backward compatibility.
1334 */
1335 flags &= LEGACY_MAP_MASK;
e4a9bc58 1336 fallthrough;
1c972597
DW
1337 case MAP_SHARED_VALIDATE:
1338 if (flags & ~flags_mask)
1339 return -EOPNOTSUPP;
dc617f29
DW
1340 if (prot & PROT_WRITE) {
1341 if (!(file->f_mode & FMODE_WRITE))
1342 return -EACCES;
1343 if (IS_SWAPFILE(file->f_mapping->host))
1344 return -ETXTBSY;
1345 }
1da177e4
LT
1346
1347 /*
1348 * Make sure we don't allow writing to an append-only
1349 * file..
1350 */
1351 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1352 return -EACCES;
1353
1da177e4
LT
1354 vm_flags |= VM_SHARED | VM_MAYSHARE;
1355 if (!(file->f_mode & FMODE_WRITE))
1356 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
e4a9bc58 1357 fallthrough;
1da177e4
LT
1358 case MAP_PRIVATE:
1359 if (!(file->f_mode & FMODE_READ))
1360 return -EACCES;
90f8572b 1361 if (path_noexec(&file->f_path)) {
80c5606c
LT
1362 if (vm_flags & VM_EXEC)
1363 return -EPERM;
1364 vm_flags &= ~VM_MAYEXEC;
1365 }
80c5606c 1366
72c2d531 1367 if (!file->f_op->mmap)
80c5606c 1368 return -ENODEV;
b2c56e4f
ON
1369 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1370 return -EINVAL;
1da177e4
LT
1371 break;
1372
1373 default:
1374 return -EINVAL;
1375 }
1376 } else {
1377 switch (flags & MAP_TYPE) {
1378 case MAP_SHARED:
b2c56e4f
ON
1379 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1380 return -EINVAL;
ce363942
TH
1381 /*
1382 * Ignore pgoff.
1383 */
1384 pgoff = 0;
1da177e4
LT
1385 vm_flags |= VM_SHARED | VM_MAYSHARE;
1386 break;
1387 case MAP_PRIVATE:
1388 /*
1389 * Set pgoff according to addr for anon_vma.
1390 */
1391 pgoff = addr >> PAGE_SHIFT;
1392 break;
1393 default:
1394 return -EINVAL;
1395 }
1396 }
1397
c22c0d63
ML
1398 /*
1399 * Set 'VM_NORESERVE' if we should not account for the
1400 * memory use of this mapping.
1401 */
1402 if (flags & MAP_NORESERVE) {
1403 /* We honor MAP_NORESERVE if allowed to overcommit */
1404 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1405 vm_flags |= VM_NORESERVE;
1406
1407 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1408 if (file && is_file_hugepages(file))
1409 vm_flags |= VM_NORESERVE;
1410 }
1411
897ab3e0 1412 addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
09a9f1d2
ML
1413 if (!IS_ERR_VALUE(addr) &&
1414 ((vm_flags & VM_LOCKED) ||
1415 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
41badc15 1416 *populate = len;
bebeb3d6 1417 return addr;
0165ab44 1418}
6be5ceb0 1419
a90f590a
DB
1420unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1421 unsigned long prot, unsigned long flags,
1422 unsigned long fd, unsigned long pgoff)
66f0dc48
HD
1423{
1424 struct file *file = NULL;
1e3ee14b 1425 unsigned long retval;
66f0dc48
HD
1426
1427 if (!(flags & MAP_ANONYMOUS)) {
120a795d 1428 audit_mmap_fd(fd, flags);
66f0dc48
HD
1429 file = fget(fd);
1430 if (!file)
1e3ee14b 1431 return -EBADF;
7bba8f0e 1432 if (is_file_hugepages(file)) {
af73e4d9 1433 len = ALIGN(len, huge_page_size(hstate_file(file)));
7bba8f0e
ZL
1434 } else if (unlikely(flags & MAP_HUGETLB)) {
1435 retval = -EINVAL;
493af578 1436 goto out_fput;
7bba8f0e 1437 }
66f0dc48 1438 } else if (flags & MAP_HUGETLB) {
c103a4dc 1439 struct hstate *hs;
af73e4d9 1440
20ac2893 1441 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
091d0d55
LZ
1442 if (!hs)
1443 return -EINVAL;
1444
1445 len = ALIGN(len, huge_page_size(hs));
66f0dc48
HD
1446 /*
1447 * VM_NORESERVE is used because the reservations will be
1448 * taken when vm_ops->mmap() is called
66f0dc48 1449 */
af73e4d9 1450 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
42d7395f 1451 VM_NORESERVE,
83c1fd76 1452 HUGETLB_ANONHUGE_INODE,
42d7395f 1453 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
66f0dc48
HD
1454 if (IS_ERR(file))
1455 return PTR_ERR(file);
1456 }
1457
9fbeb5ab 1458 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
493af578 1459out_fput:
66f0dc48
HD
1460 if (file)
1461 fput(file);
66f0dc48
HD
1462 return retval;
1463}
1464
a90f590a
DB
1465SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1466 unsigned long, prot, unsigned long, flags,
1467 unsigned long, fd, unsigned long, pgoff)
1468{
1469 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1470}
1471
a4679373
CH
1472#ifdef __ARCH_WANT_SYS_OLD_MMAP
1473struct mmap_arg_struct {
1474 unsigned long addr;
1475 unsigned long len;
1476 unsigned long prot;
1477 unsigned long flags;
1478 unsigned long fd;
1479 unsigned long offset;
1480};
1481
1482SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1483{
1484 struct mmap_arg_struct a;
1485
1486 if (copy_from_user(&a, arg, sizeof(a)))
1487 return -EFAULT;
de1741a1 1488 if (offset_in_page(a.offset))
a4679373
CH
1489 return -EINVAL;
1490
a90f590a
DB
1491 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1492 a.offset >> PAGE_SHIFT);
a4679373
CH
1493}
1494#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1495
4e950f6f 1496/*
8bb4e7a2 1497 * Some shared mappings will want the pages marked read-only
4e950f6f
AD
1498 * to track write events. If so, we'll downgrade vm_page_prot
1499 * to the private version (using protection_map[] without the
1500 * VM_SHARED bit).
1501 */
6d2329f8 1502int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
4e950f6f 1503{
ca16d140 1504 vm_flags_t vm_flags = vma->vm_flags;
8a04446a 1505 const struct vm_operations_struct *vm_ops = vma->vm_ops;
4e950f6f
AD
1506
1507 /* If it was private or non-writable, the write bit is already clear */
1508 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1509 return 0;
1510
1511 /* The backer wishes to know when pages are first written to? */
8a04446a 1512 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
4e950f6f
AD
1513 return 1;
1514
64e45507
PF
1515 /* The open routine did something to the protections that pgprot_modify
1516 * won't preserve? */
6d2329f8
AA
1517 if (pgprot_val(vm_page_prot) !=
1518 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
4e950f6f
AD
1519 return 0;
1520
f96f7a40
DH
1521 /*
1522 * Do we need to track softdirty? hugetlb does not support softdirty
1523 * tracking yet.
1524 */
1525 if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma))
64e45507
PF
1526 return 1;
1527
4e950f6f 1528 /* Specialty mapping? */
4b6e1e37 1529 if (vm_flags & VM_PFNMAP)
4e950f6f
AD
1530 return 0;
1531
1532 /* Can the mapping track the dirty pages? */
1533 return vma->vm_file && vma->vm_file->f_mapping &&
f56753ac 1534 mapping_can_writeback(vma->vm_file->f_mapping);
4e950f6f
AD
1535}
1536
fc8744ad
LT
1537/*
1538 * We account for memory if it's a private writeable mapping,
5a6fe125 1539 * not hugepages and VM_NORESERVE wasn't set.
fc8744ad 1540 */
ca16d140 1541static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
fc8744ad 1542{
5a6fe125
MG
1543 /*
1544 * hugetlb has its own accounting separate from the core VM
1545 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1546 */
1547 if (file && is_file_hugepages(file))
1548 return 0;
1549
fc8744ad
LT
1550 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1551}
1552
3499a131
LH
1553/**
1554 * unmapped_area() - Find an area between the low_limit and the high_limit with
1555 * the correct alignment and offset, all from @info. Note: current->mm is used
1556 * for the search.
1557 *
1558 * @info: The unmapped area information including the range (low_limit -
1559 * hight_limit), the alignment offset and mask.
1560 *
1561 * Return: A memory address or -ENOMEM.
1562 */
baceaf1c 1563static unsigned long unmapped_area(struct vm_unmapped_area_info *info)
db4fbfb9 1564{
3499a131 1565 unsigned long length, gap;
db4fbfb9 1566
3499a131 1567 MA_STATE(mas, &current->mm->mm_mt, 0, 0);
db4fbfb9
ML
1568
1569 /* Adjust search length to account for worst case alignment overhead */
1570 length = info->length + info->align_mask;
1571 if (length < info->length)
1572 return -ENOMEM;
1573
3499a131
LH
1574 if (mas_empty_area(&mas, info->low_limit, info->high_limit - 1,
1575 length))
db4fbfb9
ML
1576 return -ENOMEM;
1577
3499a131
LH
1578 gap = mas.index;
1579 gap += (info->align_offset - gap) & info->align_mask;
1580 return gap;
db4fbfb9
ML
1581}
1582
3499a131
LH
1583/**
1584 * unmapped_area_topdown() - Find an area between the low_limit and the
1585 * high_limit with * the correct alignment and offset at the highest available
1586 * address, all from @info. Note: current->mm is used for the search.
1587 *
1588 * @info: The unmapped area information including the range (low_limit -
1589 * hight_limit), the alignment offset and mask.
1590 *
1591 * Return: A memory address or -ENOMEM.
1592 */
baceaf1c 1593static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
db4fbfb9 1594{
3499a131 1595 unsigned long length, gap;
db4fbfb9 1596
3499a131 1597 MA_STATE(mas, &current->mm->mm_mt, 0, 0);
db4fbfb9
ML
1598 /* Adjust search length to account for worst case alignment overhead */
1599 length = info->length + info->align_mask;
1600 if (length < info->length)
1601 return -ENOMEM;
1602
3499a131
LH
1603 if (mas_empty_area_rev(&mas, info->low_limit, info->high_limit - 1,
1604 length))
db4fbfb9 1605 return -ENOMEM;
db4fbfb9 1606
3499a131
LH
1607 gap = mas.last + 1 - info->length;
1608 gap -= (gap - info->align_offset) & info->align_mask;
1609 return gap;
db4fbfb9
ML
1610}
1611
baceaf1c
JK
1612/*
1613 * Search for an unmapped address range.
1614 *
1615 * We are looking for a range that:
1616 * - does not intersect with any VMA;
1617 * - is contained within the [low_limit, high_limit) interval;
1618 * - is at least the desired size.
1619 * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
1620 */
1621unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info)
1622{
df529cab
JK
1623 unsigned long addr;
1624
baceaf1c 1625 if (info->flags & VM_UNMAPPED_AREA_TOPDOWN)
df529cab 1626 addr = unmapped_area_topdown(info);
baceaf1c 1627 else
df529cab
JK
1628 addr = unmapped_area(info);
1629
1630 trace_vm_unmapped_area(addr, info);
1631 return addr;
baceaf1c 1632}
f6795053 1633
1da177e4
LT
1634/* Get an address range which is currently unmapped.
1635 * For shmat() with addr=0.
1636 *
1637 * Ugly calling convention alert:
1638 * Return value with the low bits set means error value,
1639 * ie
1640 * if (ret & ~PAGE_MASK)
1641 * error = ret;
1642 *
1643 * This function "knows" that -ENOMEM has the bits set.
1644 */
1da177e4 1645unsigned long
4b439e25
CL
1646generic_get_unmapped_area(struct file *filp, unsigned long addr,
1647 unsigned long len, unsigned long pgoff,
1648 unsigned long flags)
1da177e4
LT
1649{
1650 struct mm_struct *mm = current->mm;
1be7107f 1651 struct vm_area_struct *vma, *prev;
db4fbfb9 1652 struct vm_unmapped_area_info info;
2cb4de08 1653 const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
1da177e4 1654
f6795053 1655 if (len > mmap_end - mmap_min_addr)
1da177e4
LT
1656 return -ENOMEM;
1657
06abdfb4
BH
1658 if (flags & MAP_FIXED)
1659 return addr;
1660
1da177e4
LT
1661 if (addr) {
1662 addr = PAGE_ALIGN(addr);
1be7107f 1663 vma = find_vma_prev(mm, addr, &prev);
f6795053 1664 if (mmap_end - len >= addr && addr >= mmap_min_addr &&
1be7107f
HD
1665 (!vma || addr + len <= vm_start_gap(vma)) &&
1666 (!prev || addr >= vm_end_gap(prev)))
1da177e4
LT
1667 return addr;
1668 }
1da177e4 1669
db4fbfb9
ML
1670 info.flags = 0;
1671 info.length = len;
4e99b021 1672 info.low_limit = mm->mmap_base;
f6795053 1673 info.high_limit = mmap_end;
db4fbfb9 1674 info.align_mask = 0;
09ef5283 1675 info.align_offset = 0;
db4fbfb9 1676 return vm_unmapped_area(&info);
1da177e4 1677}
4b439e25
CL
1678
1679#ifndef HAVE_ARCH_UNMAPPED_AREA
1680unsigned long
1681arch_get_unmapped_area(struct file *filp, unsigned long addr,
1682 unsigned long len, unsigned long pgoff,
1683 unsigned long flags)
1684{
1685 return generic_get_unmapped_area(filp, addr, len, pgoff, flags);
1686}
cc71aba3 1687#endif
1da177e4 1688
1da177e4
LT
1689/*
1690 * This mmap-allocator allocates new areas top-down from below the
1691 * stack's low limit (the base):
1692 */
1da177e4 1693unsigned long
4b439e25
CL
1694generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
1695 unsigned long len, unsigned long pgoff,
1696 unsigned long flags)
1da177e4 1697{
1be7107f 1698 struct vm_area_struct *vma, *prev;
1da177e4 1699 struct mm_struct *mm = current->mm;
db4fbfb9 1700 struct vm_unmapped_area_info info;
2cb4de08 1701 const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
1da177e4
LT
1702
1703 /* requested length too big for entire address space */
f6795053 1704 if (len > mmap_end - mmap_min_addr)
1da177e4
LT
1705 return -ENOMEM;
1706
06abdfb4
BH
1707 if (flags & MAP_FIXED)
1708 return addr;
1709
1da177e4
LT
1710 /* requesting a specific address */
1711 if (addr) {
1712 addr = PAGE_ALIGN(addr);
1be7107f 1713 vma = find_vma_prev(mm, addr, &prev);
f6795053 1714 if (mmap_end - len >= addr && addr >= mmap_min_addr &&
1be7107f
HD
1715 (!vma || addr + len <= vm_start_gap(vma)) &&
1716 (!prev || addr >= vm_end_gap(prev)))
1da177e4
LT
1717 return addr;
1718 }
1719
db4fbfb9
ML
1720 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1721 info.length = len;
2afc745f 1722 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
f6795053 1723 info.high_limit = arch_get_mmap_base(addr, mm->mmap_base);
db4fbfb9 1724 info.align_mask = 0;
09ef5283 1725 info.align_offset = 0;
db4fbfb9 1726 addr = vm_unmapped_area(&info);
b716ad95 1727
1da177e4
LT
1728 /*
1729 * A failed mmap() very likely causes application failure,
1730 * so fall back to the bottom-up function here. This scenario
1731 * can happen with large stack limits and large mmap()
1732 * allocations.
1733 */
de1741a1 1734 if (offset_in_page(addr)) {
db4fbfb9
ML
1735 VM_BUG_ON(addr != -ENOMEM);
1736 info.flags = 0;
1737 info.low_limit = TASK_UNMAPPED_BASE;
f6795053 1738 info.high_limit = mmap_end;
db4fbfb9
ML
1739 addr = vm_unmapped_area(&info);
1740 }
1da177e4
LT
1741
1742 return addr;
1743}
4b439e25
CL
1744
1745#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1746unsigned long
1747arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
1748 unsigned long len, unsigned long pgoff,
1749 unsigned long flags)
1750{
1751 return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags);
1752}
1da177e4
LT
1753#endif
1754
1da177e4
LT
1755unsigned long
1756get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1757 unsigned long pgoff, unsigned long flags)
1758{
06abdfb4
BH
1759 unsigned long (*get_area)(struct file *, unsigned long,
1760 unsigned long, unsigned long, unsigned long);
1761
9206de95
AV
1762 unsigned long error = arch_mmap_check(addr, len, flags);
1763 if (error)
1764 return error;
1765
1766 /* Careful about overflows.. */
1767 if (len > TASK_SIZE)
1768 return -ENOMEM;
1769
06abdfb4 1770 get_area = current->mm->get_unmapped_area;
c01d5b30
HD
1771 if (file) {
1772 if (file->f_op->get_unmapped_area)
1773 get_area = file->f_op->get_unmapped_area;
1774 } else if (flags & MAP_SHARED) {
1775 /*
1776 * mmap_region() will call shmem_zero_setup() to create a file,
1777 * so use shmem's get_unmapped_area in case it can be huge.
45e55300 1778 * do_mmap() will clear pgoff, so match alignment.
c01d5b30
HD
1779 */
1780 pgoff = 0;
1781 get_area = shmem_get_unmapped_area;
f35b5d7d
RR
1782 } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1783 /* Ensures that larger anonymous mappings are THP aligned. */
1784 get_area = thp_get_unmapped_area;
c01d5b30
HD
1785 }
1786
06abdfb4
BH
1787 addr = get_area(file, addr, len, pgoff, flags);
1788 if (IS_ERR_VALUE(addr))
1789 return addr;
1da177e4 1790
07ab67c8
LT
1791 if (addr > TASK_SIZE - len)
1792 return -ENOMEM;
de1741a1 1793 if (offset_in_page(addr))
07ab67c8 1794 return -EINVAL;
06abdfb4 1795
9ac4ed4b
AV
1796 error = security_mmap_addr(addr);
1797 return error ? error : addr;
1da177e4
LT
1798}
1799
1800EXPORT_SYMBOL(get_unmapped_area);
1801
abdba2dd
LH
1802/**
1803 * find_vma_intersection() - Look up the first VMA which intersects the interval
1804 * @mm: The process address space.
1805 * @start_addr: The inclusive start user address.
1806 * @end_addr: The exclusive end user address.
1807 *
1808 * Returns: The first VMA within the provided range, %NULL otherwise. Assumes
1809 * start_addr < end_addr.
1810 */
1811struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
1812 unsigned long start_addr,
1813 unsigned long end_addr)
1814{
abdba2dd
LH
1815 unsigned long index = start_addr;
1816
1817 mmap_assert_locked(mm);
7964cf8c 1818 return mt_find(&mm->mm_mt, &index, end_addr - 1);
abdba2dd
LH
1819}
1820EXPORT_SYMBOL(find_vma_intersection);
1821
be8432e7
LH
1822/**
1823 * find_vma() - Find the VMA for a given address, or the next VMA.
1824 * @mm: The mm_struct to check
1825 * @addr: The address
1826 *
1827 * Returns: The VMA associated with addr, or the next VMA.
1828 * May return %NULL in the case of no VMA at addr or above.
1829 */
48aae425 1830struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1da177e4 1831{
be8432e7 1832 unsigned long index = addr;
1da177e4 1833
5b78ed24 1834 mmap_assert_locked(mm);
7964cf8c 1835 return mt_find(&mm->mm_mt, &index, ULONG_MAX);
1da177e4 1836}
1da177e4
LT
1837EXPORT_SYMBOL(find_vma);
1838
7fdbd37d
LH
1839/**
1840 * find_vma_prev() - Find the VMA for a given address, or the next vma and
1841 * set %pprev to the previous VMA, if any.
1842 * @mm: The mm_struct to check
1843 * @addr: The address
1844 * @pprev: The pointer to set to the previous VMA
1845 *
1846 * Note that RCU lock is missing here since the external mmap_lock() is used
1847 * instead.
1848 *
1849 * Returns: The VMA associated with @addr, or the next vma.
1850 * May return %NULL in the case of no vma at addr or above.
6bd4837d 1851 */
1da177e4
LT
1852struct vm_area_struct *
1853find_vma_prev(struct mm_struct *mm, unsigned long addr,
1854 struct vm_area_struct **pprev)
1855{
6bd4837d 1856 struct vm_area_struct *vma;
7fdbd37d 1857 MA_STATE(mas, &mm->mm_mt, addr, addr);
1da177e4 1858
7fdbd37d
LH
1859 vma = mas_walk(&mas);
1860 *pprev = mas_prev(&mas, 0);
1861 if (!vma)
1862 vma = mas_next(&mas, ULONG_MAX);
6bd4837d 1863 return vma;
1da177e4
LT
1864}
1865
1866/*
1867 * Verify that the stack growth is acceptable and
1868 * update accounting. This is shared with both the
1869 * grow-up and grow-down cases.
1870 */
1be7107f
HD
1871static int acct_stack_growth(struct vm_area_struct *vma,
1872 unsigned long size, unsigned long grow)
1da177e4
LT
1873{
1874 struct mm_struct *mm = vma->vm_mm;
1be7107f 1875 unsigned long new_start;
1da177e4
LT
1876
1877 /* address space limit tests */
84638335 1878 if (!may_expand_vm(mm, vma->vm_flags, grow))
1da177e4
LT
1879 return -ENOMEM;
1880
1881 /* Stack limit test */
24c79d8e 1882 if (size > rlimit(RLIMIT_STACK))
1da177e4
LT
1883 return -ENOMEM;
1884
1885 /* mlock limit tests */
c5d8a364
ML
1886 if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT))
1887 return -ENOMEM;
1da177e4 1888
0d59a01b
AL
1889 /* Check to ensure the stack will not grow into a hugetlb-only region */
1890 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1891 vma->vm_end - size;
1892 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1893 return -EFAULT;
1894
1da177e4
LT
1895 /*
1896 * Overcommit.. This must be the final test, as it will
1897 * update security statistics.
1898 */
05fa199d 1899 if (security_vm_enough_memory_mm(mm, grow))
1da177e4
LT
1900 return -ENOMEM;
1901
1da177e4
LT
1902 return 0;
1903}
1904
46dea3d0 1905#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1da177e4 1906/*
46dea3d0
HD
1907 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1908 * vma is the last one with address > vma->vm_end. Have to extend vma.
1da177e4 1909 */
46dea3d0 1910int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1da177e4 1911{
09357814 1912 struct mm_struct *mm = vma->vm_mm;
1be7107f
HD
1913 struct vm_area_struct *next;
1914 unsigned long gap_addr;
12352d3c 1915 int error = 0;
d4af56c5 1916 MA_STATE(mas, &mm->mm_mt, 0, 0);
1da177e4
LT
1917
1918 if (!(vma->vm_flags & VM_GROWSUP))
1919 return -EFAULT;
1920
bd726c90 1921 /* Guard against exceeding limits of the address space. */
1be7107f 1922 address &= PAGE_MASK;
37511fb5 1923 if (address >= (TASK_SIZE & PAGE_MASK))
12352d3c 1924 return -ENOMEM;
bd726c90 1925 address += PAGE_SIZE;
12352d3c 1926
1be7107f
HD
1927 /* Enforce stack_guard_gap */
1928 gap_addr = address + stack_guard_gap;
bd726c90
HD
1929
1930 /* Guard against overflow */
1931 if (gap_addr < address || gap_addr > TASK_SIZE)
1932 gap_addr = TASK_SIZE;
1933
763ecb03
LH
1934 next = find_vma_intersection(mm, vma->vm_end, gap_addr);
1935 if (next && vma_is_accessible(next)) {
1be7107f
HD
1936 if (!(next->vm_flags & VM_GROWSUP))
1937 return -ENOMEM;
1938 /* Check that both stack segments have the same anon_vma? */
1939 }
1940
d4af56c5
LH
1941 if (mas_preallocate(&mas, vma, GFP_KERNEL))
1942 return -ENOMEM;
1943
12352d3c 1944 /* We must make sure the anon_vma is allocated. */
d4af56c5
LH
1945 if (unlikely(anon_vma_prepare(vma))) {
1946 mas_destroy(&mas);
1da177e4 1947 return -ENOMEM;
d4af56c5 1948 }
1da177e4
LT
1949
1950 /*
1951 * vma->vm_start/vm_end cannot change under us because the caller
c1e8d7c6 1952 * is required to hold the mmap_lock in read mode. We need the
1da177e4
LT
1953 * anon_vma lock to serialize against concurrent expand_stacks.
1954 */
12352d3c 1955 anon_vma_lock_write(vma->anon_vma);
1da177e4
LT
1956
1957 /* Somebody else might have raced and expanded it already */
1958 if (address > vma->vm_end) {
1959 unsigned long size, grow;
1960
1961 size = address - vma->vm_start;
1962 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1963
42c36f63
HD
1964 error = -ENOMEM;
1965 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1966 error = acct_stack_growth(vma, size, grow);
1967 if (!error) {
4128997b 1968 /*
524e00b3
LH
1969 * We only hold a shared mmap_lock lock here, so
1970 * we need to protect against concurrent vma
1971 * expansions. anon_vma_lock_write() doesn't
1972 * help here, as we don't guarantee that all
1973 * growable vmas in a mm share the same root
1974 * anon vma. So, we reuse mm->page_table_lock
1975 * to guard against concurrent vma expansions.
4128997b 1976 */
09357814 1977 spin_lock(&mm->page_table_lock);
87e8827b 1978 if (vma->vm_flags & VM_LOCKED)
09357814 1979 mm->locked_vm += grow;
84638335 1980 vm_stat_account(mm, vma->vm_flags, grow);
bf181b9f 1981 anon_vma_interval_tree_pre_update_vma(vma);
42c36f63 1982 vma->vm_end = address;
d4af56c5
LH
1983 /* Overwrite old entry in mtree. */
1984 vma_mas_store(vma, &mas);
bf181b9f 1985 anon_vma_interval_tree_post_update_vma(vma);
09357814 1986 spin_unlock(&mm->page_table_lock);
4128997b 1987
42c36f63
HD
1988 perf_event_mmap(vma);
1989 }
3af9e859 1990 }
1da177e4 1991 }
12352d3c 1992 anon_vma_unlock_write(vma->anon_vma);
c791576c 1993 khugepaged_enter_vma(vma, vma->vm_flags);
d4af56c5 1994 mas_destroy(&mas);
1da177e4
LT
1995 return error;
1996}
46dea3d0
HD
1997#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1998
1da177e4
LT
1999/*
2000 * vma is the first one with address < vma->vm_start. Have to extend vma.
2001 */
524e00b3 2002int expand_downwards(struct vm_area_struct *vma, unsigned long address)
1da177e4 2003{
09357814 2004 struct mm_struct *mm = vma->vm_mm;
763ecb03 2005 MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start);
1be7107f 2006 struct vm_area_struct *prev;
0a1d5299 2007 int error = 0;
1da177e4 2008
8869477a 2009 address &= PAGE_MASK;
0a1d5299
JH
2010 if (address < mmap_min_addr)
2011 return -EPERM;
8869477a 2012
1be7107f 2013 /* Enforce stack_guard_gap */
763ecb03 2014 prev = mas_prev(&mas, 0);
32e4e6d5
ON
2015 /* Check that both stack segments have the same anon_vma? */
2016 if (prev && !(prev->vm_flags & VM_GROWSDOWN) &&
3122e80e 2017 vma_is_accessible(prev)) {
32e4e6d5 2018 if (address - prev->vm_end < stack_guard_gap)
1be7107f 2019 return -ENOMEM;
1be7107f
HD
2020 }
2021
d4af56c5
LH
2022 if (mas_preallocate(&mas, vma, GFP_KERNEL))
2023 return -ENOMEM;
2024
12352d3c 2025 /* We must make sure the anon_vma is allocated. */
d4af56c5
LH
2026 if (unlikely(anon_vma_prepare(vma))) {
2027 mas_destroy(&mas);
12352d3c 2028 return -ENOMEM;
d4af56c5 2029 }
1da177e4
LT
2030
2031 /*
2032 * vma->vm_start/vm_end cannot change under us because the caller
c1e8d7c6 2033 * is required to hold the mmap_lock in read mode. We need the
1da177e4
LT
2034 * anon_vma lock to serialize against concurrent expand_stacks.
2035 */
12352d3c 2036 anon_vma_lock_write(vma->anon_vma);
1da177e4
LT
2037
2038 /* Somebody else might have raced and expanded it already */
2039 if (address < vma->vm_start) {
2040 unsigned long size, grow;
2041
2042 size = vma->vm_end - address;
2043 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2044
a626ca6a
LT
2045 error = -ENOMEM;
2046 if (grow <= vma->vm_pgoff) {
2047 error = acct_stack_growth(vma, size, grow);
2048 if (!error) {
4128997b 2049 /*
524e00b3
LH
2050 * We only hold a shared mmap_lock lock here, so
2051 * we need to protect against concurrent vma
2052 * expansions. anon_vma_lock_write() doesn't
2053 * help here, as we don't guarantee that all
2054 * growable vmas in a mm share the same root
2055 * anon vma. So, we reuse mm->page_table_lock
2056 * to guard against concurrent vma expansions.
4128997b 2057 */
09357814 2058 spin_lock(&mm->page_table_lock);
87e8827b 2059 if (vma->vm_flags & VM_LOCKED)
09357814 2060 mm->locked_vm += grow;
84638335 2061 vm_stat_account(mm, vma->vm_flags, grow);
bf181b9f 2062 anon_vma_interval_tree_pre_update_vma(vma);
a626ca6a
LT
2063 vma->vm_start = address;
2064 vma->vm_pgoff -= grow;
d4af56c5
LH
2065 /* Overwrite old entry in mtree. */
2066 vma_mas_store(vma, &mas);
bf181b9f 2067 anon_vma_interval_tree_post_update_vma(vma);
09357814 2068 spin_unlock(&mm->page_table_lock);
4128997b 2069
a626ca6a
LT
2070 perf_event_mmap(vma);
2071 }
1da177e4
LT
2072 }
2073 }
12352d3c 2074 anon_vma_unlock_write(vma->anon_vma);
c791576c 2075 khugepaged_enter_vma(vma, vma->vm_flags);
d4af56c5 2076 mas_destroy(&mas);
1da177e4
LT
2077 return error;
2078}
2079
1be7107f
HD
2080/* enforced gap between the expanding stack and other mappings. */
2081unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2082
2083static int __init cmdline_parse_stack_guard_gap(char *p)
2084{
2085 unsigned long val;
2086 char *endptr;
2087
2088 val = simple_strtoul(p, &endptr, 10);
2089 if (!*endptr)
2090 stack_guard_gap = val << PAGE_SHIFT;
2091
e6d09493 2092 return 1;
1be7107f
HD
2093}
2094__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2095
b6a2fea3
OW
2096#ifdef CONFIG_STACK_GROWSUP
2097int expand_stack(struct vm_area_struct *vma, unsigned long address)
2098{
2099 return expand_upwards(vma, address);
2100}
2101
2102struct vm_area_struct *
2103find_extend_vma(struct mm_struct *mm, unsigned long addr)
2104{
2105 struct vm_area_struct *vma, *prev;
2106
2107 addr &= PAGE_MASK;
2108 vma = find_vma_prev(mm, addr, &prev);
2109 if (vma && (vma->vm_start <= addr))
2110 return vma;
4d45e75a 2111 if (!prev || expand_stack(prev, addr))
b6a2fea3 2112 return NULL;
cea10a19 2113 if (prev->vm_flags & VM_LOCKED)
fc05f566 2114 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
b6a2fea3
OW
2115 return prev;
2116}
2117#else
2118int expand_stack(struct vm_area_struct *vma, unsigned long address)
2119{
2120 return expand_downwards(vma, address);
2121}
2122
1da177e4 2123struct vm_area_struct *
cc71aba3 2124find_extend_vma(struct mm_struct *mm, unsigned long addr)
1da177e4 2125{
cc71aba3 2126 struct vm_area_struct *vma;
1da177e4
LT
2127 unsigned long start;
2128
2129 addr &= PAGE_MASK;
cc71aba3 2130 vma = find_vma(mm, addr);
1da177e4
LT
2131 if (!vma)
2132 return NULL;
2133 if (vma->vm_start <= addr)
2134 return vma;
2135 if (!(vma->vm_flags & VM_GROWSDOWN))
2136 return NULL;
2137 start = vma->vm_start;
2138 if (expand_stack(vma, addr))
2139 return NULL;
cea10a19 2140 if (vma->vm_flags & VM_LOCKED)
fc05f566 2141 populate_vma_page_range(vma, addr, start, NULL);
1da177e4
LT
2142 return vma;
2143}
2144#endif
2145
e1d6d01a
JB
2146EXPORT_SYMBOL_GPL(find_extend_vma);
2147
1da177e4 2148/*
763ecb03
LH
2149 * Ok - we have the memory areas we should free on a maple tree so release them,
2150 * and do the vma updates.
2c0b3814
HD
2151 *
2152 * Called with the mm semaphore held.
1da177e4 2153 */
763ecb03 2154static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas)
1da177e4 2155{
4f74d2c8 2156 unsigned long nr_accounted = 0;
763ecb03 2157 struct vm_area_struct *vma;
4f74d2c8 2158
365e9c87
HD
2159 /* Update high watermark before we lower total_vm */
2160 update_hiwater_vm(mm);
763ecb03 2161 mas_for_each(mas, vma, ULONG_MAX) {
2c0b3814
HD
2162 long nrpages = vma_pages(vma);
2163
4f74d2c8
LT
2164 if (vma->vm_flags & VM_ACCOUNT)
2165 nr_accounted += nrpages;
84638335 2166 vm_stat_account(mm, vma->vm_flags, -nrpages);
763ecb03
LH
2167 remove_vma(vma);
2168 }
4f74d2c8 2169 vm_unacct_memory(nr_accounted);
1da177e4
LT
2170 validate_mm(mm);
2171}
2172
2173/*
2174 * Get rid of page table information in the indicated region.
2175 *
f10df686 2176 * Called with the mm semaphore held.
1da177e4 2177 */
763ecb03 2178static void unmap_region(struct mm_struct *mm, struct maple_tree *mt,
e0da382c 2179 struct vm_area_struct *vma, struct vm_area_struct *prev,
763ecb03 2180 struct vm_area_struct *next,
e0da382c 2181 unsigned long start, unsigned long end)
1da177e4 2182{
d16dfc55 2183 struct mmu_gather tlb;
1da177e4
LT
2184
2185 lru_add_drain();
a72afd87 2186 tlb_gather_mmu(&tlb, mm);
365e9c87 2187 update_hiwater_rss(mm);
763ecb03
LH
2188 unmap_vmas(&tlb, mt, vma, start, end);
2189 free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
6ee8630e 2190 next ? next->vm_start : USER_PGTABLES_CEILING);
ae8eba8b 2191 tlb_finish_mmu(&tlb);
1da177e4
LT
2192}
2193
1da177e4 2194/*
def5efe0
DR
2195 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2196 * has already been checked or doesn't make sense to fail.
1da177e4 2197 */
def5efe0
DR
2198int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2199 unsigned long addr, int new_below)
1da177e4 2200{
1da177e4 2201 struct vm_area_struct *new;
e3975891 2202 int err;
d4af56c5 2203 validate_mm_mt(mm);
1da177e4 2204
dd3b614f
DS
2205 if (vma->vm_ops && vma->vm_ops->may_split) {
2206 err = vma->vm_ops->may_split(vma, addr);
31383c68
DW
2207 if (err)
2208 return err;
2209 }
1da177e4 2210
3928d4f5 2211 new = vm_area_dup(vma);
1da177e4 2212 if (!new)
e3975891 2213 return -ENOMEM;
1da177e4 2214
1da177e4
LT
2215 if (new_below)
2216 new->vm_end = addr;
2217 else {
2218 new->vm_start = addr;
2219 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2220 }
2221
ef0855d3
ON
2222 err = vma_dup_policy(vma, new);
2223 if (err)
5beb4930 2224 goto out_free_vma;
1da177e4 2225
c4ea95d7
DF
2226 err = anon_vma_clone(new, vma);
2227 if (err)
5beb4930
RR
2228 goto out_free_mpol;
2229
e9714acf 2230 if (new->vm_file)
1da177e4
LT
2231 get_file(new->vm_file);
2232
2233 if (new->vm_ops && new->vm_ops->open)
2234 new->vm_ops->open(new);
2235
2236 if (new_below)
5beb4930 2237 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1da177e4
LT
2238 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2239 else
5beb4930 2240 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1da177e4 2241
5beb4930
RR
2242 /* Success. */
2243 if (!err)
2244 return 0;
2245
d4af56c5
LH
2246 /* Avoid vm accounting in close() operation */
2247 new->vm_start = new->vm_end;
2248 new->vm_pgoff = 0;
5beb4930 2249 /* Clean everything up if vma_adjust failed. */
58927533
RR
2250 if (new->vm_ops && new->vm_ops->close)
2251 new->vm_ops->close(new);
e9714acf 2252 if (new->vm_file)
5beb4930 2253 fput(new->vm_file);
2aeadc30 2254 unlink_anon_vmas(new);
5beb4930 2255 out_free_mpol:
ef0855d3 2256 mpol_put(vma_policy(new));
5beb4930 2257 out_free_vma:
3928d4f5 2258 vm_area_free(new);
d4af56c5 2259 validate_mm_mt(mm);
5beb4930 2260 return err;
1da177e4
LT
2261}
2262
659ace58
KM
2263/*
2264 * Split a vma into two pieces at address 'addr', a new vma is allocated
2265 * either for the first part or the tail.
2266 */
2267int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2268 unsigned long addr, int new_below)
2269{
2270 if (mm->map_count >= sysctl_max_map_count)
2271 return -ENOMEM;
2272
2273 return __split_vma(mm, vma, addr, new_below);
2274}
2275
763ecb03
LH
2276static inline int munmap_sidetree(struct vm_area_struct *vma,
2277 struct ma_state *mas_detach)
1da177e4 2278{
763ecb03
LH
2279 mas_set_range(mas_detach, vma->vm_start, vma->vm_end - 1);
2280 if (mas_store_gfp(mas_detach, vma, GFP_KERNEL))
2281 return -ENOMEM;
1da177e4 2282
763ecb03
LH
2283 if (vma->vm_flags & VM_LOCKED)
2284 vma->vm_mm->locked_vm -= vma_pages(vma);
1da177e4 2285
763ecb03 2286 return 0;
11f9a21a 2287}
5a28fc94 2288
11f9a21a
LH
2289/*
2290 * do_mas_align_munmap() - munmap the aligned region from @start to @end.
2291 * @mas: The maple_state, ideally set up to alter the correct tree location.
2292 * @vma: The starting vm_area_struct
2293 * @mm: The mm_struct
2294 * @start: The aligned start address to munmap.
2295 * @end: The aligned end address to munmap.
2296 * @uf: The userfaultfd list_head
2297 * @downgrade: Set to true to attempt a write downgrade of the mmap_sem
2298 *
2299 * If @downgrade is true, check return code for potential release of the lock.
2300 */
2301static int
2302do_mas_align_munmap(struct ma_state *mas, struct vm_area_struct *vma,
2303 struct mm_struct *mm, unsigned long start,
2304 unsigned long end, struct list_head *uf, bool downgrade)
2305{
763ecb03
LH
2306 struct vm_area_struct *prev, *next = NULL;
2307 struct maple_tree mt_detach;
2308 int count = 0;
11f9a21a 2309 int error = -ENOMEM;
763ecb03
LH
2310 MA_STATE(mas_detach, &mt_detach, 0, 0);
2311 mt_init_flags(&mt_detach, MT_FLAGS_LOCK_EXTERN);
2312 mt_set_external_lock(&mt_detach, &mm->mmap_lock);
d4af56c5 2313
11f9a21a 2314 if (mas_preallocate(mas, vma, GFP_KERNEL))
d4af56c5 2315 return -ENOMEM;
524e00b3 2316
11f9a21a 2317 mas->last = end - 1;
1da177e4
LT
2318 /*
2319 * If we need to split any vma, do it now to save pain later.
2320 *
2321 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2322 * unmapped vm_area_struct will remain in use: so lower split_vma
2323 * places tmp vma above, and higher split_vma places tmp vma below.
2324 */
763ecb03
LH
2325
2326 /* Does it split the first one? */
146425a3 2327 if (start > vma->vm_start) {
659ace58
KM
2328
2329 /*
2330 * Make sure that map_count on return from munmap() will
2331 * not exceed its limit; but let map_count go just above
2332 * its limit temporarily, to help free resources as expected.
2333 */
2334 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
d4af56c5 2335 goto map_count_exceeded;
659ace58 2336
763ecb03
LH
2337 /*
2338 * mas_pause() is not needed since mas->index needs to be set
2339 * differently than vma->vm_end anyways.
2340 */
659ace58 2341 error = __split_vma(mm, vma, start, 0);
1da177e4 2342 if (error)
763ecb03 2343 goto start_split_failed;
11f9a21a 2344
763ecb03
LH
2345 mas_set(mas, start);
2346 vma = mas_walk(mas);
1da177e4
LT
2347 }
2348
763ecb03
LH
2349 prev = mas_prev(mas, 0);
2350 if (unlikely((!prev)))
2351 mas_set(mas, start);
2352
2353 /*
2354 * Detach a range of VMAs from the mm. Using next as a temp variable as
2355 * it is always overwritten.
2356 */
2357 mas_for_each(mas, next, end - 1) {
2358 /* Does it split the end? */
2359 if (next->vm_end > end) {
2360 struct vm_area_struct *split;
2361
2362 error = __split_vma(mm, next, end, 1);
2363 if (error)
2364 goto end_split_failed;
11f9a21a 2365
763ecb03
LH
2366 mas_set(mas, end);
2367 split = mas_prev(mas, 0);
2368 error = munmap_sidetree(split, &mas_detach);
2369 if (error)
2370 goto munmap_sidetree_failed;
11f9a21a 2371
763ecb03
LH
2372 count++;
2373 if (vma == next)
2374 vma = split;
2375 break;
2376 }
2377 error = munmap_sidetree(next, &mas_detach);
1da177e4 2378 if (error)
763ecb03 2379 goto munmap_sidetree_failed;
11f9a21a 2380
763ecb03
LH
2381 count++;
2382#ifdef CONFIG_DEBUG_VM_MAPLE_TREE
2383 BUG_ON(next->vm_start < start);
2384 BUG_ON(next->vm_start > end);
2385#endif
1da177e4 2386 }
1da177e4 2387
763ecb03
LH
2388 if (!next)
2389 next = mas_next(mas, ULONG_MAX);
2390
2376dd7c
AA
2391 if (unlikely(uf)) {
2392 /*
2393 * If userfaultfd_unmap_prep returns an error the vmas
f0953a1b 2394 * will remain split, but userland will get a
2376dd7c
AA
2395 * highly unexpected error anyway. This is no
2396 * different than the case where the first of the two
2397 * __split_vma fails, but we don't undo the first
2398 * split, despite we could. This is unlikely enough
2399 * failure that it's not worth optimizing it for.
2400 */
69dbe6da 2401 error = userfaultfd_unmap_prep(mm, start, end, uf);
11f9a21a 2402
2376dd7c 2403 if (error)
d4af56c5 2404 goto userfaultfd_error;
2376dd7c
AA
2405 }
2406
763ecb03
LH
2407 /* Point of no return */
2408 mas_set_range(mas, start, end - 1);
2409#if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
2410 /* Make sure no VMAs are about to be lost. */
2411 {
2412 MA_STATE(test, &mt_detach, start, end - 1);
2413 struct vm_area_struct *vma_mas, *vma_test;
2414 int test_count = 0;
2415
2416 rcu_read_lock();
2417 vma_test = mas_find(&test, end - 1);
2418 mas_for_each(mas, vma_mas, end - 1) {
2419 BUG_ON(vma_mas != vma_test);
2420 test_count++;
2421 vma_test = mas_next(&test, end - 1);
2422 }
2423 rcu_read_unlock();
2424 BUG_ON(count != test_count);
2425 mas_set_range(mas, start, end - 1);
2426 }
2427#endif
11f9a21a 2428 mas_store_prealloc(mas, NULL);
763ecb03 2429 mm->map_count -= count;
11f9a21a
LH
2430 /*
2431 * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or
2432 * VM_GROWSUP VMA. Such VMAs can change their size under
2433 * down_read(mmap_lock) and collide with the VMA we are about to unmap.
2434 */
2435 if (downgrade) {
763ecb03 2436 if (next && (next->vm_flags & VM_GROWSDOWN))
11f9a21a
LH
2437 downgrade = false;
2438 else if (prev && (prev->vm_flags & VM_GROWSUP))
2439 downgrade = false;
2440 else
2441 mmap_write_downgrade(mm);
2442 }
dd2283f2 2443
763ecb03
LH
2444 unmap_region(mm, &mt_detach, vma, prev, next, start, end);
2445 /* Statistics and freeing VMAs */
2446 mas_set(&mas_detach, start);
2447 remove_mt(mm, &mas_detach);
2448 __mt_destroy(&mt_detach);
1da177e4 2449
524e00b3
LH
2450
2451 validate_mm(mm);
dd2283f2 2452 return downgrade ? 1 : 0;
d4af56c5 2453
d4af56c5 2454userfaultfd_error:
763ecb03
LH
2455munmap_sidetree_failed:
2456end_split_failed:
2457 __mt_destroy(&mt_detach);
2458start_split_failed:
2459map_count_exceeded:
11f9a21a 2460 mas_destroy(mas);
d4af56c5 2461 return error;
1da177e4 2462}
1da177e4 2463
11f9a21a
LH
2464/*
2465 * do_mas_munmap() - munmap a given range.
2466 * @mas: The maple state
2467 * @mm: The mm_struct
2468 * @start: The start address to munmap
2469 * @len: The length of the range to munmap
2470 * @uf: The userfaultfd list_head
2471 * @downgrade: set to true if the user wants to attempt to write_downgrade the
2472 * mmap_sem
2473 *
2474 * This function takes a @mas that is either pointing to the previous VMA or set
2475 * to MA_START and sets it up to remove the mapping(s). The @len will be
2476 * aligned and any arch_unmap work will be preformed.
2477 *
2478 * Returns: -EINVAL on failure, 1 on success and unlock, 0 otherwise.
2479 */
2480int do_mas_munmap(struct ma_state *mas, struct mm_struct *mm,
2481 unsigned long start, size_t len, struct list_head *uf,
2482 bool downgrade)
2483{
2484 unsigned long end;
2485 struct vm_area_struct *vma;
2486
2487 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2488 return -EINVAL;
2489
2490 end = start + PAGE_ALIGN(len);
2491 if (end == start)
2492 return -EINVAL;
2493
2494 /* arch_unmap() might do unmaps itself. */
2495 arch_unmap(mm, start, end);
2496
2497 /* Find the first overlapping VMA */
2498 vma = mas_find(mas, end - 1);
2499 if (!vma)
2500 return 0;
2501
2502 return do_mas_align_munmap(mas, vma, mm, start, end, uf, downgrade);
2503}
2504
2505/* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls.
2506 * @mm: The mm_struct
2507 * @start: The start address to munmap
2508 * @len: The length to be munmapped.
2509 * @uf: The userfaultfd list_head
2510 */
dd2283f2
YS
2511int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2512 struct list_head *uf)
2513{
11f9a21a
LH
2514 MA_STATE(mas, &mm->mm_mt, start, start);
2515
2516 return do_mas_munmap(&mas, mm, start, len, uf, false);
dd2283f2
YS
2517}
2518
e99668a5
LH
2519unsigned long mmap_region(struct file *file, unsigned long addr,
2520 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
2521 struct list_head *uf)
2522{
2523 struct mm_struct *mm = current->mm;
2524 struct vm_area_struct *vma = NULL;
2525 struct vm_area_struct *next, *prev, *merge;
2526 pgoff_t pglen = len >> PAGE_SHIFT;
2527 unsigned long charged = 0;
2528 unsigned long end = addr + len;
2529 unsigned long merge_start = addr, merge_end = end;
2530 pgoff_t vm_pgoff;
2531 int error;
2532 MA_STATE(mas, &mm->mm_mt, addr, end - 1);
2533
2534 /* Check against address space limit. */
2535 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
2536 unsigned long nr_pages;
2537
2538 /*
2539 * MAP_FIXED may remove pages of mappings that intersects with
2540 * requested mapping. Account for the pages it would unmap.
2541 */
2542 nr_pages = count_vma_pages_range(mm, addr, end);
2543
2544 if (!may_expand_vm(mm, vm_flags,
2545 (len >> PAGE_SHIFT) - nr_pages))
2546 return -ENOMEM;
2547 }
2548
2549 /* Unmap any existing mapping in the area */
11f9a21a 2550 if (do_mas_munmap(&mas, mm, addr, len, uf, false))
e99668a5
LH
2551 return -ENOMEM;
2552
2553 /*
2554 * Private writable mapping: check memory availability
2555 */
2556 if (accountable_mapping(file, vm_flags)) {
2557 charged = len >> PAGE_SHIFT;
2558 if (security_vm_enough_memory_mm(mm, charged))
2559 return -ENOMEM;
2560 vm_flags |= VM_ACCOUNT;
2561 }
2562
2563 next = mas_next(&mas, ULONG_MAX);
2564 prev = mas_prev(&mas, 0);
2565 if (vm_flags & VM_SPECIAL)
2566 goto cannot_expand;
2567
2568 /* Attempt to expand an old mapping */
2569 /* Check next */
2570 if (next && next->vm_start == end && !vma_policy(next) &&
2571 can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen,
2572 NULL_VM_UFFD_CTX, NULL)) {
2573 merge_end = next->vm_end;
2574 vma = next;
2575 vm_pgoff = next->vm_pgoff - pglen;
2576 }
2577
2578 /* Check prev */
2579 if (prev && prev->vm_end == addr && !vma_policy(prev) &&
2580 (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file,
2581 pgoff, vma->vm_userfaultfd_ctx, NULL) :
2582 can_vma_merge_after(prev, vm_flags, NULL, file, pgoff,
2583 NULL_VM_UFFD_CTX, NULL))) {
2584 merge_start = prev->vm_start;
2585 vma = prev;
2586 vm_pgoff = prev->vm_pgoff;
2587 }
2588
2589
2590 /* Actually expand, if possible */
2591 if (vma &&
2592 !vma_expand(&mas, vma, merge_start, merge_end, vm_pgoff, next)) {
2593 khugepaged_enter_vma(vma, vm_flags);
2594 goto expanded;
2595 }
2596
2597 mas.index = addr;
2598 mas.last = end - 1;
2599cannot_expand:
2600 /*
2601 * Determine the object being mapped and call the appropriate
2602 * specific mapper. the address has already been validated, but
2603 * not unmapped, but the maps are removed from the list.
2604 */
2605 vma = vm_area_alloc(mm);
2606 if (!vma) {
2607 error = -ENOMEM;
2608 goto unacct_error;
2609 }
2610
2611 vma->vm_start = addr;
2612 vma->vm_end = end;
2613 vma->vm_flags = vm_flags;
2614 vma->vm_page_prot = vm_get_page_prot(vm_flags);
2615 vma->vm_pgoff = pgoff;
2616
2617 if (file) {
2618 if (vm_flags & VM_SHARED) {
2619 error = mapping_map_writable(file->f_mapping);
2620 if (error)
2621 goto free_vma;
2622 }
2623
2624 vma->vm_file = get_file(file);
2625 error = call_mmap(file, vma);
2626 if (error)
2627 goto unmap_and_free_vma;
2628
a57b7051
LH
2629 /*
2630 * Expansion is handled above, merging is handled below.
2631 * Drivers should not alter the address of the VMA.
e99668a5 2632 */
a57b7051
LH
2633 if (WARN_ON((addr != vma->vm_start))) {
2634 error = -EINVAL;
2635 goto close_and_free_vma;
2636 }
e99668a5
LH
2637 mas_reset(&mas);
2638
2639 /*
2640 * If vm_flags changed after call_mmap(), we should try merge
2641 * vma again as we may succeed this time.
2642 */
2643 if (unlikely(vm_flags != vma->vm_flags && prev)) {
2644 merge = vma_merge(mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags,
2645 NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
2646 if (merge) {
2647 /*
2648 * ->mmap() can change vma->vm_file and fput
2649 * the original file. So fput the vma->vm_file
2650 * here or we would add an extra fput for file
2651 * and cause general protection fault
2652 * ultimately.
2653 */
2654 fput(vma->vm_file);
2655 vm_area_free(vma);
2656 vma = merge;
2657 /* Update vm_flags to pick up the change. */
e99668a5
LH
2658 vm_flags = vma->vm_flags;
2659 goto unmap_writable;
2660 }
2661 }
2662
2663 vm_flags = vma->vm_flags;
2664 } else if (vm_flags & VM_SHARED) {
2665 error = shmem_zero_setup(vma);
2666 if (error)
2667 goto free_vma;
2668 } else {
2669 vma_set_anonymous(vma);
2670 }
2671
2672 /* Allow architectures to sanity-check the vm_flags */
2673 if (!arch_validate_flags(vma->vm_flags)) {
2674 error = -EINVAL;
2675 if (file)
deb0f656 2676 goto close_and_free_vma;
cc674ab3
LZ
2677 else if (vma->vm_file)
2678 goto unmap_and_free_vma;
e99668a5
LH
2679 else
2680 goto free_vma;
2681 }
2682
2683 if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
2684 error = -ENOMEM;
2685 if (file)
5789151e 2686 goto close_and_free_vma;
cc674ab3
LZ
2687 else if (vma->vm_file)
2688 goto unmap_and_free_vma;
e99668a5
LH
2689 else
2690 goto free_vma;
2691 }
2692
2693 if (vma->vm_file)
2694 i_mmap_lock_write(vma->vm_file->f_mapping);
2695
2696 vma_mas_store(vma, &mas);
e99668a5
LH
2697 mm->map_count++;
2698 if (vma->vm_file) {
2699 if (vma->vm_flags & VM_SHARED)
2700 mapping_allow_writable(vma->vm_file->f_mapping);
2701
2702 flush_dcache_mmap_lock(vma->vm_file->f_mapping);
2703 vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap);
2704 flush_dcache_mmap_unlock(vma->vm_file->f_mapping);
2705 i_mmap_unlock_write(vma->vm_file->f_mapping);
2706 }
2707
2708 /*
2709 * vma_merge() calls khugepaged_enter_vma() either, the below
2710 * call covers the non-merge case.
2711 */
2712 khugepaged_enter_vma(vma, vma->vm_flags);
2713
2714 /* Once vma denies write, undo our temporary denial count */
2715unmap_writable:
2716 if (file && vm_flags & VM_SHARED)
2717 mapping_unmap_writable(file->f_mapping);
2718 file = vma->vm_file;
2719expanded:
2720 perf_event_mmap(vma);
2721
2722 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
2723 if (vm_flags & VM_LOCKED) {
2724 if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) ||
2725 is_vm_hugetlb_page(vma) ||
2726 vma == get_gate_vma(current->mm))
2727 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
2728 else
2729 mm->locked_vm += (len >> PAGE_SHIFT);
2730 }
2731
2732 if (file)
2733 uprobe_mmap(vma);
2734
2735 /*
2736 * New (or expanded) vma always get soft dirty status.
2737 * Otherwise user-space soft-dirty page tracker won't
2738 * be able to distinguish situation when vma area unmapped,
2739 * then new mapped in-place (which must be aimed as
2740 * a completely new data area).
2741 */
2742 vma->vm_flags |= VM_SOFTDIRTY;
2743
2744 vma_set_page_prot(vma);
2745
2746 validate_mm(mm);
2747 return addr;
2748
deb0f656
CL
2749close_and_free_vma:
2750 if (vma->vm_ops && vma->vm_ops->close)
2751 vma->vm_ops->close(vma);
e99668a5
LH
2752unmap_and_free_vma:
2753 fput(vma->vm_file);
2754 vma->vm_file = NULL;
2755
2756 /* Undo any partial mapping done by a device driver. */
763ecb03 2757 unmap_region(mm, mas.tree, vma, prev, next, vma->vm_start, vma->vm_end);
cc674ab3 2758 if (file && (vm_flags & VM_SHARED))
e99668a5
LH
2759 mapping_unmap_writable(file->f_mapping);
2760free_vma:
2761 vm_area_free(vma);
2762unacct_error:
2763 if (charged)
2764 vm_unacct_memory(charged);
2765 validate_mm(mm);
2766 return error;
2767}
2768
dd2283f2 2769static int __vm_munmap(unsigned long start, size_t len, bool downgrade)
1da177e4
LT
2770{
2771 int ret;
bfce281c 2772 struct mm_struct *mm = current->mm;
897ab3e0 2773 LIST_HEAD(uf);
11f9a21a 2774 MA_STATE(mas, &mm->mm_mt, start, start);
1da177e4 2775
d8ed45c5 2776 if (mmap_write_lock_killable(mm))
ae798783
MH
2777 return -EINTR;
2778
11f9a21a 2779 ret = do_mas_munmap(&mas, mm, start, len, &uf, downgrade);
dd2283f2 2780 /*
c1e8d7c6 2781 * Returning 1 indicates mmap_lock is downgraded.
dd2283f2
YS
2782 * But 1 is not legal return value of vm_munmap() and munmap(), reset
2783 * it to 0 before return.
2784 */
2785 if (ret == 1) {
d8ed45c5 2786 mmap_read_unlock(mm);
dd2283f2
YS
2787 ret = 0;
2788 } else
d8ed45c5 2789 mmap_write_unlock(mm);
dd2283f2 2790
897ab3e0 2791 userfaultfd_unmap_complete(mm, &uf);
1da177e4
LT
2792 return ret;
2793}
dd2283f2
YS
2794
2795int vm_munmap(unsigned long start, size_t len)
2796{
2797 return __vm_munmap(start, len, false);
2798}
a46ef99d
LT
2799EXPORT_SYMBOL(vm_munmap);
2800
2801SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2802{
ce18d171 2803 addr = untagged_addr(addr);
dd2283f2 2804 return __vm_munmap(addr, len, true);
a46ef99d 2805}
1da177e4 2806
c8d78c18
KS
2807
2808/*
2809 * Emulation of deprecated remap_file_pages() syscall.
2810 */
2811SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2812 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2813{
2814
2815 struct mm_struct *mm = current->mm;
2816 struct vm_area_struct *vma;
2817 unsigned long populate = 0;
2818 unsigned long ret = -EINVAL;
2819 struct file *file;
2820
ee65728e 2821 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n",
756a025f 2822 current->comm, current->pid);
c8d78c18
KS
2823
2824 if (prot)
2825 return ret;
2826 start = start & PAGE_MASK;
2827 size = size & PAGE_MASK;
2828
2829 if (start + size <= start)
2830 return ret;
2831
2832 /* Does pgoff wrap? */
2833 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2834 return ret;
2835
d8ed45c5 2836 if (mmap_write_lock_killable(mm))
dc0ef0df
MH
2837 return -EINTR;
2838
9b593cb2 2839 vma = vma_lookup(mm, start);
c8d78c18
KS
2840
2841 if (!vma || !(vma->vm_flags & VM_SHARED))
2842 goto out;
2843
48f7df32 2844 if (start + size > vma->vm_end) {
763ecb03
LH
2845 VMA_ITERATOR(vmi, mm, vma->vm_end);
2846 struct vm_area_struct *next, *prev = vma;
48f7df32 2847
763ecb03 2848 for_each_vma_range(vmi, next, start + size) {
48f7df32 2849 /* hole between vmas ? */
763ecb03 2850 if (next->vm_start != prev->vm_end)
48f7df32
KS
2851 goto out;
2852
2853 if (next->vm_file != vma->vm_file)
2854 goto out;
2855
2856 if (next->vm_flags != vma->vm_flags)
2857 goto out;
2858
1db43d3f
LH
2859 if (start + size <= next->vm_end)
2860 break;
2861
763ecb03 2862 prev = next;
48f7df32
KS
2863 }
2864
2865 if (!next)
2866 goto out;
c8d78c18
KS
2867 }
2868
2869 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2870 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2871 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2872
2873 flags &= MAP_NONBLOCK;
2874 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
fce000b1 2875 if (vma->vm_flags & VM_LOCKED)
c8d78c18 2876 flags |= MAP_LOCKED;
48f7df32 2877
c8d78c18 2878 file = get_file(vma->vm_file);
45e55300 2879 ret = do_mmap(vma->vm_file, start, size,
897ab3e0 2880 prot, flags, pgoff, &populate, NULL);
c8d78c18
KS
2881 fput(file);
2882out:
d8ed45c5 2883 mmap_write_unlock(mm);
c8d78c18
KS
2884 if (populate)
2885 mm_populate(ret, populate);
2886 if (!IS_ERR_VALUE(ret))
2887 ret = 0;
2888 return ret;
2889}
2890
1da177e4 2891/*
2e7ce7d3
LH
2892 * brk_munmap() - Unmap a parital vma.
2893 * @mas: The maple tree state.
2894 * @vma: The vma to be modified
2895 * @newbrk: the start of the address to unmap
2896 * @oldbrk: The end of the address to unmap
2897 * @uf: The userfaultfd list_head
2898 *
2899 * Returns: 1 on success.
2900 * unmaps a partial VMA mapping. Does not handle alignment, downgrades lock if
2901 * possible.
1da177e4 2902 */
2e7ce7d3
LH
2903static int do_brk_munmap(struct ma_state *mas, struct vm_area_struct *vma,
2904 unsigned long newbrk, unsigned long oldbrk,
2905 struct list_head *uf)
1da177e4 2906{
2e7ce7d3
LH
2907 struct mm_struct *mm = vma->vm_mm;
2908 int ret;
3a459756 2909
2e7ce7d3 2910 arch_unmap(mm, newbrk, oldbrk);
67e7c167 2911 ret = do_mas_align_munmap(mas, vma, mm, newbrk, oldbrk, uf, true);
2e7ce7d3
LH
2912 validate_mm_mt(mm);
2913 return ret;
2914}
1da177e4 2915
2e7ce7d3
LH
2916/*
2917 * do_brk_flags() - Increase the brk vma if the flags match.
2918 * @mas: The maple tree state.
2919 * @addr: The start address
2920 * @len: The length of the increase
2921 * @vma: The vma,
2922 * @flags: The VMA Flags
2923 *
2924 * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags
2925 * do not match then create a new anonymous VMA. Eventually we may be able to
2926 * do some brk-specific accounting here.
2927 */
2928static int do_brk_flags(struct ma_state *mas, struct vm_area_struct *vma,
763ecb03 2929 unsigned long addr, unsigned long len, unsigned long flags)
2e7ce7d3
LH
2930{
2931 struct mm_struct *mm = current->mm;
1da177e4 2932
2e7ce7d3
LH
2933 validate_mm_mt(mm);
2934 /*
2935 * Check against address space limits by the changed size
2936 * Note: This happens *after* clearing old mappings in some code paths.
2937 */
2938 flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
84638335 2939 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
1da177e4
LT
2940 return -ENOMEM;
2941
2942 if (mm->map_count > sysctl_max_map_count)
2943 return -ENOMEM;
2944
191c5424 2945 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
1da177e4
LT
2946 return -ENOMEM;
2947
1da177e4 2948 /*
2e7ce7d3
LH
2949 * Expand the existing vma if possible; Note that singular lists do not
2950 * occur after forking, so the expand will only happen on new VMAs.
1da177e4 2951 */
2e7ce7d3
LH
2952 if (vma &&
2953 (!vma->anon_vma || list_is_singular(&vma->anon_vma_chain)) &&
2954 ((vma->vm_flags & ~VM_SOFTDIRTY) == flags)) {
28c5609f
LH
2955 mas_set_range(mas, vma->vm_start, addr + len - 1);
2956 if (mas_preallocate(mas, vma, GFP_KERNEL))
2957 return -ENOMEM;
2958
2959 vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0);
2e7ce7d3
LH
2960 if (vma->anon_vma) {
2961 anon_vma_lock_write(vma->anon_vma);
2962 anon_vma_interval_tree_pre_update_vma(vma);
2963 }
2964 vma->vm_end = addr + len;
2965 vma->vm_flags |= VM_SOFTDIRTY;
28c5609f 2966 mas_store_prealloc(mas, vma);
2e7ce7d3
LH
2967
2968 if (vma->anon_vma) {
2969 anon_vma_interval_tree_post_update_vma(vma);
2970 anon_vma_unlock_write(vma->anon_vma);
2971 }
2972 khugepaged_enter_vma(vma, flags);
2973 goto out;
1da177e4 2974 }
2e7ce7d3
LH
2975
2976 /* create a vma struct for an anonymous mapping */
2977 vma = vm_area_alloc(mm);
2978 if (!vma)
2979 goto vma_alloc_fail;
1da177e4 2980
bfd40eaf 2981 vma_set_anonymous(vma);
1da177e4
LT
2982 vma->vm_start = addr;
2983 vma->vm_end = addr + len;
2e7ce7d3 2984 vma->vm_pgoff = addr >> PAGE_SHIFT;
1da177e4 2985 vma->vm_flags = flags;
3ed75eb8 2986 vma->vm_page_prot = vm_get_page_prot(flags);
2e7ce7d3
LH
2987 mas_set_range(mas, vma->vm_start, addr + len - 1);
2988 if (mas_store_gfp(mas, vma, GFP_KERNEL))
2989 goto mas_store_fail;
d4af56c5 2990
2e7ce7d3 2991 mm->map_count++;
1da177e4 2992out:
3af9e859 2993 perf_event_mmap(vma);
1da177e4 2994 mm->total_vm += len >> PAGE_SHIFT;
84638335 2995 mm->data_vm += len >> PAGE_SHIFT;
128557ff
ML
2996 if (flags & VM_LOCKED)
2997 mm->locked_vm += (len >> PAGE_SHIFT);
d9104d1c 2998 vma->vm_flags |= VM_SOFTDIRTY;
763ecb03 2999 validate_mm(mm);
5d22fc25 3000 return 0;
d4af56c5 3001
2e7ce7d3 3002mas_store_fail:
d4af56c5 3003 vm_area_free(vma);
2e7ce7d3
LH
3004vma_alloc_fail:
3005 vm_unacct_memory(len >> PAGE_SHIFT);
3006 return -ENOMEM;
1da177e4
LT
3007}
3008
bb177a73 3009int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
e4eb1ff6
LT
3010{
3011 struct mm_struct *mm = current->mm;
2e7ce7d3 3012 struct vm_area_struct *vma = NULL;
bb177a73 3013 unsigned long len;
5d22fc25 3014 int ret;
128557ff 3015 bool populate;
897ab3e0 3016 LIST_HEAD(uf);
2e7ce7d3 3017 MA_STATE(mas, &mm->mm_mt, addr, addr);
e4eb1ff6 3018
bb177a73
MH
3019 len = PAGE_ALIGN(request);
3020 if (len < request)
3021 return -ENOMEM;
3022 if (!len)
3023 return 0;
3024
d8ed45c5 3025 if (mmap_write_lock_killable(mm))
2d6c9282
MH
3026 return -EINTR;
3027
2e7ce7d3
LH
3028 /* Until we need other flags, refuse anything except VM_EXEC. */
3029 if ((flags & (~VM_EXEC)) != 0)
3030 return -EINVAL;
3031
3032 ret = check_brk_limits(addr, len);
3033 if (ret)
3034 goto limits_failed;
3035
11f9a21a 3036 ret = do_mas_munmap(&mas, mm, addr, len, &uf, 0);
2e7ce7d3
LH
3037 if (ret)
3038 goto munmap_failed;
3039
3040 vma = mas_prev(&mas, 0);
3041 if (!vma || vma->vm_end != addr || vma_policy(vma) ||
3042 !can_vma_merge_after(vma, flags, NULL, NULL,
3043 addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL))
3044 vma = NULL;
3045
3046 ret = do_brk_flags(&mas, vma, addr, len, flags);
128557ff 3047 populate = ((mm->def_flags & VM_LOCKED) != 0);
d8ed45c5 3048 mmap_write_unlock(mm);
897ab3e0 3049 userfaultfd_unmap_complete(mm, &uf);
5d22fc25 3050 if (populate && !ret)
128557ff 3051 mm_populate(addr, len);
e4eb1ff6 3052 return ret;
2e7ce7d3
LH
3053
3054munmap_failed:
3055limits_failed:
3056 mmap_write_unlock(mm);
3057 return ret;
e4eb1ff6 3058}
16e72e9b
DV
3059EXPORT_SYMBOL(vm_brk_flags);
3060
3061int vm_brk(unsigned long addr, unsigned long len)
3062{
3063 return vm_brk_flags(addr, len, 0);
3064}
e4eb1ff6 3065EXPORT_SYMBOL(vm_brk);
1da177e4
LT
3066
3067/* Release all mmaps. */
3068void exit_mmap(struct mm_struct *mm)
3069{
d16dfc55 3070 struct mmu_gather tlb;
ba470de4 3071 struct vm_area_struct *vma;
1da177e4 3072 unsigned long nr_accounted = 0;
763ecb03
LH
3073 MA_STATE(mas, &mm->mm_mt, 0, 0);
3074 int count = 0;
1da177e4 3075
d6dd61c8 3076 /* mm's last user has gone, and its about to be pulled down */
cddb8a5c 3077 mmu_notifier_release(mm);
d6dd61c8 3078
bf3980c8 3079 mmap_read_lock(mm);
9480c53e
JF
3080 arch_exit_mmap(mm);
3081
763ecb03 3082 vma = mas_find(&mas, ULONG_MAX);
64591e86
SB
3083 if (!vma) {
3084 /* Can happen if dup_mmap() received an OOM */
bf3980c8 3085 mmap_read_unlock(mm);
9480c53e 3086 return;
64591e86 3087 }
9480c53e 3088
1da177e4 3089 lru_add_drain();
1da177e4 3090 flush_cache_mm(mm);
d8b45053 3091 tlb_gather_mmu_fullmm(&tlb, mm);
901608d9 3092 /* update_hiwater_rss(mm) here? but nobody should be looking */
763ecb03
LH
3093 /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */
3094 unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX);
bf3980c8
SB
3095 mmap_read_unlock(mm);
3096
3097 /*
3098 * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper
b3541d91 3099 * because the memory has been already freed.
bf3980c8
SB
3100 */
3101 set_bit(MMF_OOM_SKIP, &mm->flags);
3102 mmap_write_lock(mm);
763ecb03
LH
3103 free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS,
3104 USER_PGTABLES_CEILING);
ae8eba8b 3105 tlb_finish_mmu(&tlb);
1da177e4 3106
763ecb03
LH
3107 /*
3108 * Walk the list again, actually closing and freeing it, with preemption
3109 * enabled, without holding any MM locks besides the unreachable
3110 * mmap_write_lock.
3111 */
3112 do {
4f74d2c8
LT
3113 if (vma->vm_flags & VM_ACCOUNT)
3114 nr_accounted += vma_pages(vma);
763ecb03
LH
3115 remove_vma(vma);
3116 count++;
0a3b3c25 3117 cond_resched();
763ecb03
LH
3118 } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
3119
3120 BUG_ON(count != mm->map_count);
d4af56c5
LH
3121
3122 trace_exit_mmap(mm);
3123 __mt_destroy(&mm->mm_mt);
64591e86 3124 mmap_write_unlock(mm);
4f74d2c8 3125 vm_unacct_memory(nr_accounted);
1da177e4
LT
3126}
3127
3128/* Insert vm structure into process list sorted by address
3129 * and into the inode's i_mmap tree. If vm_file is non-NULL
c8c06efa 3130 * then i_mmap_rwsem is taken here.
1da177e4 3131 */
6597d783 3132int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 3133{
d4af56c5 3134 unsigned long charged = vma_pages(vma);
1da177e4 3135
d4af56c5 3136
d0601a50 3137 if (find_vma_intersection(mm, vma->vm_start, vma->vm_end))
c9d13f5f 3138 return -ENOMEM;
d4af56c5 3139
c9d13f5f 3140 if ((vma->vm_flags & VM_ACCOUNT) &&
d4af56c5 3141 security_vm_enough_memory_mm(mm, charged))
c9d13f5f
CG
3142 return -ENOMEM;
3143
1da177e4
LT
3144 /*
3145 * The vm_pgoff of a purely anonymous vma should be irrelevant
3146 * until its first write fault, when page's anon_vma and index
3147 * are set. But now set the vm_pgoff it will almost certainly
3148 * end up with (unless mremap moves it elsewhere before that
3149 * first wfault), so /proc/pid/maps tells a consistent story.
3150 *
3151 * By setting it to reflect the virtual start address of the
3152 * vma, merges and splits can happen in a seamless way, just
3153 * using the existing file pgoff checks and manipulations.
8332326e 3154 * Similarly in do_mmap and in do_brk_flags.
1da177e4 3155 */
8a9cc3b5 3156 if (vma_is_anonymous(vma)) {
1da177e4
LT
3157 BUG_ON(vma->anon_vma);
3158 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3159 }
2b144498 3160
763ecb03 3161 if (vma_link(mm, vma)) {
d4af56c5
LH
3162 vm_unacct_memory(charged);
3163 return -ENOMEM;
3164 }
3165
1da177e4
LT
3166 return 0;
3167}
3168
3169/*
3170 * Copy the vma structure to a new location in the same mm,
3171 * prior to moving page table entries, to effect an mremap move.
3172 */
3173struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
38a76013
ML
3174 unsigned long addr, unsigned long len, pgoff_t pgoff,
3175 bool *need_rmap_locks)
1da177e4
LT
3176{
3177 struct vm_area_struct *vma = *vmap;
3178 unsigned long vma_start = vma->vm_start;
3179 struct mm_struct *mm = vma->vm_mm;
3180 struct vm_area_struct *new_vma, *prev;
948f017b 3181 bool faulted_in_anon_vma = true;
1da177e4 3182
d4af56c5 3183 validate_mm_mt(mm);
1da177e4
LT
3184 /*
3185 * If anonymous vma has not yet been faulted, update new pgoff
3186 * to match new location, to increase its chance of merging.
3187 */
ce75799b 3188 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
1da177e4 3189 pgoff = addr >> PAGE_SHIFT;
948f017b
AA
3190 faulted_in_anon_vma = false;
3191 }
1da177e4 3192
763ecb03
LH
3193 new_vma = find_vma_prev(mm, addr, &prev);
3194 if (new_vma && new_vma->vm_start < addr + len)
6597d783 3195 return NULL; /* should never get here */
524e00b3 3196
1da177e4 3197 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
19a809af 3198 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
5c26f6ac 3199 vma->vm_userfaultfd_ctx, anon_vma_name(vma));
1da177e4
LT
3200 if (new_vma) {
3201 /*
3202 * Source vma may have been merged into new_vma
3203 */
948f017b
AA
3204 if (unlikely(vma_start >= new_vma->vm_start &&
3205 vma_start < new_vma->vm_end)) {
3206 /*
3207 * The only way we can get a vma_merge with
3208 * self during an mremap is if the vma hasn't
3209 * been faulted in yet and we were allowed to
3210 * reset the dst vma->vm_pgoff to the
3211 * destination address of the mremap to allow
3212 * the merge to happen. mremap must change the
3213 * vm_pgoff linearity between src and dst vmas
3214 * (in turn preventing a vma_merge) to be
3215 * safe. It is only safe to keep the vm_pgoff
3216 * linear if there are no pages mapped yet.
3217 */
81d1b09c 3218 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
38a76013 3219 *vmap = vma = new_vma;
108d6642 3220 }
38a76013 3221 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
1da177e4 3222 } else {
3928d4f5 3223 new_vma = vm_area_dup(vma);
e3975891
CG
3224 if (!new_vma)
3225 goto out;
e3975891
CG
3226 new_vma->vm_start = addr;
3227 new_vma->vm_end = addr + len;
3228 new_vma->vm_pgoff = pgoff;
3229 if (vma_dup_policy(vma, new_vma))
3230 goto out_free_vma;
e3975891
CG
3231 if (anon_vma_clone(new_vma, vma))
3232 goto out_free_mempol;
3233 if (new_vma->vm_file)
3234 get_file(new_vma->vm_file);
3235 if (new_vma->vm_ops && new_vma->vm_ops->open)
3236 new_vma->vm_ops->open(new_vma);
763ecb03 3237 if (vma_link(mm, new_vma))
524e00b3 3238 goto out_vma_link;
e3975891 3239 *need_rmap_locks = false;
1da177e4 3240 }
d4af56c5 3241 validate_mm_mt(mm);
1da177e4 3242 return new_vma;
5beb4930 3243
524e00b3
LH
3244out_vma_link:
3245 if (new_vma->vm_ops && new_vma->vm_ops->close)
3246 new_vma->vm_ops->close(new_vma);
92b73996
LH
3247
3248 if (new_vma->vm_file)
3249 fput(new_vma->vm_file);
3250
3251 unlink_anon_vmas(new_vma);
e3975891 3252out_free_mempol:
ef0855d3 3253 mpol_put(vma_policy(new_vma));
e3975891 3254out_free_vma:
3928d4f5 3255 vm_area_free(new_vma);
e3975891 3256out:
d4af56c5 3257 validate_mm_mt(mm);
5beb4930 3258 return NULL;
1da177e4 3259}
119f657c 3260
3261/*
3262 * Return true if the calling process may expand its vm space by the passed
3263 * number of pages
3264 */
84638335 3265bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
119f657c 3266{
84638335
KK
3267 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3268 return false;
119f657c 3269
d977d56c
KK
3270 if (is_data_mapping(flags) &&
3271 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
f4fcd558
KK
3272 /* Workaround for Valgrind */
3273 if (rlimit(RLIMIT_DATA) == 0 &&
3274 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3275 return true;
57a7702b
DW
3276
3277 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3278 current->comm, current->pid,
3279 (mm->data_vm + npages) << PAGE_SHIFT,
3280 rlimit(RLIMIT_DATA),
3281 ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data");
3282
3283 if (!ignore_rlimit_data)
d977d56c
KK
3284 return false;
3285 }
119f657c 3286
84638335
KK
3287 return true;
3288}
3289
3290void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3291{
7866076b 3292 WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages);
84638335 3293
d977d56c 3294 if (is_exec_mapping(flags))
84638335 3295 mm->exec_vm += npages;
d977d56c 3296 else if (is_stack_mapping(flags))
84638335 3297 mm->stack_vm += npages;
d977d56c 3298 else if (is_data_mapping(flags))
84638335 3299 mm->data_vm += npages;
119f657c 3300}
fa5dc22f 3301
b3ec9f33 3302static vm_fault_t special_mapping_fault(struct vm_fault *vmf);
a62c34bd
AL
3303
3304/*
3305 * Having a close hook prevents vma merging regardless of flags.
3306 */
3307static void special_mapping_close(struct vm_area_struct *vma)
3308{
3309}
3310
3311static const char *special_mapping_name(struct vm_area_struct *vma)
3312{
3313 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3314}
3315
14d07113 3316static int special_mapping_mremap(struct vm_area_struct *new_vma)
b059a453
DS
3317{
3318 struct vm_special_mapping *sm = new_vma->vm_private_data;
3319
280e87e9
DS
3320 if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3321 return -EFAULT;
3322
b059a453
DS
3323 if (sm->mremap)
3324 return sm->mremap(sm, new_vma);
280e87e9 3325
b059a453
DS
3326 return 0;
3327}
3328
871402e0
DS
3329static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr)
3330{
3331 /*
3332 * Forbid splitting special mappings - kernel has expectations over
3333 * the number of pages in mapping. Together with VM_DONTEXPAND
3334 * the size of vma should stay the same over the special mapping's
3335 * lifetime.
3336 */
3337 return -EINVAL;
3338}
3339
a62c34bd
AL
3340static const struct vm_operations_struct special_mapping_vmops = {
3341 .close = special_mapping_close,
3342 .fault = special_mapping_fault,
b059a453 3343 .mremap = special_mapping_mremap,
a62c34bd 3344 .name = special_mapping_name,
af34ebeb
DS
3345 /* vDSO code relies that VVAR can't be accessed remotely */
3346 .access = NULL,
871402e0 3347 .may_split = special_mapping_split,
a62c34bd
AL
3348};
3349
3350static const struct vm_operations_struct legacy_special_mapping_vmops = {
3351 .close = special_mapping_close,
3352 .fault = special_mapping_fault,
3353};
fa5dc22f 3354
b3ec9f33 3355static vm_fault_t special_mapping_fault(struct vm_fault *vmf)
fa5dc22f 3356{
11bac800 3357 struct vm_area_struct *vma = vmf->vma;
b1d0e4f5 3358 pgoff_t pgoff;
fa5dc22f
RM
3359 struct page **pages;
3360
f872f540 3361 if (vma->vm_ops == &legacy_special_mapping_vmops) {
a62c34bd 3362 pages = vma->vm_private_data;
f872f540
AL
3363 } else {
3364 struct vm_special_mapping *sm = vma->vm_private_data;
3365
3366 if (sm->fault)
11bac800 3367 return sm->fault(sm, vmf->vma, vmf);
f872f540
AL
3368
3369 pages = sm->pages;
3370 }
a62c34bd 3371
8a9cc3b5 3372 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
b1d0e4f5 3373 pgoff--;
fa5dc22f
RM
3374
3375 if (*pages) {
3376 struct page *page = *pages;
3377 get_page(page);
b1d0e4f5
NP
3378 vmf->page = page;
3379 return 0;
fa5dc22f
RM
3380 }
3381
b1d0e4f5 3382 return VM_FAULT_SIGBUS;
fa5dc22f
RM
3383}
3384
a62c34bd
AL
3385static struct vm_area_struct *__install_special_mapping(
3386 struct mm_struct *mm,
3387 unsigned long addr, unsigned long len,
27f28b97
CG
3388 unsigned long vm_flags, void *priv,
3389 const struct vm_operations_struct *ops)
fa5dc22f 3390{
462e635e 3391 int ret;
fa5dc22f
RM
3392 struct vm_area_struct *vma;
3393
d4af56c5 3394 validate_mm_mt(mm);
490fc053 3395 vma = vm_area_alloc(mm);
fa5dc22f 3396 if (unlikely(vma == NULL))
3935ed6a 3397 return ERR_PTR(-ENOMEM);
fa5dc22f 3398
fa5dc22f
RM
3399 vma->vm_start = addr;
3400 vma->vm_end = addr + len;
3401
d9104d1c 3402 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
1fc09228 3403 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
3ed75eb8 3404 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
fa5dc22f 3405
a62c34bd
AL
3406 vma->vm_ops = ops;
3407 vma->vm_private_data = priv;
fa5dc22f 3408
462e635e
TO
3409 ret = insert_vm_struct(mm, vma);
3410 if (ret)
3411 goto out;
fa5dc22f 3412
84638335 3413 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
fa5dc22f 3414
cdd6c482 3415 perf_event_mmap(vma);
089dd79d 3416
d4af56c5 3417 validate_mm_mt(mm);
3935ed6a 3418 return vma;
462e635e
TO
3419
3420out:
3928d4f5 3421 vm_area_free(vma);
d4af56c5 3422 validate_mm_mt(mm);
3935ed6a
SS
3423 return ERR_PTR(ret);
3424}
3425
2eefd878
DS
3426bool vma_is_special_mapping(const struct vm_area_struct *vma,
3427 const struct vm_special_mapping *sm)
3428{
3429 return vma->vm_private_data == sm &&
3430 (vma->vm_ops == &special_mapping_vmops ||
3431 vma->vm_ops == &legacy_special_mapping_vmops);
3432}
3433
a62c34bd 3434/*
c1e8d7c6 3435 * Called with mm->mmap_lock held for writing.
a62c34bd
AL
3436 * Insert a new vma covering the given region, with the given flags.
3437 * Its pages are supplied by the given array of struct page *.
3438 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3439 * The region past the last page supplied will always produce SIGBUS.
3440 * The array pointer and the pages it points to are assumed to stay alive
3441 * for as long as this mapping might exist.
3442 */
3443struct vm_area_struct *_install_special_mapping(
3444 struct mm_struct *mm,
3445 unsigned long addr, unsigned long len,
3446 unsigned long vm_flags, const struct vm_special_mapping *spec)
3447{
27f28b97
CG
3448 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3449 &special_mapping_vmops);
a62c34bd
AL
3450}
3451
3935ed6a
SS
3452int install_special_mapping(struct mm_struct *mm,
3453 unsigned long addr, unsigned long len,
3454 unsigned long vm_flags, struct page **pages)
3455{
a62c34bd 3456 struct vm_area_struct *vma = __install_special_mapping(
27f28b97
CG
3457 mm, addr, len, vm_flags, (void *)pages,
3458 &legacy_special_mapping_vmops);
3935ed6a 3459
14bd5b45 3460 return PTR_ERR_OR_ZERO(vma);
fa5dc22f 3461}
7906d00c
AA
3462
3463static DEFINE_MUTEX(mm_all_locks_mutex);
3464
454ed842 3465static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
7906d00c 3466{
f808c13f 3467 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
7906d00c
AA
3468 /*
3469 * The LSB of head.next can't change from under us
3470 * because we hold the mm_all_locks_mutex.
3471 */
da1c55f1 3472 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock);
7906d00c
AA
3473 /*
3474 * We can safely modify head.next after taking the
5a505085 3475 * anon_vma->root->rwsem. If some other vma in this mm shares
7906d00c
AA
3476 * the same anon_vma we won't take it again.
3477 *
3478 * No need of atomic instructions here, head.next
3479 * can't change from under us thanks to the
5a505085 3480 * anon_vma->root->rwsem.
7906d00c
AA
3481 */
3482 if (__test_and_set_bit(0, (unsigned long *)
f808c13f 3483 &anon_vma->root->rb_root.rb_root.rb_node))
7906d00c
AA
3484 BUG();
3485 }
3486}
3487
454ed842 3488static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
7906d00c
AA
3489{
3490 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3491 /*
3492 * AS_MM_ALL_LOCKS can't change from under us because
3493 * we hold the mm_all_locks_mutex.
3494 *
3495 * Operations on ->flags have to be atomic because
3496 * even if AS_MM_ALL_LOCKS is stable thanks to the
3497 * mm_all_locks_mutex, there may be other cpus
3498 * changing other bitflags in parallel to us.
3499 */
3500 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3501 BUG();
da1c55f1 3502 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock);
7906d00c
AA
3503 }
3504}
3505
3506/*
3507 * This operation locks against the VM for all pte/vma/mm related
3508 * operations that could ever happen on a certain mm. This includes
3509 * vmtruncate, try_to_unmap, and all page faults.
3510 *
c1e8d7c6 3511 * The caller must take the mmap_lock in write mode before calling
7906d00c 3512 * mm_take_all_locks(). The caller isn't allowed to release the
c1e8d7c6 3513 * mmap_lock until mm_drop_all_locks() returns.
7906d00c 3514 *
c1e8d7c6 3515 * mmap_lock in write mode is required in order to block all operations
7906d00c 3516 * that could modify pagetables and free pages without need of
27ba0644 3517 * altering the vma layout. It's also needed in write mode to avoid new
7906d00c
AA
3518 * anon_vmas to be associated with existing vmas.
3519 *
3520 * A single task can't take more than one mm_take_all_locks() in a row
3521 * or it would deadlock.
3522 *
bf181b9f 3523 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
7906d00c
AA
3524 * mapping->flags avoid to take the same lock twice, if more than one
3525 * vma in this mm is backed by the same anon_vma or address_space.
3526 *
88f306b6
KS
3527 * We take locks in following order, accordingly to comment at beginning
3528 * of mm/rmap.c:
3529 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3530 * hugetlb mapping);
3531 * - all i_mmap_rwsem locks;
3532 * - all anon_vma->rwseml
3533 *
3534 * We can take all locks within these types randomly because the VM code
3535 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3536 * mm_all_locks_mutex.
7906d00c
AA
3537 *
3538 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3539 * that may have to take thousand of locks.
3540 *
3541 * mm_take_all_locks() can fail if it's interrupted by signals.
3542 */
3543int mm_take_all_locks(struct mm_struct *mm)
3544{
3545 struct vm_area_struct *vma;
5beb4930 3546 struct anon_vma_chain *avc;
763ecb03 3547 MA_STATE(mas, &mm->mm_mt, 0, 0);
7906d00c 3548
325bca1f 3549 mmap_assert_write_locked(mm);
7906d00c
AA
3550
3551 mutex_lock(&mm_all_locks_mutex);
3552
763ecb03 3553 mas_for_each(&mas, vma, ULONG_MAX) {
7906d00c
AA
3554 if (signal_pending(current))
3555 goto out_unlock;
88f306b6
KS
3556 if (vma->vm_file && vma->vm_file->f_mapping &&
3557 is_vm_hugetlb_page(vma))
3558 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3559 }
3560
763ecb03
LH
3561 mas_set(&mas, 0);
3562 mas_for_each(&mas, vma, ULONG_MAX) {
88f306b6
KS
3563 if (signal_pending(current))
3564 goto out_unlock;
3565 if (vma->vm_file && vma->vm_file->f_mapping &&
3566 !is_vm_hugetlb_page(vma))
454ed842 3567 vm_lock_mapping(mm, vma->vm_file->f_mapping);
7906d00c 3568 }
7cd5a02f 3569
763ecb03
LH
3570 mas_set(&mas, 0);
3571 mas_for_each(&mas, vma, ULONG_MAX) {
7cd5a02f
PZ
3572 if (signal_pending(current))
3573 goto out_unlock;
3574 if (vma->anon_vma)
5beb4930
RR
3575 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3576 vm_lock_anon_vma(mm, avc->anon_vma);
7906d00c 3577 }
7cd5a02f 3578
584cff54 3579 return 0;
7906d00c
AA
3580
3581out_unlock:
584cff54
KC
3582 mm_drop_all_locks(mm);
3583 return -EINTR;
7906d00c
AA
3584}
3585
3586static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3587{
f808c13f 3588 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
7906d00c
AA
3589 /*
3590 * The LSB of head.next can't change to 0 from under
3591 * us because we hold the mm_all_locks_mutex.
3592 *
3593 * We must however clear the bitflag before unlocking
bf181b9f 3594 * the vma so the users using the anon_vma->rb_root will
7906d00c
AA
3595 * never see our bitflag.
3596 *
3597 * No need of atomic instructions here, head.next
3598 * can't change from under us until we release the
5a505085 3599 * anon_vma->root->rwsem.
7906d00c
AA
3600 */
3601 if (!__test_and_clear_bit(0, (unsigned long *)
f808c13f 3602 &anon_vma->root->rb_root.rb_root.rb_node))
7906d00c 3603 BUG();
08b52706 3604 anon_vma_unlock_write(anon_vma);
7906d00c
AA
3605 }
3606}
3607
3608static void vm_unlock_mapping(struct address_space *mapping)
3609{
3610 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3611 /*
3612 * AS_MM_ALL_LOCKS can't change to 0 from under us
3613 * because we hold the mm_all_locks_mutex.
3614 */
83cde9e8 3615 i_mmap_unlock_write(mapping);
7906d00c
AA
3616 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3617 &mapping->flags))
3618 BUG();
3619 }
3620}
3621
3622/*
c1e8d7c6 3623 * The mmap_lock cannot be released by the caller until
7906d00c
AA
3624 * mm_drop_all_locks() returns.
3625 */
3626void mm_drop_all_locks(struct mm_struct *mm)
3627{
3628 struct vm_area_struct *vma;
5beb4930 3629 struct anon_vma_chain *avc;
763ecb03 3630 MA_STATE(mas, &mm->mm_mt, 0, 0);
7906d00c 3631
325bca1f 3632 mmap_assert_write_locked(mm);
7906d00c
AA
3633 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3634
763ecb03 3635 mas_for_each(&mas, vma, ULONG_MAX) {
7906d00c 3636 if (vma->anon_vma)
5beb4930
RR
3637 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3638 vm_unlock_anon_vma(avc->anon_vma);
7906d00c
AA
3639 if (vma->vm_file && vma->vm_file->f_mapping)
3640 vm_unlock_mapping(vma->vm_file->f_mapping);
3641 }
3642
3643 mutex_unlock(&mm_all_locks_mutex);
3644}
8feae131
DH
3645
3646/*
3edf41d8 3647 * initialise the percpu counter for VM
8feae131
DH
3648 */
3649void __init mmap_init(void)
3650{
00a62ce9
KM
3651 int ret;
3652
908c7f19 3653 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
00a62ce9 3654 VM_BUG_ON(ret);
8feae131 3655}
c9b1d098
AS
3656
3657/*
3658 * Initialise sysctl_user_reserve_kbytes.
3659 *
3660 * This is intended to prevent a user from starting a single memory hogging
3661 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3662 * mode.
3663 *
3664 * The default value is min(3% of free memory, 128MB)
3665 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3666 */
1640879a 3667static int init_user_reserve(void)
c9b1d098
AS
3668{
3669 unsigned long free_kbytes;
3670
c41f012a 3671 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
c9b1d098
AS
3672
3673 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3674 return 0;
3675}
a64fb3cd 3676subsys_initcall(init_user_reserve);
4eeab4f5
AS
3677
3678/*
3679 * Initialise sysctl_admin_reserve_kbytes.
3680 *
3681 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3682 * to log in and kill a memory hogging process.
3683 *
3684 * Systems with more than 256MB will reserve 8MB, enough to recover
3685 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3686 * only reserve 3% of free pages by default.
3687 */
1640879a 3688static int init_admin_reserve(void)
4eeab4f5
AS
3689{
3690 unsigned long free_kbytes;
3691
c41f012a 3692 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
4eeab4f5
AS
3693
3694 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3695 return 0;
3696}
a64fb3cd 3697subsys_initcall(init_admin_reserve);
1640879a
AS
3698
3699/*
3700 * Reinititalise user and admin reserves if memory is added or removed.
3701 *
3702 * The default user reserve max is 128MB, and the default max for the
3703 * admin reserve is 8MB. These are usually, but not always, enough to
3704 * enable recovery from a memory hogging process using login/sshd, a shell,
3705 * and tools like top. It may make sense to increase or even disable the
3706 * reserve depending on the existence of swap or variations in the recovery
3707 * tools. So, the admin may have changed them.
3708 *
3709 * If memory is added and the reserves have been eliminated or increased above
3710 * the default max, then we'll trust the admin.
3711 *
3712 * If memory is removed and there isn't enough free memory, then we
3713 * need to reset the reserves.
3714 *
3715 * Otherwise keep the reserve set by the admin.
3716 */
3717static int reserve_mem_notifier(struct notifier_block *nb,
3718 unsigned long action, void *data)
3719{
3720 unsigned long tmp, free_kbytes;
3721
3722 switch (action) {
3723 case MEM_ONLINE:
3724 /* Default max is 128MB. Leave alone if modified by operator. */
3725 tmp = sysctl_user_reserve_kbytes;
3726 if (0 < tmp && tmp < (1UL << 17))
3727 init_user_reserve();
3728
3729 /* Default max is 8MB. Leave alone if modified by operator. */
3730 tmp = sysctl_admin_reserve_kbytes;
3731 if (0 < tmp && tmp < (1UL << 13))
3732 init_admin_reserve();
3733
3734 break;
3735 case MEM_OFFLINE:
c41f012a 3736 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1640879a
AS
3737
3738 if (sysctl_user_reserve_kbytes > free_kbytes) {
3739 init_user_reserve();
3740 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3741 sysctl_user_reserve_kbytes);
3742 }
3743
3744 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3745 init_admin_reserve();
3746 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3747 sysctl_admin_reserve_kbytes);
3748 }
3749 break;
3750 default:
3751 break;
3752 }
3753 return NOTIFY_OK;
3754}
3755
3756static struct notifier_block reserve_mem_nb = {
3757 .notifier_call = reserve_mem_notifier,
3758};
3759
3760static int __meminit init_reserve_notifier(void)
3761{
3762 if (register_hotmemory_notifier(&reserve_mem_nb))
b1de0d13 3763 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
1640879a
AS
3764
3765 return 0;
3766}
a64fb3cd 3767subsys_initcall(init_reserve_notifier);