Version:  2.0.40 2.2.26 2.4.37 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7

Linux/mm/mmap.c

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

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us