~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/mmap.c

Version: ~ [ linux-5.16 ] ~ [ linux-5.15.13 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.90 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.170 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.224 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.261 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.296 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.298 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/shmem_fs.h>
 29 #include <linux/profile.h>
 30 #include <linux/export.h>
 31 #include <linux/mount.h>
 32 #include <linux/mempolicy.h>
 33 #include <linux/rmap.h>
 34 #include <linux/mmu_notifier.h>
 35 #include <linux/mmdebug.h>
 36 #include <linux/perf_event.h>
 37 #include <linux/audit.h>
 38 #include <linux/khugepaged.h>
 39 #include <linux/uprobes.h>
 40 #include <linux/rbtree_augmented.h>
 41 #include <linux/notifier.h>
 42 #include <linux/memory.h>
 43 #include <linux/printk.h>
 44 #include <linux/userfaultfd_k.h>
 45 #include <linux/moduleparam.h>
 46 #include <linux/pkeys.h>
 47 #include <linux/oom.h>
 48 
 49 #include <linux/uaccess.h>
 50 #include <asm/cacheflush.h>
 51 #include <asm/tlb.h>
 52 #include <asm/mmu_context.h>
 53 
 54 #include "internal.h"
 55 
 56 #ifndef arch_mmap_check
 57 #define arch_mmap_check(addr, len, flags)       (0)
 58 #endif
 59 
 60 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
 61 const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
 62 const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
 63 int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
 64 #endif
 65 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
 66 const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
 67 const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
 68 int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
 69 #endif
 70 
 71 static bool ignore_rlimit_data;
 72 core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
 73 
 74 static void unmap_region(struct mm_struct *mm,
 75                 struct vm_area_struct *vma, struct vm_area_struct *prev,
 76                 unsigned long start, unsigned long end);
 77 
 78 /* description of effects of mapping type and prot in current implementation.
 79  * this is due to the limited x86 page protection hardware.  The expected
 80  * behavior is in parens:
 81  *
 82  * map_type     prot
 83  *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
 84  * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
 85  *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
 86  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
 87  *
 88  * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
 89  *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
 90  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
 91  *
 92  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
 93  * MAP_PRIVATE:
 94  *                                                              r: (no) no
 95  *                                                              w: (no) no
 96  *                                                              x: (yes) yes
 97  */
 98 pgprot_t protection_map[16] __ro_after_init = {
 99         __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
100         __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
101 };
102 
103 pgprot_t vm_get_page_prot(unsigned long vm_flags)
104 {
105         return __pgprot(pgprot_val(protection_map[vm_flags &
106                                 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
107                         pgprot_val(arch_vm_get_page_prot(vm_flags)));
108 }
109 EXPORT_SYMBOL(vm_get_page_prot);
110 
111 static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
112 {
113         return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
114 }
115 
116 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
117 void vma_set_page_prot(struct vm_area_struct *vma)
118 {
119         unsigned long vm_flags = vma->vm_flags;
120         pgprot_t vm_page_prot;
121 
122         vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
123         if (vma_wants_writenotify(vma, vm_page_prot)) {
124                 vm_flags &= ~VM_SHARED;
125                 vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
126         }
127         /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
128         WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
129 }
130 
131 /*
132  * Requires inode->i_mapping->i_mmap_rwsem
133  */
134 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
135                 struct file *file, struct address_space *mapping)
136 {
137         if (vma->vm_flags & VM_DENYWRITE)
138                 atomic_inc(&file_inode(file)->i_writecount);
139         if (vma->vm_flags & VM_SHARED)
140                 mapping_unmap_writable(mapping);
141 
142         flush_dcache_mmap_lock(mapping);
143         vma_interval_tree_remove(vma, &mapping->i_mmap);
144         flush_dcache_mmap_unlock(mapping);
145 }
146 
147 /*
148  * Unlink a file-based vm structure from its interval tree, to hide
149  * vma from rmap and vmtruncate before freeing its page tables.
150  */
151 void unlink_file_vma(struct vm_area_struct *vma)
152 {
153         struct file *file = vma->vm_file;
154 
155         if (file) {
156                 struct address_space *mapping = file->f_mapping;
157                 i_mmap_lock_write(mapping);
158                 __remove_shared_vm_struct(vma, file, mapping);
159                 i_mmap_unlock_write(mapping);
160         }
161 }
162 
163 /*
164  * Close a vm structure and free it, returning the next.
165  */
166 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
167 {
168         struct vm_area_struct *next = vma->vm_next;
169 
170         might_sleep();
171         if (vma->vm_ops && vma->vm_ops->close)
172                 vma->vm_ops->close(vma);
173         if (vma->vm_file)
174                 fput(vma->vm_file);
175         mpol_put(vma_policy(vma));
176         kmem_cache_free(vm_area_cachep, vma);
177         return next;
178 }
179 
180 static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf);
181 
182 SYSCALL_DEFINE1(brk, unsigned long, brk)
183 {
184         unsigned long retval;
185         unsigned long newbrk, oldbrk;
186         struct mm_struct *mm = current->mm;
187         struct vm_area_struct *next;
188         unsigned long min_brk;
189         bool populate;
190         LIST_HEAD(uf);
191 
192         if (down_write_killable(&mm->mmap_sem))
193                 return -EINTR;
194 
195 #ifdef CONFIG_COMPAT_BRK
196         /*
197          * CONFIG_COMPAT_BRK can still be overridden by setting
198          * randomize_va_space to 2, which will still cause mm->start_brk
199          * to be arbitrarily shifted
200          */
201         if (current->brk_randomized)
202                 min_brk = mm->start_brk;
203         else
204                 min_brk = mm->end_data;
205 #else
206         min_brk = mm->start_brk;
207 #endif
208         if (brk < min_brk)
209                 goto out;
210 
211         /*
212          * Check against rlimit here. If this check is done later after the test
213          * of oldbrk with newbrk then it can escape the test and let the data
214          * segment grow beyond its set limit the in case where the limit is
215          * not page aligned -Ram Gupta
216          */
217         if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
218                               mm->end_data, mm->start_data))
219                 goto out;
220 
221         newbrk = PAGE_ALIGN(brk);
222         oldbrk = PAGE_ALIGN(mm->brk);
223         if (oldbrk == newbrk)
224                 goto set_brk;
225 
226         /* Always allow shrinking brk. */
227         if (brk <= mm->brk) {
228                 if (!do_munmap(mm, newbrk, oldbrk-newbrk, &uf))
229                         goto set_brk;
230                 goto out;
231         }
232 
233         /* Check against existing mmap mappings. */
234         next = find_vma(mm, oldbrk);
235         if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
236                 goto out;
237 
238         /* Ok, looks good - let it rip. */
239         if (do_brk(oldbrk, newbrk-oldbrk, &uf) < 0)
240                 goto out;
241 
242 set_brk:
243         mm->brk = brk;
244         populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
245         up_write(&mm->mmap_sem);
246         userfaultfd_unmap_complete(mm, &uf);
247         if (populate)
248                 mm_populate(oldbrk, newbrk - oldbrk);
249         return brk;
250 
251 out:
252         retval = mm->brk;
253         up_write(&mm->mmap_sem);
254         return retval;
255 }
256 
257 static long vma_compute_subtree_gap(struct vm_area_struct *vma)
258 {
259         unsigned long max, prev_end, subtree_gap;
260 
261         /*
262          * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
263          * allow two stack_guard_gaps between them here, and when choosing
264          * an unmapped area; whereas when expanding we only require one.
265          * That's a little inconsistent, but keeps the code here simpler.
266          */
267         max = vm_start_gap(vma);
268         if (vma->vm_prev) {
269                 prev_end = vm_end_gap(vma->vm_prev);
270                 if (max > prev_end)
271                         max -= prev_end;
272                 else
273                         max = 0;
274         }
275         if (vma->vm_rb.rb_left) {
276                 subtree_gap = rb_entry(vma->vm_rb.rb_left,
277                                 struct vm_area_struct, vm_rb)->rb_subtree_gap;
278                 if (subtree_gap > max)
279                         max = subtree_gap;
280         }
281         if (vma->vm_rb.rb_right) {
282                 subtree_gap = rb_entry(vma->vm_rb.rb_right,
283                                 struct vm_area_struct, vm_rb)->rb_subtree_gap;
284                 if (subtree_gap > max)
285                         max = subtree_gap;
286         }
287         return max;
288 }
289 
290 #ifdef CONFIG_DEBUG_VM_RB
291 static int browse_rb(struct mm_struct *mm)
292 {
293         struct rb_root *root = &mm->mm_rb;
294         int i = 0, j, bug = 0;
295         struct rb_node *nd, *pn = NULL;
296         unsigned long prev = 0, pend = 0;
297 
298         for (nd = rb_first(root); nd; nd = rb_next(nd)) {
299                 struct vm_area_struct *vma;
300                 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
301                 if (vma->vm_start < prev) {
302                         pr_emerg("vm_start %lx < prev %lx\n",
303                                   vma->vm_start, prev);
304                         bug = 1;
305                 }
306                 if (vma->vm_start < pend) {
307                         pr_emerg("vm_start %lx < pend %lx\n",
308                                   vma->vm_start, pend);
309                         bug = 1;
310                 }
311                 if (vma->vm_start > vma->vm_end) {
312                         pr_emerg("vm_start %lx > vm_end %lx\n",
313                                   vma->vm_start, vma->vm_end);
314                         bug = 1;
315                 }
316                 spin_lock(&mm->page_table_lock);
317                 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
318                         pr_emerg("free gap %lx, correct %lx\n",
319                                vma->rb_subtree_gap,
320                                vma_compute_subtree_gap(vma));
321                         bug = 1;
322                 }
323                 spin_unlock(&mm->page_table_lock);
324                 i++;
325                 pn = nd;
326                 prev = vma->vm_start;
327                 pend = vma->vm_end;
328         }
329         j = 0;
330         for (nd = pn; nd; nd = rb_prev(nd))
331                 j++;
332         if (i != j) {
333                 pr_emerg("backwards %d, forwards %d\n", j, i);
334                 bug = 1;
335         }
336         return bug ? -1 : i;
337 }
338 
339 static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
340 {
341         struct rb_node *nd;
342 
343         for (nd = rb_first(root); nd; nd = rb_next(nd)) {
344                 struct vm_area_struct *vma;
345                 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
346                 VM_BUG_ON_VMA(vma != ignore &&
347                         vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
348                         vma);
349         }
350 }
351 
352 static void validate_mm(struct mm_struct *mm)
353 {
354         int bug = 0;
355         int i = 0;
356         unsigned long highest_address = 0;
357         struct vm_area_struct *vma = mm->mmap;
358 
359         while (vma) {
360                 struct anon_vma *anon_vma = vma->anon_vma;
361                 struct anon_vma_chain *avc;
362 
363                 if (anon_vma) {
364                         anon_vma_lock_read(anon_vma);
365                         list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
366                                 anon_vma_interval_tree_verify(avc);
367                         anon_vma_unlock_read(anon_vma);
368                 }
369 
370                 highest_address = vm_end_gap(vma);
371                 vma = vma->vm_next;
372                 i++;
373         }
374         if (i != mm->map_count) {
375                 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
376                 bug = 1;
377         }
378         if (highest_address != mm->highest_vm_end) {
379                 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
380                           mm->highest_vm_end, highest_address);
381                 bug = 1;
382         }
383         i = browse_rb(mm);
384         if (i != mm->map_count) {
385                 if (i != -1)
386                         pr_emerg("map_count %d rb %d\n", mm->map_count, i);
387                 bug = 1;
388         }
389         VM_BUG_ON_MM(bug, mm);
390 }
391 #else
392 #define validate_mm_rb(root, ignore) do { } while (0)
393 #define validate_mm(mm) do { } while (0)
394 #endif
395 
396 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
397                      unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
398 
399 /*
400  * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
401  * vma->vm_prev->vm_end values changed, without modifying the vma's position
402  * in the rbtree.
403  */
404 static void vma_gap_update(struct vm_area_struct *vma)
405 {
406         /*
407          * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
408          * function that does exacltly what we want.
409          */
410         vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
411 }
412 
413 static inline void vma_rb_insert(struct vm_area_struct *vma,
414                                  struct rb_root *root)
415 {
416         /* All rb_subtree_gap values must be consistent prior to insertion */
417         validate_mm_rb(root, NULL);
418 
419         rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
420 }
421 
422 static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
423 {
424         /*
425          * Note rb_erase_augmented is a fairly large inline function,
426          * so make sure we instantiate it only once with our desired
427          * augmented rbtree callbacks.
428          */
429         rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
430 }
431 
432 static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
433                                                 struct rb_root *root,
434                                                 struct vm_area_struct *ignore)
435 {
436         /*
437          * All rb_subtree_gap values must be consistent prior to erase,
438          * with the possible exception of the "next" vma being erased if
439          * next->vm_start was reduced.
440          */
441         validate_mm_rb(root, ignore);
442 
443         __vma_rb_erase(vma, root);
444 }
445 
446 static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
447                                          struct rb_root *root)
448 {
449         /*
450          * All rb_subtree_gap values must be consistent prior to erase,
451          * with the possible exception of the vma being erased.
452          */
453         validate_mm_rb(root, vma);
454 
455         __vma_rb_erase(vma, root);
456 }
457 
458 /*
459  * vma has some anon_vma assigned, and is already inserted on that
460  * anon_vma's interval trees.
461  *
462  * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
463  * vma must be removed from the anon_vma's interval trees using
464  * anon_vma_interval_tree_pre_update_vma().
465  *
466  * After the update, the vma will be reinserted using
467  * anon_vma_interval_tree_post_update_vma().
468  *
469  * The entire update must be protected by exclusive mmap_sem and by
470  * the root anon_vma's mutex.
471  */
472 static inline void
473 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
474 {
475         struct anon_vma_chain *avc;
476 
477         list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
478                 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
479 }
480 
481 static inline void
482 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
483 {
484         struct anon_vma_chain *avc;
485 
486         list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
487                 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
488 }
489 
490 static int find_vma_links(struct mm_struct *mm, unsigned long addr,
491                 unsigned long end, struct vm_area_struct **pprev,
492                 struct rb_node ***rb_link, struct rb_node **rb_parent)
493 {
494         struct rb_node **__rb_link, *__rb_parent, *rb_prev;
495 
496         __rb_link = &mm->mm_rb.rb_node;
497         rb_prev = __rb_parent = NULL;
498 
499         while (*__rb_link) {
500                 struct vm_area_struct *vma_tmp;
501 
502                 __rb_parent = *__rb_link;
503                 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
504 
505                 if (vma_tmp->vm_end > addr) {
506                         /* Fail if an existing vma overlaps the area */
507                         if (vma_tmp->vm_start < end)
508                                 return -ENOMEM;
509                         __rb_link = &__rb_parent->rb_left;
510                 } else {
511                         rb_prev = __rb_parent;
512                         __rb_link = &__rb_parent->rb_right;
513                 }
514         }
515 
516         *pprev = NULL;
517         if (rb_prev)
518                 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
519         *rb_link = __rb_link;
520         *rb_parent = __rb_parent;
521         return 0;
522 }
523 
524 static unsigned long count_vma_pages_range(struct mm_struct *mm,
525                 unsigned long addr, unsigned long end)
526 {
527         unsigned long nr_pages = 0;
528         struct vm_area_struct *vma;
529 
530         /* Find first overlaping mapping */
531         vma = find_vma_intersection(mm, addr, end);
532         if (!vma)
533                 return 0;
534 
535         nr_pages = (min(end, vma->vm_end) -
536                 max(addr, vma->vm_start)) >> PAGE_SHIFT;
537 
538         /* Iterate over the rest of the overlaps */
539         for (vma = vma->vm_next; vma; vma = vma->vm_next) {
540                 unsigned long overlap_len;
541 
542                 if (vma->vm_start > end)
543                         break;
544 
545                 overlap_len = min(end, vma->vm_end) - vma->vm_start;
546                 nr_pages += overlap_len >> PAGE_SHIFT;
547         }
548 
549         return nr_pages;
550 }
551 
552 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
553                 struct rb_node **rb_link, struct rb_node *rb_parent)
554 {
555         /* Update tracking information for the gap following the new vma. */
556         if (vma->vm_next)
557                 vma_gap_update(vma->vm_next);
558         else
559                 mm->highest_vm_end = vm_end_gap(vma);
560 
561         /*
562          * vma->vm_prev wasn't known when we followed the rbtree to find the
563          * correct insertion point for that vma. As a result, we could not
564          * update the vma vm_rb parents rb_subtree_gap values on the way down.
565          * So, we first insert the vma with a zero rb_subtree_gap value
566          * (to be consistent with what we did on the way down), and then
567          * immediately update the gap to the correct value. Finally we
568          * rebalance the rbtree after all augmented values have been set.
569          */
570         rb_link_node(&vma->vm_rb, rb_parent, rb_link);
571         vma->rb_subtree_gap = 0;
572         vma_gap_update(vma);
573         vma_rb_insert(vma, &mm->mm_rb);
574 }
575 
576 static void __vma_link_file(struct vm_area_struct *vma)
577 {
578         struct file *file;
579 
580         file = vma->vm_file;
581         if (file) {
582                 struct address_space *mapping = file->f_mapping;
583 
584                 if (vma->vm_flags & VM_DENYWRITE)
585                         atomic_dec(&file_inode(file)->i_writecount);
586                 if (vma->vm_flags & VM_SHARED)
587                         atomic_inc(&mapping->i_mmap_writable);
588 
589                 flush_dcache_mmap_lock(mapping);
590                 vma_interval_tree_insert(vma, &mapping->i_mmap);
591                 flush_dcache_mmap_unlock(mapping);
592         }
593 }
594 
595 static void
596 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
597         struct vm_area_struct *prev, struct rb_node **rb_link,
598         struct rb_node *rb_parent)
599 {
600         __vma_link_list(mm, vma, prev, rb_parent);
601         __vma_link_rb(mm, vma, rb_link, rb_parent);
602 }
603 
604 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
605                         struct vm_area_struct *prev, struct rb_node **rb_link,
606                         struct rb_node *rb_parent)
607 {
608         struct address_space *mapping = NULL;
609 
610         if (vma->vm_file) {
611                 mapping = vma->vm_file->f_mapping;
612                 i_mmap_lock_write(mapping);
613         }
614 
615         __vma_link(mm, vma, prev, rb_link, rb_parent);
616         __vma_link_file(vma);
617 
618         if (mapping)
619                 i_mmap_unlock_write(mapping);
620 
621         mm->map_count++;
622         validate_mm(mm);
623 }
624 
625 /*
626  * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
627  * mm's list and rbtree.  It has already been inserted into the interval tree.
628  */
629 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
630 {
631         struct vm_area_struct *prev;
632         struct rb_node **rb_link, *rb_parent;
633 
634         if (find_vma_links(mm, vma->vm_start, vma->vm_end,
635                            &prev, &rb_link, &rb_parent))
636                 BUG();
637         __vma_link(mm, vma, prev, rb_link, rb_parent);
638         mm->map_count++;
639 }
640 
641 static __always_inline void __vma_unlink_common(struct mm_struct *mm,
642                                                 struct vm_area_struct *vma,
643                                                 struct vm_area_struct *prev,
644                                                 bool has_prev,
645                                                 struct vm_area_struct *ignore)
646 {
647         struct vm_area_struct *next;
648 
649         vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
650         next = vma->vm_next;
651         if (has_prev)
652                 prev->vm_next = next;
653         else {
654                 prev = vma->vm_prev;
655                 if (prev)
656                         prev->vm_next = next;
657                 else
658                         mm->mmap = next;
659         }
660         if (next)
661                 next->vm_prev = prev;
662 
663         /* Kill the cache */
664         vmacache_invalidate(mm);
665 }
666 
667 static inline void __vma_unlink_prev(struct mm_struct *mm,
668                                      struct vm_area_struct *vma,
669                                      struct vm_area_struct *prev)
670 {
671         __vma_unlink_common(mm, vma, prev, true, vma);
672 }
673 
674 /*
675  * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
676  * is already present in an i_mmap tree without adjusting the tree.
677  * The following helper function should be used when such adjustments
678  * are necessary.  The "insert" vma (if any) is to be inserted
679  * before we drop the necessary locks.
680  */
681 int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
682         unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
683         struct vm_area_struct *expand)
684 {
685         struct mm_struct *mm = vma->vm_mm;
686         struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
687         struct address_space *mapping = NULL;
688         struct rb_root_cached *root = NULL;
689         struct anon_vma *anon_vma = NULL;
690         struct file *file = vma->vm_file;
691         bool start_changed = false, end_changed = false;
692         long adjust_next = 0;
693         int remove_next = 0;
694 
695         if (next && !insert) {
696                 struct vm_area_struct *exporter = NULL, *importer = NULL;
697 
698                 if (end >= next->vm_end) {
699                         /*
700                          * vma expands, overlapping all the next, and
701                          * perhaps the one after too (mprotect case 6).
702                          * The only other cases that gets here are
703                          * case 1, case 7 and case 8.
704                          */
705                         if (next == expand) {
706                                 /*
707                                  * The only case where we don't expand "vma"
708                                  * and we expand "next" instead is case 8.
709                                  */
710                                 VM_WARN_ON(end != next->vm_end);
711                                 /*
712                                  * remove_next == 3 means we're
713                                  * removing "vma" and that to do so we
714                                  * swapped "vma" and "next".
715                                  */
716                                 remove_next = 3;
717                                 VM_WARN_ON(file != next->vm_file);
718                                 swap(vma, next);
719                         } else {
720                                 VM_WARN_ON(expand != vma);
721                                 /*
722                                  * case 1, 6, 7, remove_next == 2 is case 6,
723                                  * remove_next == 1 is case 1 or 7.
724                                  */
725                                 remove_next = 1 + (end > next->vm_end);
726                                 VM_WARN_ON(remove_next == 2 &&
727                                            end != next->vm_next->vm_end);
728                                 VM_WARN_ON(remove_next == 1 &&
729                                            end != next->vm_end);
730                                 /* trim end to next, for case 6 first pass */
731                                 end = next->vm_end;
732                         }
733 
734                         exporter = next;
735                         importer = vma;
736 
737                         /*
738                          * If next doesn't have anon_vma, import from vma after
739                          * next, if the vma overlaps with it.
740                          */
741                         if (remove_next == 2 && !next->anon_vma)
742                                 exporter = next->vm_next;
743 
744                 } else if (end > next->vm_start) {
745                         /*
746                          * vma expands, overlapping part of the next:
747                          * mprotect case 5 shifting the boundary up.
748                          */
749                         adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
750                         exporter = next;
751                         importer = vma;
752                         VM_WARN_ON(expand != importer);
753                 } else if (end < vma->vm_end) {
754                         /*
755                          * vma shrinks, and !insert tells it's not
756                          * split_vma inserting another: so it must be
757                          * mprotect case 4 shifting the boundary down.
758                          */
759                         adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
760                         exporter = vma;
761                         importer = next;
762                         VM_WARN_ON(expand != importer);
763                 }
764 
765                 /*
766                  * Easily overlooked: when mprotect shifts the boundary,
767                  * make sure the expanding vma has anon_vma set if the
768                  * shrinking vma had, to cover any anon pages imported.
769                  */
770                 if (exporter && exporter->anon_vma && !importer->anon_vma) {
771                         int error;
772 
773                         importer->anon_vma = exporter->anon_vma;
774                         error = anon_vma_clone(importer, exporter);
775                         if (error)
776                                 return error;
777                 }
778         }
779 again:
780         vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
781 
782         if (file) {
783                 mapping = file->f_mapping;
784                 root = &mapping->i_mmap;
785                 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
786 
787                 if (adjust_next)
788                         uprobe_munmap(next, next->vm_start, next->vm_end);
789 
790                 i_mmap_lock_write(mapping);
791                 if (insert) {
792                         /*
793                          * Put into interval tree now, so instantiated pages
794                          * are visible to arm/parisc __flush_dcache_page
795                          * throughout; but we cannot insert into address
796                          * space until vma start or end is updated.
797                          */
798                         __vma_link_file(insert);
799                 }
800         }
801 
802         anon_vma = vma->anon_vma;
803         if (!anon_vma && adjust_next)
804                 anon_vma = next->anon_vma;
805         if (anon_vma) {
806                 VM_WARN_ON(adjust_next && next->anon_vma &&
807                            anon_vma != next->anon_vma);
808                 anon_vma_lock_write(anon_vma);
809                 anon_vma_interval_tree_pre_update_vma(vma);
810                 if (adjust_next)
811                         anon_vma_interval_tree_pre_update_vma(next);
812         }
813 
814         if (root) {
815                 flush_dcache_mmap_lock(mapping);
816                 vma_interval_tree_remove(vma, root);
817                 if (adjust_next)
818                         vma_interval_tree_remove(next, root);
819         }
820 
821         if (start != vma->vm_start) {
822                 vma->vm_start = start;
823                 start_changed = true;
824         }
825         if (end != vma->vm_end) {
826                 vma->vm_end = end;
827                 end_changed = true;
828         }
829         vma->vm_pgoff = pgoff;
830         if (adjust_next) {
831                 next->vm_start += adjust_next << PAGE_SHIFT;
832                 next->vm_pgoff += adjust_next;
833         }
834 
835         if (root) {
836                 if (adjust_next)
837                         vma_interval_tree_insert(next, root);
838                 vma_interval_tree_insert(vma, root);
839                 flush_dcache_mmap_unlock(mapping);
840         }
841 
842         if (remove_next) {
843                 /*
844                  * vma_merge has merged next into vma, and needs
845                  * us to remove next before dropping the locks.
846                  */
847                 if (remove_next != 3)
848                         __vma_unlink_prev(mm, next, vma);
849                 else
850                         /*
851                          * vma is not before next if they've been
852                          * swapped.
853                          *
854                          * pre-swap() next->vm_start was reduced so
855                          * tell validate_mm_rb to ignore pre-swap()
856                          * "next" (which is stored in post-swap()
857                          * "vma").
858                          */
859                         __vma_unlink_common(mm, next, NULL, false, vma);
860                 if (file)
861                         __remove_shared_vm_struct(next, file, mapping);
862         } else if (insert) {
863                 /*
864                  * split_vma has split insert from vma, and needs
865                  * us to insert it before dropping the locks
866                  * (it may either follow vma or precede it).
867                  */
868                 __insert_vm_struct(mm, insert);
869         } else {
870                 if (start_changed)
871                         vma_gap_update(vma);
872                 if (end_changed) {
873                         if (!next)
874                                 mm->highest_vm_end = vm_end_gap(vma);
875                         else if (!adjust_next)
876                                 vma_gap_update(next);
877                 }
878         }
879 
880         if (anon_vma) {
881                 anon_vma_interval_tree_post_update_vma(vma);
882                 if (adjust_next)
883                         anon_vma_interval_tree_post_update_vma(next);
884                 anon_vma_unlock_write(anon_vma);
885         }
886         if (mapping)
887                 i_mmap_unlock_write(mapping);
888 
889         if (root) {
890                 uprobe_mmap(vma);
891 
892                 if (adjust_next)
893                         uprobe_mmap(next);
894         }
895 
896         if (remove_next) {
897                 if (file) {
898                         uprobe_munmap(next, next->vm_start, next->vm_end);
899                         fput(file);
900                 }
901                 if (next->anon_vma)
902                         anon_vma_merge(vma, next);
903                 mm->map_count--;
904                 mpol_put(vma_policy(next));
905                 kmem_cache_free(vm_area_cachep, next);
906                 /*
907                  * In mprotect's case 6 (see comments on vma_merge),
908                  * we must remove another next too. It would clutter
909                  * up the code too much to do both in one go.
910                  */
911                 if (remove_next != 3) {
912                         /*
913                          * If "next" was removed and vma->vm_end was
914                          * expanded (up) over it, in turn
915                          * "next->vm_prev->vm_end" changed and the
916                          * "vma->vm_next" gap must be updated.
917                          */
918                         next = vma->vm_next;
919                 } else {
920                         /*
921                          * For the scope of the comment "next" and
922                          * "vma" considered pre-swap(): if "vma" was
923                          * removed, next->vm_start was expanded (down)
924                          * over it and the "next" gap must be updated.
925                          * Because of the swap() the post-swap() "vma"
926                          * actually points to pre-swap() "next"
927                          * (post-swap() "next" as opposed is now a
928                          * dangling pointer).
929                          */
930                         next = vma;
931                 }
932                 if (remove_next == 2) {
933                         remove_next = 1;
934                         end = next->vm_end;
935                         goto again;
936                 }
937                 else if (next)
938                         vma_gap_update(next);
939                 else {
940                         /*
941                          * If remove_next == 2 we obviously can't
942                          * reach this path.
943                          *
944                          * If remove_next == 3 we can't reach this
945                          * path because pre-swap() next is always not
946                          * NULL. pre-swap() "next" is not being
947                          * removed and its next->vm_end is not altered
948                          * (and furthermore "end" already matches
949                          * next->vm_end in remove_next == 3).
950                          *
951                          * We reach this only in the remove_next == 1
952                          * case if the "next" vma that was removed was
953                          * the highest vma of the mm. However in such
954                          * case next->vm_end == "end" and the extended
955                          * "vma" has vma->vm_end == next->vm_end so
956                          * mm->highest_vm_end doesn't need any update
957                          * in remove_next == 1 case.
958                          */
959                         VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
960                 }
961         }
962         if (insert && file)
963                 uprobe_mmap(insert);
964 
965         validate_mm(mm);
966 
967         return 0;
968 }
969 
970 /*
971  * If the vma has a ->close operation then the driver probably needs to release
972  * per-vma resources, so we don't attempt to merge those.
973  */
974 static inline int is_mergeable_vma(struct vm_area_struct *vma,
975                                 struct file *file, unsigned long vm_flags,
976                                 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
977 {
978         /*
979          * VM_SOFTDIRTY should not prevent from VMA merging, if we
980          * match the flags but dirty bit -- the caller should mark
981          * merged VMA as dirty. If dirty bit won't be excluded from
982          * comparison, we increase pressue on the memory system forcing
983          * the kernel to generate new VMAs when old one could be
984          * extended instead.
985          */
986         if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
987                 return 0;
988         if (vma->vm_file != file)
989                 return 0;
990         if (vma->vm_ops && vma->vm_ops->close)
991                 return 0;
992         if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
993                 return 0;
994         return 1;
995 }
996 
997 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
998                                         struct anon_vma *anon_vma2,
999                                         struct vm_area_struct *vma)
1000 {
1001         /*
1002          * The list_is_singular() test is to avoid merging VMA cloned from
1003          * parents. This can improve scalability caused by anon_vma lock.
1004          */
1005         if ((!anon_vma1 || !anon_vma2) && (!vma ||
1006                 list_is_singular(&vma->anon_vma_chain)))
1007                 return 1;
1008         return anon_vma1 == anon_vma2;
1009 }
1010 
1011 /*
1012  * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1013  * in front of (at a lower virtual address and file offset than) the vma.
1014  *
1015  * We cannot merge two vmas if they have differently assigned (non-NULL)
1016  * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1017  *
1018  * We don't check here for the merged mmap wrapping around the end of pagecache
1019  * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1020  * wrap, nor mmaps which cover the final page at index -1UL.
1021  */
1022 static int
1023 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
1024                      struct anon_vma *anon_vma, struct file *file,
1025                      pgoff_t vm_pgoff,
1026                      struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1027 {
1028         if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
1029             is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1030                 if (vma->vm_pgoff == vm_pgoff)
1031                         return 1;
1032         }
1033         return 0;
1034 }
1035 
1036 /*
1037  * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1038  * beyond (at a higher virtual address and file offset than) the vma.
1039  *
1040  * We cannot merge two vmas if they have differently assigned (non-NULL)
1041  * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1042  */
1043 static int
1044 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
1045                     struct anon_vma *anon_vma, struct file *file,
1046                     pgoff_t vm_pgoff,
1047                     struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1048 {
1049         if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
1050             is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1051                 pgoff_t vm_pglen;
1052                 vm_pglen = vma_pages(vma);
1053                 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1054                         return 1;
1055         }
1056         return 0;
1057 }
1058 
1059 /*
1060  * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1061  * whether that can be merged with its predecessor or its successor.
1062  * Or both (it neatly fills a hole).
1063  *
1064  * In most cases - when called for mmap, brk or mremap - [addr,end) is
1065  * certain not to be mapped by the time vma_merge is called; but when
1066  * called for mprotect, it is certain to be already mapped (either at
1067  * an offset within prev, or at the start of next), and the flags of
1068  * this area are about to be changed to vm_flags - and the no-change
1069  * case has already been eliminated.
1070  *
1071  * The following mprotect cases have to be considered, where AAAA is
1072  * the area passed down from mprotect_fixup, never extending beyond one
1073  * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1074  *
1075  *     AAAA             AAAA                AAAA          AAAA
1076  *    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPNNNNXXXX
1077  *    cannot merge    might become    might become    might become
1078  *                    PPNNNNNNNNNN    PPPPPPPPPPNN    PPPPPPPPPPPP 6 or
1079  *    mmap, brk or    case 4 below    case 5 below    PPPPPPPPXXXX 7 or
1080  *    mremap move:                                    PPPPXXXXXXXX 8
1081  *        AAAA
1082  *    PPPP    NNNN    PPPPPPPPPPPP    PPPPPPPPNNNN    PPPPNNNNNNNN
1083  *    might become    case 1 below    case 2 below    case 3 below
1084  *
1085  * It is important for case 8 that the the vma NNNN overlapping the
1086  * region AAAA is never going to extended over XXXX. Instead XXXX must
1087  * be extended in region AAAA and NNNN must be removed. This way in
1088  * all cases where vma_merge succeeds, the moment vma_adjust drops the
1089  * rmap_locks, the properties of the merged vma will be already
1090  * correct for the whole merged range. Some of those properties like
1091  * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1092  * be correct for the whole merged range immediately after the
1093  * rmap_locks are released. Otherwise if XXXX would be removed and
1094  * NNNN would be extended over the XXXX range, remove_migration_ptes
1095  * or other rmap walkers (if working on addresses beyond the "end"
1096  * parameter) may establish ptes with the wrong permissions of NNNN
1097  * instead of the right permissions of XXXX.
1098  */
1099 struct vm_area_struct *vma_merge(struct mm_struct *mm,
1100                         struct vm_area_struct *prev, unsigned long addr,
1101                         unsigned long end, unsigned long vm_flags,
1102                         struct anon_vma *anon_vma, struct file *file,
1103                         pgoff_t pgoff, struct mempolicy *policy,
1104                         struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1105 {
1106         pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1107         struct vm_area_struct *area, *next;
1108         int err;
1109 
1110         /*
1111          * We later require that vma->vm_flags == vm_flags,
1112          * so this tests vma->vm_flags & VM_SPECIAL, too.
1113          */
1114         if (vm_flags & VM_SPECIAL)
1115                 return NULL;
1116 
1117         if (prev)
1118                 next = prev->vm_next;
1119         else
1120                 next = mm->mmap;
1121         area = next;
1122         if (area && area->vm_end == end)                /* cases 6, 7, 8 */
1123                 next = next->vm_next;
1124 
1125         /* verify some invariant that must be enforced by the caller */
1126         VM_WARN_ON(prev && addr <= prev->vm_start);
1127         VM_WARN_ON(area && end > area->vm_end);
1128         VM_WARN_ON(addr >= end);
1129 
1130         /*
1131          * Can it merge with the predecessor?
1132          */
1133         if (prev && prev->vm_end == addr &&
1134                         mpol_equal(vma_policy(prev), policy) &&
1135                         can_vma_merge_after(prev, vm_flags,
1136                                             anon_vma, file, pgoff,
1137                                             vm_userfaultfd_ctx)) {
1138                 /*
1139                  * OK, it can.  Can we now merge in the successor as well?
1140                  */
1141                 if (next && end == next->vm_start &&
1142                                 mpol_equal(policy, vma_policy(next)) &&
1143                                 can_vma_merge_before(next, vm_flags,
1144                                                      anon_vma, file,
1145                                                      pgoff+pglen,
1146                                                      vm_userfaultfd_ctx) &&
1147                                 is_mergeable_anon_vma(prev->anon_vma,
1148                                                       next->anon_vma, NULL)) {
1149                                                         /* cases 1, 6 */
1150                         err = __vma_adjust(prev, prev->vm_start,
1151                                          next->vm_end, prev->vm_pgoff, NULL,
1152                                          prev);
1153                 } else                                  /* cases 2, 5, 7 */
1154                         err = __vma_adjust(prev, prev->vm_start,
1155                                          end, prev->vm_pgoff, NULL, prev);
1156                 if (err)
1157                         return NULL;
1158                 khugepaged_enter_vma_merge(prev, vm_flags);
1159                 return prev;
1160         }
1161 
1162         /*
1163          * Can this new request be merged in front of next?
1164          */
1165         if (next && end == next->vm_start &&
1166                         mpol_equal(policy, vma_policy(next)) &&
1167                         can_vma_merge_before(next, vm_flags,
1168                                              anon_vma, file, pgoff+pglen,
1169                                              vm_userfaultfd_ctx)) {
1170                 if (prev && addr < prev->vm_end)        /* case 4 */
1171                         err = __vma_adjust(prev, prev->vm_start,
1172                                          addr, prev->vm_pgoff, NULL, next);
1173                 else {                                  /* cases 3, 8 */
1174                         err = __vma_adjust(area, addr, next->vm_end,
1175                                          next->vm_pgoff - pglen, NULL, next);
1176                         /*
1177                          * In case 3 area is already equal to next and
1178                          * this is a noop, but in case 8 "area" has
1179                          * been removed and next was expanded over it.
1180                          */
1181                         area = next;
1182                 }
1183                 if (err)
1184                         return NULL;
1185                 khugepaged_enter_vma_merge(area, vm_flags);
1186                 return area;
1187         }
1188 
1189         return NULL;
1190 }
1191 
1192 /*
1193  * Rough compatbility check to quickly see if it's even worth looking
1194  * at sharing an anon_vma.
1195  *
1196  * They need to have the same vm_file, and the flags can only differ
1197  * in things that mprotect may change.
1198  *
1199  * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1200  * we can merge the two vma's. For example, we refuse to merge a vma if
1201  * there is a vm_ops->close() function, because that indicates that the
1202  * driver is doing some kind of reference counting. But that doesn't
1203  * really matter for the anon_vma sharing case.
1204  */
1205 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1206 {
1207         return a->vm_end == b->vm_start &&
1208                 mpol_equal(vma_policy(a), vma_policy(b)) &&
1209                 a->vm_file == b->vm_file &&
1210                 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
1211                 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1212 }
1213 
1214 /*
1215  * Do some basic sanity checking to see if we can re-use the anon_vma
1216  * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1217  * the same as 'old', the other will be the new one that is trying
1218  * to share the anon_vma.
1219  *
1220  * NOTE! This runs with mm_sem held for reading, so it is possible that
1221  * the anon_vma of 'old' is concurrently in the process of being set up
1222  * by another page fault trying to merge _that_. But that's ok: if it
1223  * is being set up, that automatically means that it will be a singleton
1224  * acceptable for merging, so we can do all of this optimistically. But
1225  * we do that READ_ONCE() to make sure that we never re-load the pointer.
1226  *
1227  * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1228  * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1229  * is to return an anon_vma that is "complex" due to having gone through
1230  * a fork).
1231  *
1232  * We also make sure that the two vma's are compatible (adjacent,
1233  * and with the same memory policies). That's all stable, even with just
1234  * a read lock on the mm_sem.
1235  */
1236 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1237 {
1238         if (anon_vma_compatible(a, b)) {
1239                 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1240 
1241                 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1242                         return anon_vma;
1243         }
1244         return NULL;
1245 }
1246 
1247 /*
1248  * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1249  * neighbouring vmas for a suitable anon_vma, before it goes off
1250  * to allocate a new anon_vma.  It checks because a repetitive
1251  * sequence of mprotects and faults may otherwise lead to distinct
1252  * anon_vmas being allocated, preventing vma merge in subsequent
1253  * mprotect.
1254  */
1255 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1256 {
1257         struct anon_vma *anon_vma;
1258         struct vm_area_struct *near;
1259 
1260         near = vma->vm_next;
1261         if (!near)
1262                 goto try_prev;
1263 
1264         anon_vma = reusable_anon_vma(near, vma, near);
1265         if (anon_vma)
1266                 return anon_vma;
1267 try_prev:
1268         near = vma->vm_prev;
1269         if (!near)
1270                 goto none;
1271 
1272         anon_vma = reusable_anon_vma(near, near, vma);
1273         if (anon_vma)
1274                 return anon_vma;
1275 none:
1276         /*
1277          * There's no absolute need to look only at touching neighbours:
1278          * we could search further afield for "compatible" anon_vmas.
1279          * But it would probably just be a waste of time searching,
1280          * or lead to too many vmas hanging off the same anon_vma.
1281          * We're trying to allow mprotect remerging later on,
1282          * not trying to minimize memory used for anon_vmas.
1283          */
1284         return NULL;
1285 }
1286 
1287 /*
1288  * If a hint addr is less than mmap_min_addr change hint to be as
1289  * low as possible but still greater than mmap_min_addr
1290  */
1291 static inline unsigned long round_hint_to_min(unsigned long hint)
1292 {
1293         hint &= PAGE_MASK;
1294         if (((void *)hint != NULL) &&
1295             (hint < mmap_min_addr))
1296                 return PAGE_ALIGN(mmap_min_addr);
1297         return hint;
1298 }
1299 
1300 static inline int mlock_future_check(struct mm_struct *mm,
1301                                      unsigned long flags,
1302                                      unsigned long len)
1303 {
1304         unsigned long locked, lock_limit;
1305 
1306         /*  mlock MCL_FUTURE? */
1307         if (flags & VM_LOCKED) {
1308                 locked = len >> PAGE_SHIFT;
1309                 locked += mm->locked_vm;
1310                 lock_limit = rlimit(RLIMIT_MEMLOCK);
1311                 lock_limit >>= PAGE_SHIFT;
1312                 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1313                         return -EAGAIN;
1314         }
1315         return 0;
1316 }
1317 
1318 static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1319 {
1320         if (S_ISREG(inode->i_mode))
1321                 return MAX_LFS_FILESIZE;
1322 
1323         if (S_ISBLK(inode->i_mode))
1324                 return MAX_LFS_FILESIZE;
1325 
1326         /* Special "we do even unsigned file positions" case */
1327         if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1328                 return 0;
1329 
1330         /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1331         return ULONG_MAX;
1332 }
1333 
1334 static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1335                                 unsigned long pgoff, unsigned long len)
1336 {
1337         u64 maxsize = file_mmap_size_max(file, inode);
1338 
1339         if (maxsize && len > maxsize)
1340                 return false;
1341         maxsize -= len;
1342         if (pgoff > maxsize >> PAGE_SHIFT)
1343                 return false;
1344         return true;
1345 }
1346 
1347 /*
1348  * The caller must hold down_write(&current->mm->mmap_sem).
1349  */
1350 unsigned long do_mmap(struct file *file, unsigned long addr,
1351                         unsigned long len, unsigned long prot,
1352                         unsigned long flags, vm_flags_t vm_flags,
1353                         unsigned long pgoff, unsigned long *populate,
1354                         struct list_head *uf)
1355 {
1356         struct mm_struct *mm = current->mm;
1357         int pkey = 0;
1358 
1359         *populate = 0;
1360 
1361         if (!len)
1362                 return -EINVAL;
1363 
1364         /*
1365          * Does the application expect PROT_READ to imply PROT_EXEC?
1366          *
1367          * (the exception is when the underlying filesystem is noexec
1368          *  mounted, in which case we dont add PROT_EXEC.)
1369          */
1370         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1371                 if (!(file && path_noexec(&file->f_path)))
1372                         prot |= PROT_EXEC;
1373 
1374         if (!(flags & MAP_FIXED))
1375                 addr = round_hint_to_min(addr);
1376 
1377         /* Careful about overflows.. */
1378         len = PAGE_ALIGN(len);
1379         if (!len)
1380                 return -ENOMEM;
1381 
1382         /* offset overflow? */
1383         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1384                 return -EOVERFLOW;
1385 
1386         /* Too many mappings? */
1387         if (mm->map_count > sysctl_max_map_count)
1388                 return -ENOMEM;
1389 
1390         /* Obtain the address to map to. we verify (or select) it and ensure
1391          * that it represents a valid section of the address space.
1392          */
1393         addr = get_unmapped_area(file, addr, len, pgoff, flags);
1394         if (offset_in_page(addr))
1395                 return addr;
1396 
1397         if (prot == PROT_EXEC) {
1398                 pkey = execute_only_pkey(mm);
1399                 if (pkey < 0)
1400                         pkey = 0;
1401         }
1402 
1403         /* Do simple checking here so the lower-level routines won't have
1404          * to. we assume access permissions have been handled by the open
1405          * of the memory object, so we don't do any here.
1406          */
1407         vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1408                         mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1409 
1410         if (flags & MAP_LOCKED)
1411                 if (!can_do_mlock())
1412                         return -EPERM;
1413 
1414         if (mlock_future_check(mm, vm_flags, len))
1415                 return -EAGAIN;
1416 
1417         if (file) {
1418                 struct inode *inode = file_inode(file);
1419                 unsigned long flags_mask;
1420 
1421                 if (!file_mmap_ok(file, inode, pgoff, len))
1422                         return -EOVERFLOW;
1423 
1424                 flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
1425 
1426                 switch (flags & MAP_TYPE) {
1427                 case MAP_SHARED:
1428                         /*
1429                          * Force use of MAP_SHARED_VALIDATE with non-legacy
1430                          * flags. E.g. MAP_SYNC is dangerous to use with
1431                          * MAP_SHARED as you don't know which consistency model
1432                          * you will get. We silently ignore unsupported flags
1433                          * with MAP_SHARED to preserve backward compatibility.
1434                          */
1435                         flags &= LEGACY_MAP_MASK;
1436                         /* fall through */
1437                 case MAP_SHARED_VALIDATE:
1438                         if (flags & ~flags_mask)
1439                                 return -EOPNOTSUPP;
1440                         if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1441                                 return -EACCES;
1442 
1443                         /*
1444                          * Make sure we don't allow writing to an append-only
1445                          * file..
1446                          */
1447                         if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1448                                 return -EACCES;
1449 
1450                         /*
1451                          * Make sure there are no mandatory locks on the file.
1452                          */
1453                         if (locks_verify_locked(file))
1454                                 return -EAGAIN;
1455 
1456                         vm_flags |= VM_SHARED | VM_MAYSHARE;
1457                         if (!(file->f_mode & FMODE_WRITE))
1458                                 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1459 
1460                         /* fall through */
1461                 case MAP_PRIVATE:
1462                         if (!(file->f_mode & FMODE_READ))
1463                                 return -EACCES;
1464                         if (path_noexec(&file->f_path)) {
1465                                 if (vm_flags & VM_EXEC)
1466                                         return -EPERM;
1467                                 vm_flags &= ~VM_MAYEXEC;
1468                         }
1469 
1470                         if (!file->f_op->mmap)
1471                                 return -ENODEV;
1472                         if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1473                                 return -EINVAL;
1474                         break;
1475 
1476                 default:
1477                         return -EINVAL;
1478                 }
1479         } else {
1480                 switch (flags & MAP_TYPE) {
1481                 case MAP_SHARED:
1482                         if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1483                                 return -EINVAL;
1484                         /*
1485                          * Ignore pgoff.
1486                          */
1487                         pgoff = 0;
1488                         vm_flags |= VM_SHARED | VM_MAYSHARE;
1489                         break;
1490                 case MAP_PRIVATE:
1491                         /*
1492                          * Set pgoff according to addr for anon_vma.
1493                          */
1494                         pgoff = addr >> PAGE_SHIFT;
1495                         break;
1496                 default:
1497                         return -EINVAL;
1498                 }
1499         }
1500 
1501         /*
1502          * Set 'VM_NORESERVE' if we should not account for the
1503          * memory use of this mapping.
1504          */
1505         if (flags & MAP_NORESERVE) {
1506                 /* We honor MAP_NORESERVE if allowed to overcommit */
1507                 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1508                         vm_flags |= VM_NORESERVE;
1509 
1510                 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1511                 if (file && is_file_hugepages(file))
1512                         vm_flags |= VM_NORESERVE;
1513         }
1514 
1515         addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
1516         if (!IS_ERR_VALUE(addr) &&
1517             ((vm_flags & VM_LOCKED) ||
1518              (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1519                 *populate = len;
1520         return addr;
1521 }
1522 
1523 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1524                 unsigned long, prot, unsigned long, flags,
1525                 unsigned long, fd, unsigned long, pgoff)
1526 {
1527         struct file *file = NULL;
1528         unsigned long retval;
1529 
1530         if (!(flags & MAP_ANONYMOUS)) {
1531                 audit_mmap_fd(fd, flags);
1532                 file = fget(fd);
1533                 if (!file)
1534                         return -EBADF;
1535                 if (is_file_hugepages(file))
1536                         len = ALIGN(len, huge_page_size(hstate_file(file)));
1537                 retval = -EINVAL;
1538                 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1539                         goto out_fput;
1540         } else if (flags & MAP_HUGETLB) {
1541                 struct user_struct *user = NULL;
1542                 struct hstate *hs;
1543 
1544                 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1545                 if (!hs)
1546                         return -EINVAL;
1547 
1548                 len = ALIGN(len, huge_page_size(hs));
1549                 /*
1550                  * VM_NORESERVE is used because the reservations will be
1551                  * taken when vm_ops->mmap() is called
1552                  * A dummy user value is used because we are not locking
1553                  * memory so no accounting is necessary
1554                  */
1555                 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1556                                 VM_NORESERVE,
1557                                 &user, HUGETLB_ANONHUGE_INODE,
1558                                 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1559                 if (IS_ERR(file))
1560                         return PTR_ERR(file);
1561         }
1562 
1563         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1564 
1565         retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1566 out_fput:
1567         if (file)
1568                 fput(file);
1569         return retval;
1570 }
1571 
1572 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1573 struct mmap_arg_struct {
1574         unsigned long addr;
1575         unsigned long len;
1576         unsigned long prot;
1577         unsigned long flags;
1578         unsigned long fd;
1579         unsigned long offset;
1580 };
1581 
1582 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1583 {
1584         struct mmap_arg_struct a;
1585 
1586         if (copy_from_user(&a, arg, sizeof(a)))
1587                 return -EFAULT;
1588         if (offset_in_page(a.offset))
1589                 return -EINVAL;
1590 
1591         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1592                               a.offset >> PAGE_SHIFT);
1593 }
1594 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1595 
1596 /*
1597  * Some shared mappigns will want the pages marked read-only
1598  * to track write events. If so, we'll downgrade vm_page_prot
1599  * to the private version (using protection_map[] without the
1600  * VM_SHARED bit).
1601  */
1602 int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
1603 {
1604         vm_flags_t vm_flags = vma->vm_flags;
1605         const struct vm_operations_struct *vm_ops = vma->vm_ops;
1606 
1607         /* If it was private or non-writable, the write bit is already clear */
1608         if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1609                 return 0;
1610 
1611         /* The backer wishes to know when pages are first written to? */
1612         if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
1613                 return 1;
1614 
1615         /* The open routine did something to the protections that pgprot_modify
1616          * won't preserve? */
1617         if (pgprot_val(vm_page_prot) !=
1618             pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
1619                 return 0;
1620 
1621         /* Do we need to track softdirty? */
1622         if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1623                 return 1;
1624 
1625         /* Specialty mapping? */
1626         if (vm_flags & VM_PFNMAP)
1627                 return 0;
1628 
1629         /* Can the mapping track the dirty pages? */
1630         return vma->vm_file && vma->vm_file->f_mapping &&
1631                 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1632 }
1633 
1634 /*
1635  * We account for memory if it's a private writeable mapping,
1636  * not hugepages and VM_NORESERVE wasn't set.
1637  */
1638 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1639 {
1640         /*
1641          * hugetlb has its own accounting separate from the core VM
1642          * VM_HUGETLB may not be set yet so we cannot check for that flag.
1643          */
1644         if (file && is_file_hugepages(file))
1645                 return 0;
1646 
1647         return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1648 }
1649 
1650 unsigned long mmap_region(struct file *file, unsigned long addr,
1651                 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
1652                 struct list_head *uf)
1653 {
1654         struct mm_struct *mm = current->mm;
1655         struct vm_area_struct *vma, *prev;
1656         int error;
1657         struct rb_node **rb_link, *rb_parent;
1658         unsigned long charged = 0;
1659 
1660         /* Check against address space limit. */
1661         if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
1662                 unsigned long nr_pages;
1663 
1664                 /*
1665                  * MAP_FIXED may remove pages of mappings that intersects with
1666                  * requested mapping. Account for the pages it would unmap.
1667                  */
1668                 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1669 
1670                 if (!may_expand_vm(mm, vm_flags,
1671                                         (len >> PAGE_SHIFT) - nr_pages))
1672                         return -ENOMEM;
1673         }
1674 
1675         /* Clear old maps */
1676         while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1677                               &rb_parent)) {
1678                 if (do_munmap(mm, addr, len, uf))
1679                         return -ENOMEM;
1680         }
1681 
1682         /*
1683          * Private writable mapping: check memory availability
1684          */
1685         if (accountable_mapping(file, vm_flags)) {
1686                 charged = len >> PAGE_SHIFT;
1687                 if (security_vm_enough_memory_mm(mm, charged))
1688                         return -ENOMEM;
1689                 vm_flags |= VM_ACCOUNT;
1690         }
1691 
1692         /*
1693          * Can we just expand an old mapping?
1694          */
1695         vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1696                         NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
1697         if (vma)
1698                 goto out;
1699 
1700         /*
1701          * Determine the object being mapped and call the appropriate
1702          * specific mapper. the address has already been validated, but
1703          * not unmapped, but the maps are removed from the list.
1704          */
1705         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1706         if (!vma) {
1707                 error = -ENOMEM;
1708                 goto unacct_error;
1709         }
1710 
1711         vma->vm_mm = mm;
1712         vma->vm_start = addr;
1713         vma->vm_end = addr + len;
1714         vma->vm_flags = vm_flags;
1715         vma->vm_page_prot = vm_get_page_prot(vm_flags);
1716         vma->vm_pgoff = pgoff;
1717         INIT_LIST_HEAD(&vma->anon_vma_chain);
1718 
1719         if (file) {
1720                 if (vm_flags & VM_DENYWRITE) {
1721                         error = deny_write_access(file);
1722                         if (error)
1723                                 goto free_vma;
1724                 }
1725                 if (vm_flags & VM_SHARED) {
1726                         error = mapping_map_writable(file->f_mapping);
1727                         if (error)
1728                                 goto allow_write_and_free_vma;
1729                 }
1730 
1731                 /* ->mmap() can change vma->vm_file, but must guarantee that
1732                  * vma_link() below can deny write-access if VM_DENYWRITE is set
1733                  * and map writably if VM_SHARED is set. This usually means the
1734                  * new file must not have been exposed to user-space, yet.
1735                  */
1736                 vma->vm_file = get_file(file);
1737                 error = call_mmap(file, vma);
1738                 if (error)
1739                         goto unmap_and_free_vma;
1740 
1741                 /* Can addr have changed??
1742                  *
1743                  * Answer: Yes, several device drivers can do it in their
1744                  *         f_op->mmap method. -DaveM
1745                  * Bug: If addr is changed, prev, rb_link, rb_parent should
1746                  *      be updated for vma_link()
1747                  */
1748                 WARN_ON_ONCE(addr != vma->vm_start);
1749 
1750                 addr = vma->vm_start;
1751                 vm_flags = vma->vm_flags;
1752         } else if (vm_flags & VM_SHARED) {
1753                 error = shmem_zero_setup(vma);
1754                 if (error)
1755                         goto free_vma;
1756         }
1757 
1758         vma_link(mm, vma, prev, rb_link, rb_parent);
1759         /* Once vma denies write, undo our temporary denial count */
1760         if (file) {
1761                 if (vm_flags & VM_SHARED)
1762                         mapping_unmap_writable(file->f_mapping);
1763                 if (vm_flags & VM_DENYWRITE)
1764                         allow_write_access(file);
1765         }
1766         file = vma->vm_file;
1767 out:
1768         perf_event_mmap(vma);
1769 
1770         vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
1771         if (vm_flags & VM_LOCKED) {
1772                 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1773                                         vma == get_gate_vma(current->mm)))
1774                         mm->locked_vm += (len >> PAGE_SHIFT);
1775                 else
1776                         vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
1777         }
1778 
1779         if (file)
1780                 uprobe_mmap(vma);
1781 
1782         /*
1783          * New (or expanded) vma always get soft dirty status.
1784          * Otherwise user-space soft-dirty page tracker won't
1785          * be able to distinguish situation when vma area unmapped,
1786          * then new mapped in-place (which must be aimed as
1787          * a completely new data area).
1788          */
1789         vma->vm_flags |= VM_SOFTDIRTY;
1790 
1791         vma_set_page_prot(vma);
1792 
1793         return addr;
1794 
1795 unmap_and_free_vma:
1796         vma->vm_file = NULL;
1797         fput(file);
1798 
1799         /* Undo any partial mapping done by a device driver. */
1800         unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1801         charged = 0;
1802         if (vm_flags & VM_SHARED)
1803                 mapping_unmap_writable(file->f_mapping);
1804 allow_write_and_free_vma:
1805         if (vm_flags & VM_DENYWRITE)
1806                 allow_write_access(file);
1807 free_vma:
1808         kmem_cache_free(vm_area_cachep, vma);
1809 unacct_error:
1810         if (charged)
1811                 vm_unacct_memory(charged);
1812         return error;
1813 }
1814 
1815 unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1816 {
1817         /*
1818          * We implement the search by looking for an rbtree node that
1819          * immediately follows a suitable gap. That is,
1820          * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1821          * - gap_end   = vma->vm_start        >= info->low_limit  + length;
1822          * - gap_end - gap_start >= length
1823          */
1824 
1825         struct mm_struct *mm = current->mm;
1826         struct vm_area_struct *vma;
1827         unsigned long length, low_limit, high_limit, gap_start, gap_end;
1828 
1829         /* Adjust search length to account for worst case alignment overhead */
1830         length = info->length + info->align_mask;
1831         if (length < info->length)
1832                 return -ENOMEM;
1833 
1834         /* Adjust search limits by the desired length */
1835         if (info->high_limit < length)
1836                 return -ENOMEM;
1837         high_limit = info->high_limit - length;
1838 
1839         if (info->low_limit > high_limit)
1840                 return -ENOMEM;
1841         low_limit = info->low_limit + length;
1842 
1843         /* Check if rbtree root looks promising */
1844         if (RB_EMPTY_ROOT(&mm->mm_rb))
1845                 goto check_highest;
1846         vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1847         if (vma->rb_subtree_gap < length)
1848                 goto check_highest;
1849 
1850         while (true) {
1851                 /* Visit left subtree if it looks promising */
1852                 gap_end = vm_start_gap(vma);
1853                 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1854                         struct vm_area_struct *left =
1855                                 rb_entry(vma->vm_rb.rb_left,
1856                                          struct vm_area_struct, vm_rb);
1857                         if (left->rb_subtree_gap >= length) {
1858                                 vma = left;
1859                                 continue;
1860                         }
1861                 }
1862 
1863                 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
1864 check_current:
1865                 /* Check if current node has a suitable gap */
1866                 if (gap_start > high_limit)
1867                         return -ENOMEM;
1868                 if (gap_end >= low_limit &&
1869                     gap_end > gap_start && gap_end - gap_start >= length)
1870                         goto found;
1871 
1872                 /* Visit right subtree if it looks promising */
1873                 if (vma->vm_rb.rb_right) {
1874                         struct vm_area_struct *right =
1875                                 rb_entry(vma->vm_rb.rb_right,
1876                                          struct vm_area_struct, vm_rb);
1877                         if (right->rb_subtree_gap >= length) {
1878                                 vma = right;
1879                                 continue;
1880                         }
1881                 }
1882 
1883                 /* Go back up the rbtree to find next candidate node */
1884                 while (true) {
1885                         struct rb_node *prev = &vma->vm_rb;
1886                         if (!rb_parent(prev))
1887                                 goto check_highest;
1888                         vma = rb_entry(rb_parent(prev),
1889                                        struct vm_area_struct, vm_rb);
1890                         if (prev == vma->vm_rb.rb_left) {
1891                                 gap_start = vm_end_gap(vma->vm_prev);
1892                                 gap_end = vm_start_gap(vma);
1893                                 goto check_current;
1894                         }
1895                 }
1896         }
1897 
1898 check_highest:
1899         /* Check highest gap, which does not precede any rbtree node */
1900         gap_start = mm->highest_vm_end;
1901         gap_end = ULONG_MAX;  /* Only for VM_BUG_ON below */
1902         if (gap_start > high_limit)
1903                 return -ENOMEM;
1904 
1905 found:
1906         /* We found a suitable gap. Clip it with the original low_limit. */
1907         if (gap_start < info->low_limit)
1908                 gap_start = info->low_limit;
1909 
1910         /* Adjust gap address to the desired alignment */
1911         gap_start += (info->align_offset - gap_start) & info->align_mask;
1912 
1913         VM_BUG_ON(gap_start + info->length > info->high_limit);
1914         VM_BUG_ON(gap_start + info->length > gap_end);
1915         return gap_start;
1916 }
1917 
1918 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1919 {
1920         struct mm_struct *mm = current->mm;
1921         struct vm_area_struct *vma;
1922         unsigned long length, low_limit, high_limit, gap_start, gap_end;
1923 
1924         /* Adjust search length to account for worst case alignment overhead */
1925         length = info->length + info->align_mask;
1926         if (length < info->length)
1927                 return -ENOMEM;
1928 
1929         /*
1930          * Adjust search limits by the desired length.
1931          * See implementation comment at top of unmapped_area().
1932          */
1933         gap_end = info->high_limit;
1934         if (gap_end < length)
1935                 return -ENOMEM;
1936         high_limit = gap_end - length;
1937 
1938         if (info->low_limit > high_limit)
1939                 return -ENOMEM;
1940         low_limit = info->low_limit + length;
1941 
1942         /* Check highest gap, which does not precede any rbtree node */
1943         gap_start = mm->highest_vm_end;
1944         if (gap_start <= high_limit)
1945                 goto found_highest;
1946 
1947         /* Check if rbtree root looks promising */
1948         if (RB_EMPTY_ROOT(&mm->mm_rb))
1949                 return -ENOMEM;
1950         vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1951         if (vma->rb_subtree_gap < length)
1952                 return -ENOMEM;
1953 
1954         while (true) {
1955                 /* Visit right subtree if it looks promising */
1956                 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
1957                 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1958                         struct vm_area_struct *right =
1959                                 rb_entry(vma->vm_rb.rb_right,
1960                                          struct vm_area_struct, vm_rb);
1961                         if (right->rb_subtree_gap >= length) {
1962                                 vma = right;
1963                                 continue;
1964                         }
1965                 }
1966 
1967 check_current:
1968                 /* Check if current node has a suitable gap */
1969                 gap_end = vm_start_gap(vma);
1970                 if (gap_end < low_limit)
1971                         return -ENOMEM;
1972                 if (gap_start <= high_limit &&
1973                     gap_end > gap_start && gap_end - gap_start >= length)
1974                         goto found;
1975 
1976                 /* Visit left subtree if it looks promising */
1977                 if (vma->vm_rb.rb_left) {
1978                         struct vm_area_struct *left =
1979                                 rb_entry(vma->vm_rb.rb_left,
1980                                          struct vm_area_struct, vm_rb);
1981                         if (left->rb_subtree_gap >= length) {
1982                                 vma = left;
1983                                 continue;
1984                         }
1985                 }
1986 
1987                 /* Go back up the rbtree to find next candidate node */
1988                 while (true) {
1989                         struct rb_node *prev = &vma->vm_rb;
1990                         if (!rb_parent(prev))
1991                                 return -ENOMEM;
1992                         vma = rb_entry(rb_parent(prev),
1993                                        struct vm_area_struct, vm_rb);
1994                         if (prev == vma->vm_rb.rb_right) {
1995                                 gap_start = vma->vm_prev ?
1996                                         vm_end_gap(vma->vm_prev) : 0;
1997                                 goto check_current;
1998                         }
1999                 }
2000         }
2001 
2002 found:
2003         /* We found a suitable gap. Clip it with the original high_limit. */
2004         if (gap_end > info->high_limit)
2005                 gap_end = info->high_limit;
2006 
2007 found_highest:
2008         /* Compute highest gap address at the desired alignment */
2009         gap_end -= info->length;
2010         gap_end -= (gap_end - info->align_offset) & info->align_mask;
2011 
2012         VM_BUG_ON(gap_end < info->low_limit);
2013         VM_BUG_ON(gap_end < gap_start);
2014         return gap_end;
2015 }
2016 
2017 /* Get an address range which is currently unmapped.
2018  * For shmat() with addr=0.
2019  *
2020  * Ugly calling convention alert:
2021  * Return value with the low bits set means error value,
2022  * ie
2023  *      if (ret & ~PAGE_MASK)
2024  *              error = ret;
2025  *
2026  * This function "knows" that -ENOMEM has the bits set.
2027  */
2028 #ifndef HAVE_ARCH_UNMAPPED_AREA
2029 unsigned long
2030 arch_get_unmapped_area(struct file *filp, unsigned long addr,
2031                 unsigned long len, unsigned long pgoff, unsigned long flags)
2032 {
2033         struct mm_struct *mm = current->mm;
2034         struct vm_area_struct *vma, *prev;
2035         struct vm_unmapped_area_info info;
2036 
2037         if (len > TASK_SIZE - mmap_min_addr)
2038                 return -ENOMEM;
2039 
2040         if (flags & MAP_FIXED)
2041                 return addr;
2042 
2043         if (addr) {
2044                 addr = PAGE_ALIGN(addr);
2045                 vma = find_vma_prev(mm, addr, &prev);
2046                 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
2047                     (!vma || addr + len <= vm_start_gap(vma)) &&
2048                     (!prev || addr >= vm_end_gap(prev)))
2049                         return addr;
2050         }
2051 
2052         info.flags = 0;
2053         info.length = len;
2054         info.low_limit = mm->mmap_base;
2055         info.high_limit = TASK_SIZE;
2056         info.align_mask = 0;
2057         return vm_unmapped_area(&info);
2058 }
2059 #endif
2060 
2061 /*
2062  * This mmap-allocator allocates new areas top-down from below the
2063  * stack's low limit (the base):
2064  */
2065 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2066 unsigned long
2067 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
2068                           const unsigned long len, const unsigned long pgoff,
2069                           const unsigned long flags)
2070 {
2071         struct vm_area_struct *vma, *prev;
2072         struct mm_struct *mm = current->mm;
2073         unsigned long addr = addr0;
2074         struct vm_unmapped_area_info info;
2075 
2076         /* requested length too big for entire address space */
2077         if (len > TASK_SIZE - mmap_min_addr)
2078                 return -ENOMEM;
2079 
2080         if (flags & MAP_FIXED)
2081                 return addr;
2082 
2083         /* requesting a specific address */
2084         if (addr) {
2085                 addr = PAGE_ALIGN(addr);
2086                 vma = find_vma_prev(mm, addr, &prev);
2087                 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
2088                                 (!vma || addr + len <= vm_start_gap(vma)) &&
2089                                 (!prev || addr >= vm_end_gap(prev)))
2090                         return addr;
2091         }
2092 
2093         info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2094         info.length = len;
2095         info.low_limit = max(PAGE_SIZE, mmap_min_addr);
2096         info.high_limit = mm->mmap_base;
2097         info.align_mask = 0;
2098         addr = vm_unmapped_area(&info);
2099 
2100         /*
2101          * A failed mmap() very likely causes application failure,
2102          * so fall back to the bottom-up function here. This scenario
2103          * can happen with large stack limits and large mmap()
2104          * allocations.
2105          */
2106         if (offset_in_page(addr)) {
2107                 VM_BUG_ON(addr != -ENOMEM);
2108                 info.flags = 0;
2109                 info.low_limit = TASK_UNMAPPED_BASE;
2110                 info.high_limit = TASK_SIZE;
2111                 addr = vm_unmapped_area(&info);
2112         }
2113 
2114         return addr;
2115 }
2116 #endif
2117 
2118 unsigned long
2119 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2120                 unsigned long pgoff, unsigned long flags)
2121 {
2122         unsigned long (*get_area)(struct file *, unsigned long,
2123                                   unsigned long, unsigned long, unsigned long);
2124 
2125         unsigned long error = arch_mmap_check(addr, len, flags);
2126         if (error)
2127                 return error;
2128 
2129         /* Careful about overflows.. */
2130         if (len > TASK_SIZE)
2131                 return -ENOMEM;
2132 
2133         get_area = current->mm->get_unmapped_area;
2134         if (file) {
2135                 if (file->f_op->get_unmapped_area)
2136                         get_area = file->f_op->get_unmapped_area;
2137         } else if (flags & MAP_SHARED) {
2138                 /*
2139                  * mmap_region() will call shmem_zero_setup() to create a file,
2140                  * so use shmem's get_unmapped_area in case it can be huge.
2141                  * do_mmap_pgoff() will clear pgoff, so match alignment.
2142                  */
2143                 pgoff = 0;
2144                 get_area = shmem_get_unmapped_area;
2145         }
2146 
2147         addr = get_area(file, addr, len, pgoff, flags);
2148         if (IS_ERR_VALUE(addr))
2149                 return addr;
2150 
2151         if (addr > TASK_SIZE - len)
2152                 return -ENOMEM;
2153         if (offset_in_page(addr))
2154                 return -EINVAL;
2155 
2156         error = security_mmap_addr(addr);
2157         return error ? error : addr;
2158 }
2159 
2160 EXPORT_SYMBOL(get_unmapped_area);
2161 
2162 /* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
2163 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
2164 {
2165         struct rb_node *rb_node;
2166         struct vm_area_struct *vma;
2167 
2168         /* Check the cache first. */
2169         vma = vmacache_find(mm, addr);
2170         if (likely(vma))
2171                 return vma;
2172 
2173         rb_node = mm->mm_rb.rb_node;
2174 
2175         while (rb_node) {
2176                 struct vm_area_struct *tmp;
2177 
2178                 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2179 
2180                 if (tmp->vm_end > addr) {
2181                         vma = tmp;
2182                         if (tmp->vm_start <= addr)
2183                                 break;
2184                         rb_node = rb_node->rb_left;
2185                 } else
2186                         rb_node = rb_node->rb_right;
2187         }
2188 
2189         if (vma)
2190                 vmacache_update(addr, vma);
2191         return vma;
2192 }
2193 
2194 EXPORT_SYMBOL(find_vma);
2195 
2196 /*
2197  * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2198  */
2199 struct vm_area_struct *
2200 find_vma_prev(struct mm_struct *mm, unsigned long addr,
2201                         struct vm_area_struct **pprev)
2202 {
2203         struct vm_area_struct *vma;
2204 
2205         vma = find_vma(mm, addr);
2206         if (vma) {
2207                 *pprev = vma->vm_prev;
2208         } else {
2209                 struct rb_node *rb_node = mm->mm_rb.rb_node;
2210                 *pprev = NULL;
2211                 while (rb_node) {
2212                         *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2213                         rb_node = rb_node->rb_right;
2214                 }
2215         }
2216         return vma;
2217 }
2218 
2219 /*
2220  * Verify that the stack growth is acceptable and
2221  * update accounting. This is shared with both the
2222  * grow-up and grow-down cases.
2223  */
2224 static int acct_stack_growth(struct vm_area_struct *vma,
2225                              unsigned long size, unsigned long grow)
2226 {
2227         struct mm_struct *mm = vma->vm_mm;
2228         unsigned long new_start;
2229 
2230         /* address space limit tests */
2231         if (!may_expand_vm(mm, vma->vm_flags, grow))
2232                 return -ENOMEM;
2233 
2234         /* Stack limit test */
2235         if (size > rlimit(RLIMIT_STACK))
2236                 return -ENOMEM;
2237 
2238         /* mlock limit tests */
2239         if (vma->vm_flags & VM_LOCKED) {
2240                 unsigned long locked;
2241                 unsigned long limit;
2242                 locked = mm->locked_vm + grow;
2243                 limit = rlimit(RLIMIT_MEMLOCK);
2244                 limit >>= PAGE_SHIFT;
2245                 if (locked > limit && !capable(CAP_IPC_LOCK))
2246                         return -ENOMEM;
2247         }
2248 
2249         /* Check to ensure the stack will not grow into a hugetlb-only region */
2250         new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2251                         vma->vm_end - size;
2252         if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2253                 return -EFAULT;
2254 
2255         /*
2256          * Overcommit..  This must be the final test, as it will
2257          * update security statistics.
2258          */
2259         if (security_vm_enough_memory_mm(mm, grow))
2260                 return -ENOMEM;
2261 
2262         return 0;
2263 }
2264 
2265 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2266 /*
2267  * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2268  * vma is the last one with address > vma->vm_end.  Have to extend vma.
2269  */
2270 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2271 {
2272         struct mm_struct *mm = vma->vm_mm;
2273         struct vm_area_struct *next;
2274         unsigned long gap_addr;
2275         int error = 0;
2276 
2277         if (!(vma->vm_flags & VM_GROWSUP))
2278                 return -EFAULT;
2279 
2280         /* Guard against exceeding limits of the address space. */
2281         address &= PAGE_MASK;
2282         if (address >= (TASK_SIZE & PAGE_MASK))
2283                 return -ENOMEM;
2284         address += PAGE_SIZE;
2285 
2286         /* Enforce stack_guard_gap */
2287         gap_addr = address + stack_guard_gap;
2288 
2289         /* Guard against overflow */
2290         if (gap_addr < address || gap_addr > TASK_SIZE)
2291                 gap_addr = TASK_SIZE;
2292 
2293         next = vma->vm_next;
2294         if (next && next->vm_start < gap_addr &&
2295                         (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
2296                 if (!(next->vm_flags & VM_GROWSUP))
2297                         return -ENOMEM;
2298                 /* Check that both stack segments have the same anon_vma? */
2299         }
2300 
2301         /* We must make sure the anon_vma is allocated. */
2302         if (unlikely(anon_vma_prepare(vma)))
2303                 return -ENOMEM;
2304 
2305         /*
2306          * vma->vm_start/vm_end cannot change under us because the caller
2307          * is required to hold the mmap_sem in read mode.  We need the
2308          * anon_vma lock to serialize against concurrent expand_stacks.
2309          */
2310         anon_vma_lock_write(vma->anon_vma);
2311 
2312         /* Somebody else might have raced and expanded it already */
2313         if (address > vma->vm_end) {
2314                 unsigned long size, grow;
2315 
2316                 size = address - vma->vm_start;
2317                 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2318 
2319                 error = -ENOMEM;
2320                 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2321                         error = acct_stack_growth(vma, size, grow);
2322                         if (!error) {
2323                                 /*
2324                                  * vma_gap_update() doesn't support concurrent
2325                                  * updates, but we only hold a shared mmap_sem
2326                                  * lock here, so we need to protect against
2327                                  * concurrent vma expansions.
2328                                  * anon_vma_lock_write() doesn't help here, as
2329                                  * we don't guarantee that all growable vmas
2330                                  * in a mm share the same root anon vma.
2331                                  * So, we reuse mm->page_table_lock to guard
2332                                  * against concurrent vma expansions.
2333                                  */
2334                                 spin_lock(&mm->page_table_lock);
2335                                 if (vma->vm_flags & VM_LOCKED)
2336                                         mm->locked_vm += grow;
2337                                 vm_stat_account(mm, vma->vm_flags, grow);
2338                                 anon_vma_interval_tree_pre_update_vma(vma);
2339                                 vma->vm_end = address;
2340                                 anon_vma_interval_tree_post_update_vma(vma);
2341                                 if (vma->vm_next)
2342                                         vma_gap_update(vma->vm_next);
2343                                 else
2344                                         mm->highest_vm_end = vm_end_gap(vma);
2345                                 spin_unlock(&mm->page_table_lock);
2346 
2347                                 perf_event_mmap(vma);
2348                         }
2349                 }
2350         }
2351         anon_vma_unlock_write(vma->anon_vma);
2352         khugepaged_enter_vma_merge(vma, vma->vm_flags);
2353         validate_mm(mm);
2354         return error;
2355 }
2356 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2357 
2358 /*
2359  * vma is the first one with address < vma->vm_start.  Have to extend vma.
2360  */
2361 int expand_downwards(struct vm_area_struct *vma,
2362                                    unsigned long address)
2363 {
2364         struct mm_struct *mm = vma->vm_mm;
2365         struct vm_area_struct *prev;
2366         int error;
2367 
2368         address &= PAGE_MASK;
2369         error = security_mmap_addr(address);
2370         if (error)
2371                 return error;
2372 
2373         /* Enforce stack_guard_gap */
2374         prev = vma->vm_prev;
2375         /* Check that both stack segments have the same anon_vma? */
2376         if (prev && !(prev->vm_flags & VM_GROWSDOWN) &&
2377                         (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
2378                 if (address - prev->vm_end < stack_guard_gap)
2379                         return -ENOMEM;
2380         }
2381 
2382         /* We must make sure the anon_vma is allocated. */
2383         if (unlikely(anon_vma_prepare(vma)))
2384                 return -ENOMEM;
2385 
2386         /*
2387          * vma->vm_start/vm_end cannot change under us because the caller
2388          * is required to hold the mmap_sem in read mode.  We need the
2389          * anon_vma lock to serialize against concurrent expand_stacks.
2390          */
2391         anon_vma_lock_write(vma->anon_vma);
2392 
2393         /* Somebody else might have raced and expanded it already */
2394         if (address < vma->vm_start) {
2395                 unsigned long size, grow;
2396 
2397                 size = vma->vm_end - address;
2398                 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2399 
2400                 error = -ENOMEM;
2401                 if (grow <= vma->vm_pgoff) {
2402                         error = acct_stack_growth(vma, size, grow);
2403                         if (!error) {
2404                                 /*
2405                                  * vma_gap_update() doesn't support concurrent
2406                                  * updates, but we only hold a shared mmap_sem
2407                                  * lock here, so we need to protect against
2408                                  * concurrent vma expansions.
2409                                  * anon_vma_lock_write() doesn't help here, as
2410                                  * we don't guarantee that all growable vmas
2411                                  * in a mm share the same root anon vma.
2412                                  * So, we reuse mm->page_table_lock to guard
2413                                  * against concurrent vma expansions.
2414                                  */
2415                                 spin_lock(&mm->page_table_lock);
2416                                 if (vma->vm_flags & VM_LOCKED)
2417                                         mm->locked_vm += grow;
2418                                 vm_stat_account(mm, vma->vm_flags, grow);
2419                                 anon_vma_interval_tree_pre_update_vma(vma);
2420                                 vma->vm_start = address;
2421                                 vma->vm_pgoff -= grow;
2422                                 anon_vma_interval_tree_post_update_vma(vma);
2423                                 vma_gap_update(vma);
2424                                 spin_unlock(&mm->page_table_lock);
2425 
2426                                 perf_event_mmap(vma);
2427                         }
2428                 }
2429         }
2430         anon_vma_unlock_write(vma->anon_vma);
2431         khugepaged_enter_vma_merge(vma, vma->vm_flags);
2432         validate_mm(mm);
2433         return error;
2434 }
2435 
2436 /* enforced gap between the expanding stack and other mappings. */
2437 unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2438 
2439 static int __init cmdline_parse_stack_guard_gap(char *p)
2440 {
2441         unsigned long val;
2442         char *endptr;
2443 
2444         val = simple_strtoul(p, &endptr, 10);
2445         if (!*endptr)
2446                 stack_guard_gap = val << PAGE_SHIFT;
2447 
2448         return 0;
2449 }
2450 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2451 
2452 #ifdef CONFIG_STACK_GROWSUP
2453 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2454 {
2455         return expand_upwards(vma, address);
2456 }
2457 
2458 struct vm_area_struct *
2459 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2460 {
2461         struct vm_area_struct *vma, *prev;
2462 
2463         addr &= PAGE_MASK;
2464         vma = find_vma_prev(mm, addr, &prev);
2465         if (vma && (vma->vm_start <= addr))
2466                 return vma;
2467         if (!prev || expand_stack(prev, addr))
2468                 return NULL;
2469         if (prev->vm_flags & VM_LOCKED)
2470                 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
2471         return prev;
2472 }
2473 #else
2474 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2475 {
2476         return expand_downwards(vma, address);
2477 }
2478 
2479 struct vm_area_struct *
2480 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2481 {
2482         struct vm_area_struct *vma;
2483         unsigned long start;
2484 
2485         addr &= PAGE_MASK;
2486         vma = find_vma(mm, addr);
2487         if (!vma)
2488                 return NULL;
2489         if (vma->vm_start <= addr)
2490                 return vma;
2491         if (!(vma->vm_flags & VM_GROWSDOWN))
2492                 return NULL;
2493         start = vma->vm_start;
2494         if (expand_stack(vma, addr))
2495                 return NULL;
2496         if (vma->vm_flags & VM_LOCKED)
2497                 populate_vma_page_range(vma, addr, start, NULL);
2498         return vma;
2499 }
2500 #endif
2501 
2502 EXPORT_SYMBOL_GPL(find_extend_vma);
2503 
2504 /*
2505  * Ok - we have the memory areas we should free on the vma list,
2506  * so release them, and do the vma updates.
2507  *
2508  * Called with the mm semaphore held.
2509  */
2510 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2511 {
2512         unsigned long nr_accounted = 0;
2513 
2514         /* Update high watermark before we lower total_vm */
2515         update_hiwater_vm(mm);
2516         do {
2517                 long nrpages = vma_pages(vma);
2518 
2519                 if (vma->vm_flags & VM_ACCOUNT)
2520                         nr_accounted += nrpages;
2521                 vm_stat_account(mm, vma->vm_flags, -nrpages);
2522                 vma = remove_vma(vma);
2523         } while (vma);
2524         vm_unacct_memory(nr_accounted);
2525         validate_mm(mm);
2526 }
2527 
2528 /*
2529  * Get rid of page table information in the indicated region.
2530  *
2531  * Called with the mm semaphore held.
2532  */
2533 static void unmap_region(struct mm_struct *mm,
2534                 struct vm_area_struct *vma, struct vm_area_struct *prev,
2535                 unsigned long start, unsigned long end)
2536 {
2537         struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
2538         struct mmu_gather tlb;
2539 
2540         lru_add_drain();
2541         tlb_gather_mmu(&tlb, mm, start, end);
2542         update_hiwater_rss(mm);
2543         unmap_vmas(&tlb, vma, start, end);
2544         free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2545                                  next ? next->vm_start : USER_PGTABLES_CEILING);
2546         tlb_finish_mmu(&tlb, start, end);
2547 }
2548 
2549 /*
2550  * Create a list of vma's touched by the unmap, removing them from the mm's
2551  * vma list as we go..
2552  */
2553 static void
2554 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2555         struct vm_area_struct *prev, unsigned long end)
2556 {
2557         struct vm_area_struct **insertion_point;
2558         struct vm_area_struct *tail_vma = NULL;
2559 
2560         insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2561         vma->vm_prev = NULL;
2562         do {
2563                 vma_rb_erase(vma, &mm->mm_rb);
2564                 mm->map_count--;
2565                 tail_vma = vma;
2566                 vma = vma->vm_next;
2567         } while (vma && vma->vm_start < end);
2568         *insertion_point = vma;
2569         if (vma) {
2570                 vma->vm_prev = prev;
2571                 vma_gap_update(vma);
2572         } else
2573                 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
2574         tail_vma->vm_next = NULL;
2575 
2576         /* Kill the cache */
2577         vmacache_invalidate(mm);
2578 }
2579 
2580 /*
2581  * __split_vma() bypasses sysctl_max_map_count checking.  We use this where it
2582  * has already been checked or doesn't make sense to fail.
2583  */
2584 int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2585                 unsigned long addr, int new_below)
2586 {
2587         struct vm_area_struct *new;
2588         int err;
2589 
2590         if (vma->vm_ops && vma->vm_ops->split) {
2591                 err = vma->vm_ops->split(vma, addr);
2592                 if (err)
2593                         return err;
2594         }
2595 
2596         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2597         if (!new)
2598                 return -ENOMEM;
2599 
2600         /* most fields are the same, copy all, and then fixup */
2601         *new = *vma;
2602 
2603         INIT_LIST_HEAD(&new->anon_vma_chain);
2604 
2605         if (new_below)
2606                 new->vm_end = addr;
2607         else {
2608                 new->vm_start = addr;
2609                 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2610         }
2611 
2612         err = vma_dup_policy(vma, new);
2613         if (err)
2614                 goto out_free_vma;
2615 
2616         err = anon_vma_clone(new, vma);
2617         if (err)
2618                 goto out_free_mpol;
2619 
2620         if (new->vm_file)
2621                 get_file(new->vm_file);
2622 
2623         if (new->vm_ops && new->vm_ops->open)
2624                 new->vm_ops->open(new);
2625 
2626         if (new_below)
2627                 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2628                         ((addr - new->vm_start) >> PAGE_SHIFT), new);
2629         else
2630                 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2631 
2632         /* Success. */
2633         if (!err)
2634                 return 0;
2635 
2636         /* Clean everything up if vma_adjust failed. */
2637         if (new->vm_ops && new->vm_ops->close)
2638                 new->vm_ops->close(new);
2639         if (new->vm_file)
2640                 fput(new->vm_file);
2641         unlink_anon_vmas(new);
2642  out_free_mpol:
2643         mpol_put(vma_policy(new));
2644  out_free_vma:
2645         kmem_cache_free(vm_area_cachep, new);
2646         return err;
2647 }
2648 
2649 /*
2650  * Split a vma into two pieces at address 'addr', a new vma is allocated
2651  * either for the first part or the tail.
2652  */
2653 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2654               unsigned long addr, int new_below)
2655 {
2656         if (mm->map_count >= sysctl_max_map_count)
2657                 return -ENOMEM;
2658 
2659         return __split_vma(mm, vma, addr, new_below);
2660 }
2661 
2662 /* Munmap is split into 2 main parts -- this part which finds
2663  * what needs doing, and the areas themselves, which do the
2664  * work.  This now handles partial unmappings.
2665  * Jeremy Fitzhardinge <jeremy@goop.org>
2666  */
2667 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2668               struct list_head *uf)
2669 {
2670         unsigned long end;
2671         struct vm_area_struct *vma, *prev, *last;
2672 
2673         if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2674                 return -EINVAL;
2675 
2676         len = PAGE_ALIGN(len);
2677         if (len == 0)
2678                 return -EINVAL;
2679 
2680         /* Find the first overlapping VMA */
2681         vma = find_vma(mm, start);
2682         if (!vma)
2683                 return 0;
2684         prev = vma->vm_prev;
2685         /* we have  start < vma->vm_end  */
2686 
2687         /* if it doesn't overlap, we have nothing.. */
2688         end = start + len;
2689         if (vma->vm_start >= end)
2690                 return 0;
2691 
2692         /*
2693          * If we need to split any vma, do it now to save pain later.
2694          *
2695          * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2696          * unmapped vm_area_struct will remain in use: so lower split_vma
2697          * places tmp vma above, and higher split_vma places tmp vma below.
2698          */
2699         if (start > vma->vm_start) {
2700                 int error;
2701 
2702                 /*
2703                  * Make sure that map_count on return from munmap() will
2704                  * not exceed its limit; but let map_count go just above
2705                  * its limit temporarily, to help free resources as expected.
2706                  */
2707                 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2708                         return -ENOMEM;
2709 
2710                 error = __split_vma(mm, vma, start, 0);
2711                 if (error)
2712                         return error;
2713                 prev = vma;
2714         }
2715 
2716         /* Does it split the last one? */
2717         last = find_vma(mm, end);
2718         if (last && end > last->vm_start) {
2719                 int error = __split_vma(mm, last, end, 1);
2720                 if (error)
2721                         return error;
2722         }
2723         vma = prev ? prev->vm_next : mm->mmap;
2724 
2725         if (unlikely(uf)) {
2726                 /*
2727                  * If userfaultfd_unmap_prep returns an error the vmas
2728                  * will remain splitted, but userland will get a
2729                  * highly unexpected error anyway. This is no
2730                  * different than the case where the first of the two
2731                  * __split_vma fails, but we don't undo the first
2732                  * split, despite we could. This is unlikely enough
2733                  * failure that it's not worth optimizing it for.
2734                  */
2735                 int error = userfaultfd_unmap_prep(vma, start, end, uf);
2736                 if (error)
2737                         return error;
2738         }
2739 
2740         /*
2741          * unlock any mlock()ed ranges before detaching vmas
2742          */
2743         if (mm->locked_vm) {
2744                 struct vm_area_struct *tmp = vma;
2745                 while (tmp && tmp->vm_start < end) {
2746                         if (tmp->vm_flags & VM_LOCKED) {
2747                                 mm->locked_vm -= vma_pages(tmp);
2748                                 munlock_vma_pages_all(tmp);
2749                         }
2750                         tmp = tmp->vm_next;
2751                 }
2752         }
2753 
2754         /*
2755          * Remove the vma's, and unmap the actual pages
2756          */
2757         detach_vmas_to_be_unmapped(mm, vma, prev, end);
2758         unmap_region(mm, vma, prev, start, end);
2759 
2760         arch_unmap(mm, vma, start, end);
2761 
2762         /* Fix up all other VM information */
2763         remove_vma_list(mm, vma);
2764 
2765         return 0;
2766 }
2767 
2768 int vm_munmap(unsigned long start, size_t len)
2769 {
2770         int ret;
2771         struct mm_struct *mm = current->mm;
2772         LIST_HEAD(uf);
2773 
2774         if (down_write_killable(&mm->mmap_sem))
2775                 return -EINTR;
2776 
2777         ret = do_munmap(mm, start, len, &uf);
2778         up_write(&mm->mmap_sem);
2779         userfaultfd_unmap_complete(mm, &uf);
2780         return ret;
2781 }
2782 EXPORT_SYMBOL(vm_munmap);
2783 
2784 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2785 {
2786         profile_munmap(addr);
2787         return vm_munmap(addr, len);
2788 }
2789 
2790 
2791 /*
2792  * Emulation of deprecated remap_file_pages() syscall.
2793  */
2794 SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2795                 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2796 {
2797 
2798         struct mm_struct *mm = current->mm;
2799         struct vm_area_struct *vma;
2800         unsigned long populate = 0;
2801         unsigned long ret = -EINVAL;
2802         struct file *file;
2803 
2804         pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2805                      current->comm, current->pid);
2806 
2807         if (prot)
2808                 return ret;
2809         start = start & PAGE_MASK;
2810         size = size & PAGE_MASK;
2811 
2812         if (start + size <= start)
2813                 return ret;
2814 
2815         /* Does pgoff wrap? */
2816         if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2817                 return ret;
2818 
2819         if (down_write_killable(&mm->mmap_sem))
2820                 return -EINTR;
2821 
2822         vma = find_vma(mm, start);
2823 
2824         if (!vma || !(vma->vm_flags & VM_SHARED))
2825                 goto out;
2826 
2827         if (start < vma->vm_start)
2828                 goto out;
2829 
2830         if (start + size > vma->vm_end) {
2831                 struct vm_area_struct *next;
2832 
2833                 for (next = vma->vm_next; next; next = next->vm_next) {
2834                         /* hole between vmas ? */
2835                         if (next->vm_start != next->vm_prev->vm_end)
2836                                 goto out;
2837 
2838                         if (next->vm_file != vma->vm_file)
2839                                 goto out;
2840 
2841                         if (next->vm_flags != vma->vm_flags)
2842                                 goto out;
2843 
2844                         if (start + size <= next->vm_end)
2845                                 break;
2846                 }
2847 
2848                 if (!next)
2849                         goto out;
2850         }
2851 
2852         prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2853         prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2854         prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2855 
2856         flags &= MAP_NONBLOCK;
2857         flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2858         if (vma->vm_flags & VM_LOCKED) {
2859                 struct vm_area_struct *tmp;
2860                 flags |= MAP_LOCKED;
2861 
2862                 /* drop PG_Mlocked flag for over-mapped range */
2863                 for (tmp = vma; tmp->vm_start >= start + size;
2864                                 tmp = tmp->vm_next) {
2865                         /*
2866                          * Split pmd and munlock page on the border
2867                          * of the range.
2868                          */
2869                         vma_adjust_trans_huge(tmp, start, start + size, 0);
2870 
2871                         munlock_vma_pages_range(tmp,
2872                                         max(tmp->vm_start, start),
2873                                         min(tmp->vm_end, start + size));
2874                 }
2875         }
2876 
2877         file = get_file(vma->vm_file);
2878         ret = do_mmap_pgoff(vma->vm_file, start, size,
2879                         prot, flags, pgoff, &populate, NULL);
2880         fput(file);
2881 out:
2882         up_write(&mm->mmap_sem);
2883         if (populate)
2884                 mm_populate(ret, populate);
2885         if (!IS_ERR_VALUE(ret))
2886                 ret = 0;
2887         return ret;
2888 }
2889 
2890 static inline void verify_mm_writelocked(struct mm_struct *mm)
2891 {
2892 #ifdef CONFIG_DEBUG_VM
2893         if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2894                 WARN_ON(1);
2895                 up_read(&mm->mmap_sem);
2896         }
2897 #endif
2898 }
2899 
2900 /*
2901  *  this is really a simplified "do_mmap".  it only handles
2902  *  anonymous maps.  eventually we may be able to do some
2903  *  brk-specific accounting here.
2904  */
2905 static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, struct list_head *uf)
2906 {
2907         struct mm_struct *mm = current->mm;
2908         struct vm_area_struct *vma, *prev;
2909         unsigned long len;
2910         struct rb_node **rb_link, *rb_parent;
2911         pgoff_t pgoff = addr >> PAGE_SHIFT;
2912         int error;
2913 
2914         len = PAGE_ALIGN(request);
2915         if (len < request)
2916                 return -ENOMEM;
2917         if (!len)
2918                 return 0;
2919 
2920         /* Until we need other flags, refuse anything except VM_EXEC. */
2921         if ((flags & (~VM_EXEC)) != 0)
2922                 return -EINVAL;
2923         flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2924 
2925         error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2926         if (offset_in_page(error))
2927                 return error;
2928 
2929         error = mlock_future_check(mm, mm->def_flags, len);
2930         if (error)
2931                 return error;
2932 
2933         /*
2934          * mm->mmap_sem is required to protect against another thread
2935          * changing the mappings in case we sleep.
2936          */
2937         verify_mm_writelocked(mm);
2938 
2939         /*
2940          * Clear old maps.  this also does some error checking for us
2941          */
2942         while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2943                               &rb_parent)) {
2944                 if (do_munmap(mm, addr, len, uf))
2945                         return -ENOMEM;
2946         }
2947 
2948         /* Check against address space limits *after* clearing old maps... */
2949         if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
2950                 return -ENOMEM;
2951 
2952         if (mm->map_count > sysctl_max_map_count)
2953                 return -ENOMEM;
2954 
2955         if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2956                 return -ENOMEM;
2957 
2958         /* Can we just expand an old private anonymous mapping? */
2959         vma = vma_merge(mm, prev, addr, addr + len, flags,
2960                         NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
2961         if (vma)
2962                 goto out;
2963 
2964         /*
2965          * create a vma struct for an anonymous mapping
2966          */
2967         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2968         if (!vma) {
2969                 vm_unacct_memory(len >> PAGE_SHIFT);
2970                 return -ENOMEM;
2971         }
2972 
2973         INIT_LIST_HEAD(&vma->anon_vma_chain);
2974         vma->vm_mm = mm;
2975         vma->vm_start = addr;
2976         vma->vm_end = addr + len;
2977         vma->vm_pgoff = pgoff;
2978         vma->vm_flags = flags;
2979         vma->vm_page_prot = vm_get_page_prot(flags);
2980         vma_link(mm, vma, prev, rb_link, rb_parent);
2981 out:
2982         perf_event_mmap(vma);
2983         mm->total_vm += len >> PAGE_SHIFT;
2984         mm->data_vm += len >> PAGE_SHIFT;
2985         if (flags & VM_LOCKED)
2986                 mm->locked_vm += (len >> PAGE_SHIFT);
2987         vma->vm_flags |= VM_SOFTDIRTY;
2988         return 0;
2989 }
2990 
2991 static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf)
2992 {
2993         return do_brk_flags(addr, len, 0, uf);
2994 }
2995 
2996 int vm_brk_flags(unsigned long addr, unsigned long len, unsigned long flags)
2997 {
2998         struct mm_struct *mm = current->mm;
2999         int ret;
3000         bool populate;
3001         LIST_HEAD(uf);
3002 
3003         if (down_write_killable(&mm->mmap_sem))
3004                 return -EINTR;
3005 
3006         ret = do_brk_flags(addr, len, flags, &uf);
3007         populate = ((mm->def_flags & VM_LOCKED) != 0);
3008         up_write(&mm->mmap_sem);
3009         userfaultfd_unmap_complete(mm, &uf);
3010         if (populate && !ret)
3011                 mm_populate(addr, len);
3012         return ret;
3013 }
3014 EXPORT_SYMBOL(vm_brk_flags);
3015 
3016 int vm_brk(unsigned long addr, unsigned long len)
3017 {
3018         return vm_brk_flags(addr, len, 0);
3019 }
3020 EXPORT_SYMBOL(vm_brk);
3021 
3022 /* Release all mmaps. */
3023 void exit_mmap(struct mm_struct *mm)
3024 {
3025         struct mmu_gather tlb;
3026         struct vm_area_struct *vma;
3027         unsigned long nr_accounted = 0;
3028 
3029         /* mm's last user has gone, and its about to be pulled down */
3030         mmu_notifier_release(mm);
3031 
3032         if (unlikely(mm_is_oom_victim(mm))) {
3033                 /*
3034                  * Manually reap the mm to free as much memory as possible.
3035                  * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3036                  * this mm from further consideration.  Taking mm->mmap_sem for
3037                  * write after setting MMF_OOM_SKIP will guarantee that the oom
3038                  * reaper will not run on this mm again after mmap_sem is
3039                  * dropped.
3040                  *
3041                  * Nothing can be holding mm->mmap_sem here and the above call
3042                  * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3043                  * __oom_reap_task_mm() will not block.
3044                  *
3045                  * This needs to be done before calling munlock_vma_pages_all(),
3046                  * which clears VM_LOCKED, otherwise the oom reaper cannot
3047                  * reliably test it.
3048                  */
3049                 mutex_lock(&oom_lock);
3050                 __oom_reap_task_mm(mm);
3051                 mutex_unlock(&oom_lock);
3052 
3053                 set_bit(MMF_OOM_SKIP, &mm->flags);
3054                 down_write(&mm->mmap_sem);
3055                 up_write(&mm->mmap_sem);
3056         }
3057 
3058         if (mm->locked_vm) {
3059                 vma = mm->mmap;
3060                 while (vma) {
3061                         if (vma->vm_flags & VM_LOCKED)
3062                                 munlock_vma_pages_all(vma);
3063                         vma = vma->vm_next;
3064                 }
3065         }
3066 
3067         arch_exit_mmap(mm);
3068 
3069         vma = mm->mmap;
3070         if (!vma)       /* Can happen if dup_mmap() received an OOM */
3071                 return;
3072 
3073         lru_add_drain();
3074         flush_cache_mm(mm);
3075         tlb_gather_mmu(&tlb, mm, 0, -1);
3076         /* update_hiwater_rss(mm) here? but nobody should be looking */
3077         /* Use -1 here to ensure all VMAs in the mm are unmapped */
3078         unmap_vmas(&tlb, vma, 0, -1);
3079         free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
3080         tlb_finish_mmu(&tlb, 0, -1);
3081 
3082         /*
3083          * Walk the list again, actually closing and freeing it,
3084          * with preemption enabled, without holding any MM locks.
3085          */
3086         while (vma) {
3087                 if (vma->vm_flags & VM_ACCOUNT)
3088                         nr_accounted += vma_pages(vma);
3089                 vma = remove_vma(vma);
3090         }
3091         vm_unacct_memory(nr_accounted);
3092 }
3093 
3094 /* Insert vm structure into process list sorted by address
3095  * and into the inode's i_mmap tree.  If vm_file is non-NULL
3096  * then i_mmap_rwsem is taken here.
3097  */
3098 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
3099 {
3100         struct vm_area_struct *prev;
3101         struct rb_node **rb_link, *rb_parent;
3102 
3103         if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3104                            &prev, &rb_link, &rb_parent))
3105                 return -ENOMEM;
3106         if ((vma->vm_flags & VM_ACCOUNT) &&
3107              security_vm_enough_memory_mm(mm, vma_pages(vma)))
3108                 return -ENOMEM;
3109 
3110         /*
3111          * The vm_pgoff of a purely anonymous vma should be irrelevant
3112          * until its first write fault, when page's anon_vma and index
3113          * are set.  But now set the vm_pgoff it will almost certainly
3114          * end up with (unless mremap moves it elsewhere before that
3115          * first wfault), so /proc/pid/maps tells a consistent story.
3116          *
3117          * By setting it to reflect the virtual start address of the
3118          * vma, merges and splits can happen in a seamless way, just
3119          * using the existing file pgoff checks and manipulations.
3120          * Similarly in do_mmap_pgoff and in do_brk.
3121          */
3122         if (vma_is_anonymous(vma)) {
3123                 BUG_ON(vma->anon_vma);
3124                 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3125         }
3126 
3127         vma_link(mm, vma, prev, rb_link, rb_parent);
3128         return 0;
3129 }
3130 
3131 /*
3132  * Copy the vma structure to a new location in the same mm,
3133  * prior to moving page table entries, to effect an mremap move.
3134  */
3135 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
3136         unsigned long addr, unsigned long len, pgoff_t pgoff,
3137         bool *need_rmap_locks)
3138 {
3139         struct vm_area_struct *vma = *vmap;
3140         unsigned long vma_start = vma->vm_start;
3141         struct mm_struct *mm = vma->vm_mm;
3142         struct vm_area_struct *new_vma, *prev;
3143         struct rb_node **rb_link, *rb_parent;
3144         bool faulted_in_anon_vma = true;
3145 
3146         /*
3147          * If anonymous vma has not yet been faulted, update new pgoff
3148          * to match new location, to increase its chance of merging.
3149          */
3150         if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
3151                 pgoff = addr >> PAGE_SHIFT;
3152                 faulted_in_anon_vma = false;
3153         }
3154 
3155         if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3156                 return NULL;    /* should never get here */
3157         new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
3158                             vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3159                             vma->vm_userfaultfd_ctx);
3160         if (new_vma) {
3161                 /*
3162                  * Source vma may have been merged into new_vma
3163                  */
3164                 if (unlikely(vma_start >= new_vma->vm_start &&
3165                              vma_start < new_vma->vm_end)) {
3166                         /*
3167                          * The only way we can get a vma_merge with
3168                          * self during an mremap is if the vma hasn't
3169                          * been faulted in yet and we were allowed to
3170                          * reset the dst vma->vm_pgoff to the
3171                          * destination address of the mremap to allow
3172                          * the merge to happen. mremap must change the
3173                          * vm_pgoff linearity between src and dst vmas
3174                          * (in turn preventing a vma_merge) to be
3175                          * safe. It is only safe to keep the vm_pgoff
3176                          * linear if there are no pages mapped yet.
3177                          */
3178                         VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
3179                         *vmap = vma = new_vma;
3180                 }
3181                 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
3182         } else {
3183                 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
3184                 if (!new_vma)
3185                         goto out;
3186                 *new_vma = *vma;
3187                 new_vma->vm_start = addr;
3188                 new_vma->vm_end = addr + len;
3189                 new_vma->vm_pgoff = pgoff;
3190                 if (vma_dup_policy(vma, new_vma))
3191                         goto out_free_vma;
3192                 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3193                 if (anon_vma_clone(new_vma, vma))
3194                         goto out_free_mempol;
3195                 if (new_vma->vm_file)
3196                         get_file(new_vma->vm_file);
3197                 if (new_vma->vm_ops && new_vma->vm_ops->open)
3198                         new_vma->vm_ops->open(new_vma);
3199                 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3200                 *need_rmap_locks = false;
3201         }
3202         return new_vma;
3203 
3204 out_free_mempol:
3205         mpol_put(vma_policy(new_vma));
3206 out_free_vma:
3207         kmem_cache_free(vm_area_cachep, new_vma);
3208 out:
3209         return NULL;
3210 }
3211 
3212 /*
3213  * Return true if the calling process may expand its vm space by the passed
3214  * number of pages
3215  */
3216 bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
3217 {
3218         if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3219                 return false;
3220 
3221         if (is_data_mapping(flags) &&
3222             mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
3223                 /* Workaround for Valgrind */
3224                 if (rlimit(RLIMIT_DATA) == 0 &&
3225                     mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3226                         return true;
3227                 if (!ignore_rlimit_data) {
3228                         pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n",
3229                                      current->comm, current->pid,
3230                                      (mm->data_vm + npages) << PAGE_SHIFT,
3231                                      rlimit(RLIMIT_DATA));
3232                         return false;
3233                 }
3234         }
3235 
3236         return true;
3237 }
3238 
3239 void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3240 {
3241         mm->total_vm += npages;
3242 
3243         if (is_exec_mapping(flags))
3244                 mm->exec_vm += npages;
3245         else if (is_stack_mapping(flags))
3246                 mm->stack_vm += npages;
3247         else if (is_data_mapping(flags))
3248                 mm->data_vm += npages;
3249 }
3250 
3251 static int special_mapping_fault(struct vm_fault *vmf);
3252 
3253 /*
3254  * Having a close hook prevents vma merging regardless of flags.
3255  */
3256 static void special_mapping_close(struct vm_area_struct *vma)
3257 {
3258 }
3259 
3260 static const char *special_mapping_name(struct vm_area_struct *vma)
3261 {
3262         return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3263 }
3264 
3265 static int special_mapping_mremap(struct vm_area_struct *new_vma)
3266 {
3267         struct vm_special_mapping *sm = new_vma->vm_private_data;
3268 
3269         if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3270                 return -EFAULT;
3271 
3272         if (sm->mremap)
3273                 return sm->mremap(sm, new_vma);
3274 
3275         return 0;
3276 }
3277 
3278 static const struct vm_operations_struct special_mapping_vmops = {
3279         .close = special_mapping_close,
3280         .fault = special_mapping_fault,
3281         .mremap = special_mapping_mremap,
3282         .name = special_mapping_name,
3283 };
3284 
3285 static const struct vm_operations_struct legacy_special_mapping_vmops = {
3286         .close = special_mapping_close,
3287         .fault = special_mapping_fault,
3288 };
3289 
3290 static int special_mapping_fault(struct vm_fault *vmf)
3291 {
3292         struct vm_area_struct *vma = vmf->vma;
3293         pgoff_t pgoff;
3294         struct page **pages;
3295 
3296         if (vma->vm_ops == &legacy_special_mapping_vmops) {
3297                 pages = vma->vm_private_data;
3298         } else {
3299                 struct vm_special_mapping *sm = vma->vm_private_data;
3300 
3301                 if (sm->fault)
3302                         return sm->fault(sm, vmf->vma, vmf);
3303 
3304                 pages = sm->pages;
3305         }
3306 
3307         for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
3308                 pgoff--;
3309 
3310         if (*pages) {
3311                 struct page *page = *pages;
3312                 get_page(page);
3313                 vmf->page = page;
3314                 return 0;
3315         }
3316 
3317         return VM_FAULT_SIGBUS;
3318 }
3319 
3320 static struct vm_area_struct *__install_special_mapping(
3321         struct mm_struct *mm,
3322         unsigned long addr, unsigned long len,
3323         unsigned long vm_flags, void *priv,
3324         const struct vm_operations_struct *ops)
3325 {
3326         int ret;
3327         struct vm_area_struct *vma;
3328 
3329         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3330         if (unlikely(vma == NULL))
3331                 return ERR_PTR(-ENOMEM);
3332 
3333         INIT_LIST_HEAD(&vma->anon_vma_chain);
3334         vma->vm_mm = mm;
3335         vma->vm_start = addr;
3336         vma->vm_end = addr + len;
3337 
3338         vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
3339         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3340 
3341         vma->vm_ops = ops;
3342         vma->vm_private_data = priv;
3343 
3344         ret = insert_vm_struct(mm, vma);
3345         if (ret)
3346                 goto out;
3347 
3348         vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
3349 
3350         perf_event_mmap(vma);
3351 
3352         return vma;
3353 
3354 out:
3355         kmem_cache_free(vm_area_cachep, vma);
3356         return ERR_PTR(ret);
3357 }
3358 
3359 bool vma_is_special_mapping(const struct vm_area_struct *vma,
3360         const struct vm_special_mapping *sm)
3361 {
3362         return vma->vm_private_data == sm &&
3363                 (vma->vm_ops == &special_mapping_vmops ||
3364                  vma->vm_ops == &legacy_special_mapping_vmops);
3365 }
3366 
3367 /*
3368  * Called with mm->mmap_sem held for writing.
3369  * Insert a new vma covering the given region, with the given flags.
3370  * Its pages are supplied by the given array of struct page *.
3371  * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3372  * The region past the last page supplied will always produce SIGBUS.
3373  * The array pointer and the pages it points to are assumed to stay alive
3374  * for as long as this mapping might exist.
3375  */
3376 struct vm_area_struct *_install_special_mapping(
3377         struct mm_struct *mm,
3378         unsigned long addr, unsigned long len,
3379         unsigned long vm_flags, const struct vm_special_mapping *spec)
3380 {
3381         return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3382                                         &special_mapping_vmops);
3383 }
3384 
3385 int install_special_mapping(struct mm_struct *mm,
3386                             unsigned long addr, unsigned long len,
3387                             unsigned long vm_flags, struct page **pages)
3388 {
3389         struct vm_area_struct *vma = __install_special_mapping(
3390                 mm, addr, len, vm_flags, (void *)pages,
3391                 &legacy_special_mapping_vmops);
3392 
3393         return PTR_ERR_OR_ZERO(vma);
3394 }
3395 
3396 static DEFINE_MUTEX(mm_all_locks_mutex);
3397 
3398 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3399 {
3400         if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3401                 /*
3402                  * The LSB of head.next can't change from under us
3403                  * because we hold the mm_all_locks_mutex.
3404                  */
3405                 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
3406                 /*
3407                  * We can safely modify head.next after taking the
3408                  * anon_vma->root->rwsem. If some other vma in this mm shares
3409                  * the same anon_vma we won't take it again.
3410                  *
3411                  * No need of atomic instructions here, head.next
3412                  * can't change from under us thanks to the
3413                  * anon_vma->root->rwsem.
3414                  */
3415                 if (__test_and_set_bit(0, (unsigned long *)
3416                                        &anon_vma->root->rb_root.rb_root.rb_node))
3417                         BUG();
3418         }
3419 }
3420 
3421 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3422 {
3423         if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3424                 /*
3425                  * AS_MM_ALL_LOCKS can't change from under us because
3426                  * we hold the mm_all_locks_mutex.
3427                  *
3428                  * Operations on ->flags have to be atomic because
3429                  * even if AS_MM_ALL_LOCKS is stable thanks to the
3430                  * mm_all_locks_mutex, there may be other cpus
3431                  * changing other bitflags in parallel to us.
3432                  */
3433                 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3434                         BUG();
3435                 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
3436         }
3437 }
3438 
3439 /*
3440  * This operation locks against the VM for all pte/vma/mm related
3441  * operations that could ever happen on a certain mm. This includes
3442  * vmtruncate, try_to_unmap, and all page faults.
3443  *
3444  * The caller must take the mmap_sem in write mode before calling
3445  * mm_take_all_locks(). The caller isn't allowed to release the
3446  * mmap_sem until mm_drop_all_locks() returns.
3447  *
3448  * mmap_sem in write mode is required in order to block all operations
3449  * that could modify pagetables and free pages without need of
3450  * altering the vma layout. It's also needed in write mode to avoid new
3451  * anon_vmas to be associated with existing vmas.
3452  *
3453  * A single task can't take more than one mm_take_all_locks() in a row
3454  * or it would deadlock.
3455  *
3456  * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3457  * mapping->flags avoid to take the same lock twice, if more than one
3458  * vma in this mm is backed by the same anon_vma or address_space.
3459  *
3460  * We take locks in following order, accordingly to comment at beginning
3461  * of mm/rmap.c:
3462  *   - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3463  *     hugetlb mapping);
3464  *   - all i_mmap_rwsem locks;
3465  *   - all anon_vma->rwseml
3466  *
3467  * We can take all locks within these types randomly because the VM code
3468  * doesn't nest them and we protected from parallel mm_take_all_locks() by
3469  * mm_all_locks_mutex.
3470  *
3471  * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3472  * that may have to take thousand of locks.
3473  *
3474  * mm_take_all_locks() can fail if it's interrupted by signals.
3475  */
3476 int mm_take_all_locks(struct mm_struct *mm)
3477 {
3478         struct vm_area_struct *vma;
3479         struct anon_vma_chain *avc;
3480 
3481         BUG_ON(down_read_trylock(&mm->mmap_sem));
3482 
3483         mutex_lock(&mm_all_locks_mutex);
3484 
3485         for (vma = mm->mmap; vma; vma = vma->vm_next) {
3486                 if (signal_pending(current))
3487                         goto out_unlock;
3488                 if (vma->vm_file && vma->vm_file->f_mapping &&
3489                                 is_vm_hugetlb_page(vma))
3490                         vm_lock_mapping(mm, vma->vm_file->f_mapping);
3491         }
3492 
3493         for (vma = mm->mmap; vma; vma = vma->vm_next) {
3494                 if (signal_pending(current))
3495                         goto out_unlock;
3496                 if (vma->vm_file && vma->vm_file->f_mapping &&
3497                                 !is_vm_hugetlb_page(vma))
3498                         vm_lock_mapping(mm, vma->vm_file->f_mapping);
3499         }
3500 
3501         for (vma = mm->mmap; vma; vma = vma->vm_next) {
3502                 if (signal_pending(current))
3503                         goto out_unlock;
3504                 if (vma->anon_vma)
3505                         list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3506                                 vm_lock_anon_vma(mm, avc->anon_vma);
3507         }
3508 
3509         return 0;
3510 
3511 out_unlock:
3512         mm_drop_all_locks(mm);
3513         return -EINTR;
3514 }
3515 
3516 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3517 {
3518         if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3519                 /*
3520                  * The LSB of head.next can't change to 0 from under
3521                  * us because we hold the mm_all_locks_mutex.
3522                  *
3523                  * We must however clear the bitflag before unlocking
3524                  * the vma so the users using the anon_vma->rb_root will
3525                  * never see our bitflag.
3526                  *
3527                  * No need of atomic instructions here, head.next
3528                  * can't change from under us until we release the
3529                  * anon_vma->root->rwsem.
3530                  */
3531                 if (!__test_and_clear_bit(0, (unsigned long *)
3532                                           &anon_vma->root->rb_root.rb_root.rb_node))
3533                         BUG();
3534                 anon_vma_unlock_write(anon_vma);
3535         }
3536 }
3537 
3538 static void vm_unlock_mapping(struct address_space *mapping)
3539 {
3540         if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3541                 /*
3542                  * AS_MM_ALL_LOCKS can't change to 0 from under us
3543                  * because we hold the mm_all_locks_mutex.
3544                  */
3545                 i_mmap_unlock_write(mapping);
3546                 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3547                                         &mapping->flags))
3548                         BUG();
3549         }
3550 }
3551 
3552 /*
3553  * The mmap_sem cannot be released by the caller until
3554  * mm_drop_all_locks() returns.
3555  */
3556 void mm_drop_all_locks(struct mm_struct *mm)
3557 {
3558         struct vm_area_struct *vma;
3559         struct anon_vma_chain *avc;
3560 
3561         BUG_ON(down_read_trylock(&mm->mmap_sem));
3562         BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3563 
3564         for (vma = mm->mmap; vma; vma = vma->vm_next) {
3565                 if (vma->anon_vma)
3566                         list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3567                                 vm_unlock_anon_vma(avc->anon_vma);
3568                 if (vma->vm_file && vma->vm_file->f_mapping)
3569                         vm_unlock_mapping(vma->vm_file->f_mapping);
3570         }
3571 
3572         mutex_unlock(&mm_all_locks_mutex);
3573 }
3574 
3575 /*
3576  * initialise the percpu counter for VM
3577  */
3578 void __init mmap_init(void)
3579 {
3580         int ret;
3581 
3582         ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3583         VM_BUG_ON(ret);
3584 }
3585 
3586 /*
3587  * Initialise sysctl_user_reserve_kbytes.
3588  *
3589  * This is intended to prevent a user from starting a single memory hogging
3590  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3591  * mode.
3592  *
3593  * The default value is min(3% of free memory, 128MB)
3594  * 128MB is enough to recover with sshd/login, bash, and top/kill.
3595  */
3596 static int init_user_reserve(void)
3597 {
3598         unsigned long free_kbytes;
3599 
3600         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3601 
3602         sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3603         return 0;
3604 }
3605 subsys_initcall(init_user_reserve);
3606 
3607 /*
3608  * Initialise sysctl_admin_reserve_kbytes.
3609  *
3610  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3611  * to log in and kill a memory hogging process.
3612  *
3613  * Systems with more than 256MB will reserve 8MB, enough to recover
3614  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3615  * only reserve 3% of free pages by default.
3616  */
3617 static int init_admin_reserve(void)
3618 {
3619         unsigned long free_kbytes;
3620 
3621         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3622 
3623         sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3624         return 0;
3625 }
3626 subsys_initcall(init_admin_reserve);
3627 
3628 /*
3629  * Reinititalise user and admin reserves if memory is added or removed.
3630  *
3631  * The default user reserve max is 128MB, and the default max for the
3632  * admin reserve is 8MB. These are usually, but not always, enough to
3633  * enable recovery from a memory hogging process using login/sshd, a shell,
3634  * and tools like top. It may make sense to increase or even disable the
3635  * reserve depending on the existence of swap or variations in the recovery
3636  * tools. So, the admin may have changed them.
3637  *
3638  * If memory is added and the reserves have been eliminated or increased above
3639  * the default max, then we'll trust the admin.
3640  *
3641  * If memory is removed and there isn't enough free memory, then we
3642  * need to reset the reserves.
3643  *
3644  * Otherwise keep the reserve set by the admin.
3645  */
3646 static int reserve_mem_notifier(struct notifier_block *nb,
3647                              unsigned long action, void *data)
3648 {
3649         unsigned long tmp, free_kbytes;
3650 
3651         switch (action) {
3652         case MEM_ONLINE:
3653                 /* Default max is 128MB. Leave alone if modified by operator. */
3654                 tmp = sysctl_user_reserve_kbytes;
3655                 if (0 < tmp && tmp < (1UL << 17))
3656                         init_user_reserve();
3657 
3658                 /* Default max is 8MB.  Leave alone if modified by operator. */
3659                 tmp = sysctl_admin_reserve_kbytes;
3660                 if (0 < tmp && tmp < (1UL << 13))
3661                         init_admin_reserve();
3662 
3663                 break;
3664         case MEM_OFFLINE:
3665                 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3666 
3667                 if (sysctl_user_reserve_kbytes > free_kbytes) {
3668                         init_user_reserve();
3669                         pr_info("vm.user_reserve_kbytes reset to %lu\n",
3670                                 sysctl_user_reserve_kbytes);
3671                 }
3672 
3673                 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3674                         init_admin_reserve();
3675                         pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3676                                 sysctl_admin_reserve_kbytes);
3677                 }
3678                 break;
3679         default:
3680                 break;
3681         }
3682         return NOTIFY_OK;
3683 }
3684 
3685 static struct notifier_block reserve_mem_nb = {
3686         .notifier_call = reserve_mem_notifier,
3687 };
3688 
3689 static int __meminit init_reserve_notifier(void)
3690 {
3691         if (register_hotmemory_notifier(&reserve_mem_nb))
3692                 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3693 
3694         return 0;
3695 }
3696 subsys_initcall(init_reserve_notifier);
3697 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp