| blob | da9990acc08b2d8014e342771a52650b47ee05fc |
1 /*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
7 */
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/backing-dev.h>
14 #include <linux/mm.h>
15 #include <linux/vmacache.h>
16 #include <linux/shm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/syscalls.h>
21 #include <linux/capability.h>
22 #include <linux/init.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/hugetlb.h>
28 #include <linux/profile.h>
29 #include <linux/export.h>
30 #include <linux/mount.h>
31 #include <linux/mempolicy.h>
32 #include <linux/rmap.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/mmdebug.h>
35 #include <linux/perf_event.h>
36 #include <linux/audit.h>
37 #include <linux/khugepaged.h>
38 #include <linux/uprobes.h>
39 #include <linux/rbtree_augmented.h>
40 #include <linux/sched/sysctl.h>
41 #include <linux/notifier.h>
42 #include <linux/memory.h>
43 #include <linux/printk.h>
45 #include <asm/uaccess.h>
46 #include <asm/cacheflush.h>
47 #include <asm/tlb.h>
48 #include <asm/mmu_context.h>
52 #ifndef arch_mmap_check
53 #define arch_mmap_check(addr, len, flags) (0)
54 #endif
56 #ifndef arch_rebalance_pgtables
57 #define arch_rebalance_pgtables(addr, len) (addr)
58 #endif
64 /* description of effects of mapping type and prot in current implementation.
65 * this is due to the limited x86 page protection hardware. The expected
66 * behavior is in parens:
67 *
68 * map_type prot
69 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
70 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (yes) yes w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
73 *
74 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
75 * w: (no) no w: (no) no w: (copy) copy w: (no) no
76 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
77 *
78 */
82 };
85 {
89 }
93 {
95 }
97 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
99 {
106 vm_flags);
107 }
108 }
117 /*
118 * Make sure vm_committed_as in one cacheline and not cacheline shared with
119 * other variables. It can be updated by several CPUs frequently.
120 */
123 /*
124 * The global memory commitment made in the system can be a metric
125 * that can be used to drive ballooning decisions when Linux is hosted
126 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
127 * balancing memory across competing virtual machines that are hosted.
128 * Several metrics drive this policy engine including the guest reported
129 * memory commitment.
130 */
132 {
134 }
137 /*
138 * Check that a process has enough memory to allocate a new virtual
139 * mapping. 0 means there is enough memory for the allocation to
140 * succeed and -ENOMEM implies there is not.
141 *
142 * We currently support three overcommit policies, which are set via the
143 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
144 *
145 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
146 * Additional code 2002 Jul 20 by Robert Love.
147 *
148 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
149 *
150 * Note this is a helper function intended to be used by LSMs which
151 * wish to use this logic.
152 */
154 {
163 /*
164 * Sometimes we want to use more memory than we have
165 */
173 /*
174 * shmem pages shouldn't be counted as free in this
175 * case, they can't be purged, only swapped out, and
176 * that won't affect the overall amount of available
177 * memory in the system.
178 */
183 /*
184 * Any slabs which are created with the
185 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
186 * which are reclaimable, under pressure. The dentry
187 * cache and most inode caches should fall into this
188 */
191 /*
192 * Leave reserved pages. The pages are not for anonymous pages.
193 */
196 else
199 /*
200 * Reserve some for root
201 */
209 }
212 /*
213 * Reserve some for root
214 */
218 /*
219 * Don't let a single process grow so big a user can't recover
220 */
224 }
228 error:
232 }
234 /*
235 * Requires inode->i_mapping->i_mmap_rwsem
236 */
239 {
248 }
250 /*
251 * Unlink a file-based vm structure from its interval tree, to hide
252 * vma from rmap and vmtruncate before freeing its page tables.
253 */
255 {
263 }
264 }
266 /*
267 * Close a vm structure and free it, returning the next.
268 */
270 {
281 }
286 {
295 #ifdef CONFIG_COMPAT_BRK
296 /*
297 * CONFIG_COMPAT_BRK can still be overridden by setting
298 * randomize_va_space to 2, which will still cause mm->start_brk
299 * to be arbitrarily shifted
300 */
303 else
305 #else
307 #endif
311 /*
312 * Check against rlimit here. If this check is done later after the test
313 * of oldbrk with newbrk then it can escape the test and let the data
314 * segment grow beyond its set limit the in case where the limit is
315 * not page aligned -Ram Gupta
316 */
326 /* Always allow shrinking brk. */
331 }
333 /* Check against existing mmap mappings. */
337 /* Ok, looks good - let it rip. */
341 set_brk:
349 out:
353 }
356 {
366 }
372 }
374 }
376 #ifdef CONFIG_DEBUG_VM_RB
378 {
390 }
395 }
400 }
406 }
407 i++;
411 }
414 j++;
418 }
420 }
423 {
431 vma);
432 }
433 }
436 {
451 i++;
452 }
456 }
461 }
467 }
469 }
470 #else
471 #define validate_mm_rb(root, ignore) do { } while (0)
472 #define validate_mm(mm) do { } while (0)
473 #endif
478 /*
479 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
480 * vma->vm_prev->vm_end values changed, without modifying the vma's position
481 * in the rbtree.
482 */
484 {
485 /*
486 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
487 * function that does exacltly what we want.
488 */
490 }
494 {
495 /* All rb_subtree_gap values must be consistent prior to insertion */
499 }
502 {
503 /*
504 * All rb_subtree_gap values must be consistent prior to erase,
505 * with the possible exception of the vma being erased.
506 */
509 /*
510 * Note rb_erase_augmented is a fairly large inline function,
511 * so make sure we instantiate it only once with our desired
512 * augmented rbtree callbacks.
513 */
515 }
517 /*
518 * vma has some anon_vma assigned, and is already inserted on that
519 * anon_vma's interval trees.
520 *
521 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
522 * vma must be removed from the anon_vma's interval trees using
523 * anon_vma_interval_tree_pre_update_vma().
524 *
525 * After the update, the vma will be reinserted using
526 * anon_vma_interval_tree_post_update_vma().
527 *
528 * The entire update must be protected by exclusive mmap_sem and by
529 * the root anon_vma's mutex.
530 */
533 {
538 }
542 {
547 }
552 {
565 /* Fail if an existing vma overlaps the area */
572 }
573 }
581 }
585 {
589 /* Find first overlaping mapping */
597 /* Iterate over the rest of the overlaps */
606 }
609 }
613 {
614 /* Update tracking information for the gap following the new vma. */
617 else
620 /*
621 * vma->vm_prev wasn't known when we followed the rbtree to find the
622 * correct insertion point for that vma. As a result, we could not
623 * update the vma vm_rb parents rb_subtree_gap values on the way down.
624 * So, we first insert the vma with a zero rb_subtree_gap value
625 * (to be consistent with what we did on the way down), and then
626 * immediately update the gap to the correct value. Finally we
627 * rebalance the rbtree after all augmented values have been set.
628 */
633 }
636 {
651 }
652 }
654 static void
658 {
661 }
666 {
672 }
682 }
684 /*
685 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
686 * mm's list and rbtree. It has already been inserted into the interval tree.
687 */
689 {
698 }
703 {
711 /* Kill the cache */
713 }
715 /*
716 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
717 * is already present in an i_mmap tree without adjusting the tree.
718 * The following helper function should be used when such adjustments
719 * are necessary. The "insert" vma (if any) is to be inserted
720 * before we drop the necessary locks.
721 */
724 {
740 /*
741 * vma expands, overlapping all the next, and
742 * perhaps the one after too (mprotect case 6).
743 */
749 /*
750 * vma expands, overlapping part of the next:
751 * mprotect case 5 shifting the boundary up.
752 */
757 /*
758 * vma shrinks, and !insert tells it's not
759 * split_vma inserting another: so it must be
760 * mprotect case 4 shifting the boundary down.
761 */
765 }
767 /*
768 * Easily overlooked: when mprotect shifts the boundary,
769 * make sure the expanding vma has anon_vma set if the
770 * shrinking vma had, to cover any anon pages imported.
771 */
780 }
781 }
782 }
794 /*
795 * Put into interval tree now, so instantiated pages
796 * are visible to arm/parisc __flush_dcache_page
797 * throughout; but we cannot insert into address
798 * space until vma start or end is updated.
799 */
801 }
802 }
816 }
823 }
828 }
832 }
837 }
844 }
847 /*
848 * vma_merge has merged next into vma, and needs
849 * us to remove next before dropping the locks.
850 */
855 /*
856 * split_vma has split insert from vma, and needs
857 * us to insert it before dropping the locks
858 * (it may either follow vma or precede it).
859 */
869 }
870 }
877 }
886 }
892 }
898 /*
899 * In mprotect's case 6 (see comments on vma_merge),
900 * we must remove another next too. It would clutter
901 * up the code too much to do both in one go.
902 */
908 else
910 }
917 }
919 /*
920 * If the vma has a ->close operation then the driver probably needs to release
921 * per-vma resources, so we don't attempt to merge those.
922 */
925 {
926 /*
927 * VM_SOFTDIRTY should not prevent from VMA merging, if we
928 * match the flags but dirty bit -- the caller should mark
929 * merged VMA as dirty. If dirty bit won't be excluded from
930 * comparison, we increase pressue on the memory system forcing
931 * the kernel to generate new VMAs when old one could be
932 * extended instead.
933 */
941 }
946 {
947 /*
948 * The list_is_singular() test is to avoid merging VMA cloned from
949 * parents. This can improve scalability caused by anon_vma lock.
950 */
955 }
957 /*
958 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
959 * in front of (at a lower virtual address and file offset than) the vma.
960 *
961 * We cannot merge two vmas if they have differently assigned (non-NULL)
962 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
963 *
964 * We don't check here for the merged mmap wrapping around the end of pagecache
965 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
966 * wrap, nor mmaps which cover the final page at index -1UL.
967 */
968 static int
971 {
976 }
978 }
980 /*
981 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
982 * beyond (at a higher virtual address and file offset than) the vma.
983 *
984 * We cannot merge two vmas if they have differently assigned (non-NULL)
985 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
986 */
987 static int
990 {
993 pgoff_t vm_pglen;
997 }
999 }
1001 /*
1002 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1003 * whether that can be merged with its predecessor or its successor.
1004 * Or both (it neatly fills a hole).
1005 *
1006 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1007 * certain not to be mapped by the time vma_merge is called; but when
1008 * called for mprotect, it is certain to be already mapped (either at
1009 * an offset within prev, or at the start of next), and the flags of
1010 * this area are about to be changed to vm_flags - and the no-change
1011 * case has already been eliminated.
1012 *
1013 * The following mprotect cases have to be considered, where AAAA is
1014 * the area passed down from mprotect_fixup, never extending beyond one
1015 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1016 *
1017 * AAAA AAAA AAAA AAAA
1018 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1019 * cannot merge might become might become might become
1020 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1021 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1022 * mremap move: PPPPNNNNNNNN 8
1023 * AAAA
1024 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1025 * might become case 1 below case 2 below case 3 below
1026 *
1027 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1028 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1029 */
1035 {
1040 /*
1041 * We later require that vma->vm_flags == vm_flags,
1042 * so this tests vma->vm_flags & VM_SPECIAL, too.
1043 */
1049 else
1055 /*
1056 * Can it merge with the predecessor?
1057 */
1062 /*
1063 * OK, it can. Can we now merge in the successor as well?
1064 */
1071 /* cases 1, 6 */
1081 }
1083 /*
1084 * Can this new request be merged in front of next?
1085 */
1100 }
1103 }
1105 /*
1106 * Rough compatbility check to quickly see if it's even worth looking
1107 * at sharing an anon_vma.
1108 *
1109 * They need to have the same vm_file, and the flags can only differ
1110 * in things that mprotect may change.
1111 *
1112 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1113 * we can merge the two vma's. For example, we refuse to merge a vma if
1114 * there is a vm_ops->close() function, because that indicates that the
1115 * driver is doing some kind of reference counting. But that doesn't
1116 * really matter for the anon_vma sharing case.
1117 */
1119 {
1125 }
1127 /*
1128 * Do some basic sanity checking to see if we can re-use the anon_vma
1129 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1130 * the same as 'old', the other will be the new one that is trying
1131 * to share the anon_vma.
1132 *
1133 * NOTE! This runs with mm_sem held for reading, so it is possible that
1134 * the anon_vma of 'old' is concurrently in the process of being set up
1135 * by another page fault trying to merge _that_. But that's ok: if it
1136 * is being set up, that automatically means that it will be a singleton
1137 * acceptable for merging, so we can do all of this optimistically. But
1138 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1139 *
1140 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1141 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1142 * is to return an anon_vma that is "complex" due to having gone through
1143 * a fork).
1144 *
1145 * We also make sure that the two vma's are compatible (adjacent,
1146 * and with the same memory policies). That's all stable, even with just
1147 * a read lock on the mm_sem.
1148 */
1149 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1150 {
1156 }
1158 }
1160 /*
1161 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1162 * neighbouring vmas for a suitable anon_vma, before it goes off
1163 * to allocate a new anon_vma. It checks because a repetitive
1164 * sequence of mprotects and faults may otherwise lead to distinct
1165 * anon_vmas being allocated, preventing vma merge in subsequent
1166 * mprotect.
1167 */
1169 {
1180 try_prev:
1188 none:
1189 /*
1190 * There's no absolute need to look only at touching neighbours:
1191 * we could search further afield for "compatible" anon_vmas.
1192 * But it would probably just be a waste of time searching,
1193 * or lead to too many vmas hanging off the same anon_vma.
1194 * We're trying to allow mprotect remerging later on,
1195 * not trying to minimize memory used for anon_vmas.
1196 */
1198 }
1200 #ifdef CONFIG_PROC_FS
1203 {
1204 const unsigned long stack_flags
1215 }
1218 /*
1219 * If a hint addr is less than mmap_min_addr change hint to be as
1220 * low as possible but still greater than mmap_min_addr
1221 */
1223 {
1229 }
1234 {
1237 /* mlock MCL_FUTURE? */
1245 }
1247 }
1249 /*
1250 * The caller must hold down_write(¤t->mm->mmap_sem).
1251 */
1257 {
1259 vm_flags_t vm_flags;
1263 /*
1264 * Does the application expect PROT_READ to imply PROT_EXEC?
1265 *
1266 * (the exception is when the underlying filesystem is noexec
1267 * mounted, in which case we dont add PROT_EXEC.)
1268 */
1279 /* Careful about overflows.. */
1284 /* offset overflow? */
1288 /* Too many mappings? */
1292 /* Obtain the address to map to. we verify (or select) it and ensure
1293 * that it represents a valid section of the address space.
1294 */
1299 /* Do simple checking here so the lower-level routines won't have
1300 * to. we assume access permissions have been handled by the open
1301 * of the memory object, so we don't do any here.
1302 */
1321 /*
1322 * Make sure we don't allow writing to an append-only
1323 * file..
1324 */
1328 /*
1329 * Make sure there are no mandatory locks on the file.
1330 */
1338 /* fall through */
1346 }
1356 }
1362 /*
1363 * Ignore pgoff.
1364 */
1369 /*
1370 * Set pgoff according to addr for anon_vma.
1371 */
1376 }
1377 }
1379 /*
1380 * Set 'VM_NORESERVE' if we should not account for the
1381 * memory use of this mapping.
1382 */
1384 /* We honor MAP_NORESERVE if allowed to overcommit */
1388 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1391 }
1399 }
1404 {
1427 /*
1428 * VM_NORESERVE is used because the reservations will be
1429 * taken when vm_ops->mmap() is called
1430 * A dummy user value is used because we are not locking
1431 * memory so no accounting is necessary
1432 */
1434 VM_NORESERVE,
1439 }
1444 out_fput:
1447 out:
1449 }
1451 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1459 };
1462 {
1472 }
1475 /*
1476 * Some shared mappigns will want the pages marked read-only
1477 * to track write events. If so, we'll downgrade vm_page_prot
1478 * to the private version (using protection_map[] without the
1479 * VM_SHARED bit).
1480 */
1482 {
1485 /* If it was private or non-writable, the write bit is already clear */
1489 /* The backer wishes to know when pages are first written to? */
1493 /* The open routine did something to the protections that pgprot_modify
1494 * won't preserve? */
1499 /* Do we need to track softdirty? */
1503 /* Specialty mapping? */
1507 /* Can the mapping track the dirty pages? */
1510 }
1512 /*
1513 * We account for memory if it's a private writeable mapping,
1514 * not hugepages and VM_NORESERVE wasn't set.
1515 */
1517 {
1518 /*
1519 * hugetlb has its own accounting separate from the core VM
1520 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1521 */
1526 }
1530 {
1537 /* Check against address space limit. */
1541 /*
1542 * MAP_FIXED may remove pages of mappings that intersects with
1543 * requested mapping. Account for the pages it would unmap.
1544 */
1552 }
1554 /* Clear old maps */
1556 munmap_back:
1561 }
1563 /*
1564 * Private writable mapping: check memory availability
1565 */
1571 }
1573 /*
1574 * Can we just expand an old mapping?
1575 */
1580 /*
1581 * Determine the object being mapped and call the appropriate
1582 * specific mapper. the address has already been validated, but
1583 * not unmapped, but the maps are removed from the list.
1584 */
1589 }
1604 }
1609 }
1611 /* ->mmap() can change vma->vm_file, but must guarantee that
1612 * vma_link() below can deny write-access if VM_DENYWRITE is set
1613 * and map writably if VM_SHARED is set. This usually means the
1614 * new file must not have been exposed to user-space, yet.
1615 */
1621 /* Can addr have changed??
1622 *
1623 * Answer: Yes, several device drivers can do it in their
1624 * f_op->mmap method. -DaveM
1625 * Bug: If addr is changed, prev, rb_link, rb_parent should
1626 * be updated for vma_link()
1627 */
1636 }
1639 /* Once vma denies write, undo our temporary denial count */
1645 }
1647 out:
1655 else
1657 }
1662 /*
1663 * New (or expanded) vma always get soft dirty status.
1664 * Otherwise user-space soft-dirty page tracker won't
1665 * be able to distinguish situation when vma area unmapped,
1666 * then new mapped in-place (which must be aimed as
1667 * a completely new data area).
1668 */
1675 unmap_and_free_vma:
1679 /* Undo any partial mapping done by a device driver. */
1684 allow_write_and_free_vma:
1687 free_vma:
1689 unacct_error:
1693 }
1696 {
1697 /*
1698 * We implement the search by looking for an rbtree node that
1699 * immediately follows a suitable gap. That is,
1700 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1701 * - gap_end = vma->vm_start >= info->low_limit + length;
1702 * - gap_end - gap_start >= length
1703 */
1709 /* Adjust search length to account for worst case alignment overhead */
1714 /* Adjust search limits by the desired length */
1723 /* Check if rbtree root looks promising */
1731 /* Visit left subtree if it looks promising */
1740 }
1741 }
1744 check_current:
1745 /* Check if current node has a suitable gap */
1751 /* Visit right subtree if it looks promising */
1759 }
1760 }
1762 /* Go back up the rbtree to find next candidate node */
1773 }
1774 }
1775 }
1777 check_highest:
1778 /* Check highest gap, which does not precede any rbtree node */
1784 found:
1785 /* We found a suitable gap. Clip it with the original low_limit. */
1789 /* Adjust gap address to the desired alignment */
1795 }
1798 {
1803 /* Adjust search length to account for worst case alignment overhead */
1808 /*
1809 * Adjust search limits by the desired length.
1810 * See implementation comment at top of unmapped_area().
1811 */
1821 /* Check highest gap, which does not precede any rbtree node */
1826 /* Check if rbtree root looks promising */
1834 /* Visit right subtree if it looks promising */
1843 }
1844 }
1846 check_current:
1847 /* Check if current node has a suitable gap */
1854 /* Visit left subtree if it looks promising */
1862 }
1863 }
1865 /* Go back up the rbtree to find next candidate node */
1876 }
1877 }
1878 }
1880 found:
1881 /* We found a suitable gap. Clip it with the original high_limit. */
1885 found_highest:
1886 /* Compute highest gap address at the desired alignment */
1893 }
1895 /* Get an address range which is currently unmapped.
1896 * For shmat() with addr=0.
1897 *
1898 * Ugly calling convention alert:
1899 * Return value with the low bits set means error value,
1900 * ie
1901 * if (ret & ~PAGE_MASK)
1902 * error = ret;
1903 *
1904 * This function "knows" that -ENOMEM has the bits set.
1905 */
1906 #ifndef HAVE_ARCH_UNMAPPED_AREA
1907 unsigned long
1910 {
1927 }
1935 }
1936 #endif
1938 /*
1939 * This mmap-allocator allocates new areas top-down from below the
1940 * stack's low limit (the base):
1941 */
1942 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1943 unsigned long
1947 {
1953 /* requested length too big for entire address space */
1960 /* requesting a specific address */
1967 }
1976 /*
1977 * A failed mmap() very likely causes application failure,
1978 * so fall back to the bottom-up function here. This scenario
1979 * can happen with large stack limits and large mmap()
1980 * allocations.
1981 */
1988 }
1991 }
1992 #endif
1994 unsigned long
1997 {
2005 /* Careful about overflows.. */
2024 }
2028 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2030 {
2034 /* Check the cache first. */
2054 }
2059 }
2063 /*
2064 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2065 */
2069 {
2081 }
2082 }
2084 }
2086 /*
2087 * Verify that the stack growth is acceptable and
2088 * update accounting. This is shared with both the
2089 * grow-up and grow-down cases.
2090 */
2091 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
2092 {
2097 /* address space limit tests */
2101 /* Stack limit test */
2108 /* mlock limit tests */
2117 }
2119 /* Check to ensure the stack will not grow into a hugetlb-only region */
2125 /*
2126 * Overcommit.. This must be the final test, as it will
2127 * update security statistics.
2128 */
2132 /* Ok, everything looks good - let it rip */
2137 }
2139 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2140 /*
2141 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2142 * vma is the last one with address > vma->vm_end. Have to extend vma.
2143 */
2145 {
2151 /*
2152 * We must make sure the anon_vma is allocated
2153 * so that the anon_vma locking is not a noop.
2154 */
2159 /*
2160 * vma->vm_start/vm_end cannot change under us because the caller
2161 * is required to hold the mmap_sem in read mode. We need the
2162 * anon_vma lock to serialize against concurrent expand_stacks.
2163 * Also guard against wrapping around to address 0.
2164 */
2170 }
2173 /* Somebody else might have raced and expanded it already */
2184 /*
2185 * vma_gap_update() doesn't support concurrent
2186 * updates, but we only hold a shared mmap_sem
2187 * lock here, so we need to protect against
2188 * concurrent vma expansions.
2189 * vma_lock_anon_vma() doesn't help here, as
2190 * we don't guarantee that all growable vmas
2191 * in a mm share the same root anon vma.
2192 * So, we reuse mm->page_table_lock to guard
2193 * against concurrent vma expansions.
2194 */
2201 else
2206 }
2207 }
2208 }
2213 }
2216 /*
2217 * vma is the first one with address < vma->vm_start. Have to extend vma.
2218 */
2221 {
2224 /*
2225 * We must make sure the anon_vma is allocated
2226 * so that the anon_vma locking is not a noop.
2227 */
2238 /*
2239 * vma->vm_start/vm_end cannot change under us because the caller
2240 * is required to hold the mmap_sem in read mode. We need the
2241 * anon_vma lock to serialize against concurrent expand_stacks.
2242 */
2244 /* Somebody else might have raced and expanded it already */
2255 /*
2256 * vma_gap_update() doesn't support concurrent
2257 * updates, but we only hold a shared mmap_sem
2258 * lock here, so we need to protect against
2259 * concurrent vma expansions.
2260 * vma_lock_anon_vma() doesn't help here, as
2261 * we don't guarantee that all growable vmas
2262 * in a mm share the same root anon vma.
2263 * So, we reuse mm->page_table_lock to guard
2264 * against concurrent vma expansions.
2265 */
2275 }
2276 }
2277 }
2282 }
2284 /*
2285 * Note how expand_stack() refuses to expand the stack all the way to
2286 * abut the next virtual mapping, *unless* that mapping itself is also
2287 * a stack mapping. We want to leave room for a guard page, after all
2288 * (the guard page itself is not added here, that is done by the
2289 * actual page faulting logic)
2290 *
2291 * This matches the behavior of the guard page logic (see mm/memory.c:
2292 * check_stack_guard_page()), which only allows the guard page to be
2293 * removed under these circumstances.
2294 */
2295 #ifdef CONFIG_STACK_GROWSUP
2297 {
2305 }
2307 }
2311 {
2323 }
2324 #else
2326 {
2334 }
2336 }
2340 {
2358 }
2359 #endif
2363 /*
2364 * Ok - we have the memory areas we should free on the vma list,
2365 * so release them, and do the vma updates.
2366 *
2367 * Called with the mm semaphore held.
2368 */
2370 {
2373 /* Update high watermark before we lower total_vm */
2385 }
2387 /*
2388 * Get rid of page table information in the indicated region.
2389 *
2390 * Called with the mm semaphore held.
2391 */
2395 {
2406 }
2408 /*
2409 * Create a list of vma's touched by the unmap, removing them from the mm's
2410 * vma list as we go..
2411 */
2412 static void
2415 {
2435 /* Kill the cache */
2437 }
2439 /*
2440 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2441 * munmap path where it doesn't make sense to fail.
2442 */
2445 {
2457 /* most fields are the same, copy all, and then fixup */
2467 }
2486 else
2489 /* Success. */
2493 /* Clean everything up if vma_adjust failed. */
2499 out_free_mpol:
2501 out_free_vma:
2503 out_err:
2505 }
2507 /*
2508 * Split a vma into two pieces at address 'addr', a new vma is allocated
2509 * either for the first part or the tail.
2510 */
2513 {
2518 }
2520 /* Munmap is split into 2 main parts -- this part which finds
2521 * what needs doing, and the areas themselves, which do the
2522 * work. This now handles partial unmappings.
2523 * Jeremy Fitzhardinge <jeremy@goop.org>
2524 */
2526 {
2537 /* Find the first overlapping VMA */
2542 /* we have start < vma->vm_end */
2544 /* if it doesn't overlap, we have nothing.. */
2549 /*
2550 * If we need to split any vma, do it now to save pain later.
2551 *
2552 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2553 * unmapped vm_area_struct will remain in use: so lower split_vma
2554 * places tmp vma above, and higher split_vma places tmp vma below.
2555 */
2559 /*
2560 * Make sure that map_count on return from munmap() will
2561 * not exceed its limit; but let map_count go just above
2562 * its limit temporarily, to help free resources as expected.
2563 */
2571 }
2573 /* Does it split the last one? */
2579 }
2582 /*
2583 * unlock any mlock()ed ranges before detaching vmas
2584 */
2591 }
2593 }
2594 }
2596 /*
2597 * Remove the vma's, and unmap the actual pages
2598 */
2604 /* Fix up all other VM information */
2608 }
2611 {
2619 }
2623 {
2626 }
2629 /*
2630 * Emulation of deprecated remap_file_pages() syscall.
2631 */
2634 {
2654 /* Does pgoff wrap? */
2670 }
2680 /* drop PG_Mlocked flag for over-mapped range */
2682 }
2688 out:
2695 }
2698 {
2699 #ifdef CONFIG_DEBUG_VM
2703 }
2704 #endif
2705 }
2707 /*
2708 * this is really a simplified "do_mmap". it only handles
2709 * anonymous maps. eventually we may be able to do some
2710 * brk-specific accounting here.
2711 */
2713 {
2735 /*
2736 * mm->mmap_sem is required to protect against another thread
2737 * changing the mappings in case we sleep.
2738 */
2741 /*
2742 * Clear old maps. this also does some error checking for us
2743 */
2744 munmap_back:
2749 }
2751 /* Check against address space limits *after* clearing old maps... */
2761 /* Can we just expand an old private anonymous mapping? */
2767 /*
2768 * create a vma struct for an anonymous mapping
2769 */
2774 }
2784 out:
2791 }
2794 {
2806 }
2809 /* Release all mmaps. */
2811 {
2816 /* mm's last user has gone, and its about to be pulled down */
2825 }
2826 }
2837 /* update_hiwater_rss(mm) here? but nobody should be looking */
2838 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2844 /*
2845 * Walk the list again, actually closing and freeing it,
2846 * with preemption enabled, without holding any MM locks.
2847 */
2852 }
2854 }
2856 /* Insert vm structure into process list sorted by address
2857 * and into the inode's i_mmap tree. If vm_file is non-NULL
2858 * then i_mmap_rwsem is taken here.
2859 */
2861 {
2865 /*
2866 * The vm_pgoff of a purely anonymous vma should be irrelevant
2867 * until its first write fault, when page's anon_vma and index
2868 * are set. But now set the vm_pgoff it will almost certainly
2869 * end up with (unless mremap moves it elsewhere before that
2870 * first wfault), so /proc/pid/maps tells a consistent story.
2871 *
2872 * By setting it to reflect the virtual start address of the
2873 * vma, merges and splits can happen in a seamless way, just
2874 * using the existing file pgoff checks and manipulations.
2875 * Similarly in do_mmap_pgoff and in do_brk.
2876 */
2880 }
2890 }
2892 /*
2893 * Copy the vma structure to a new location in the same mm,
2894 * prior to moving page table entries, to effect an mremap move.
2895 */
2899 {
2907 /*
2908 * If anonymous vma has not yet been faulted, update new pgoff
2909 * to match new location, to increase its chance of merging.
2910 */
2914 }
2921 /*
2922 * Source vma may have been merged into new_vma
2923 */
2926 /*
2927 * The only way we can get a vma_merge with
2928 * self during an mremap is if the vma hasn't
2929 * been faulted in yet and we were allowed to
2930 * reset the dst vma->vm_pgoff to the
2931 * destination address of the mremap to allow
2932 * the merge to happen. mremap must change the
2933 * vm_pgoff linearity between src and dst vmas
2934 * (in turn preventing a vma_merge) to be
2935 * safe. It is only safe to keep the vm_pgoff
2936 * linear if there are no pages mapped yet.
2937 */
2940 }
2960 }
2961 }
2964 out_free_mempol:
2966 out_free_vma:
2969 }
2971 /*
2972 * Return true if the calling process may expand its vm space by the passed
2973 * number of pages
2974 */
2976 {
2985 }
2990 /*
2991 * Having a close hook prevents vma merging regardless of flags.
2992 */
2994 {
2995 }
2998 {
3000 }
3006 };
3011 };
3015 {
3016 pgoff_t pgoff;
3019 /*
3020 * special mappings have no vm_file, and in that case, the mm
3021 * uses vm_pgoff internally. So we have to subtract it from here.
3022 * We are allowed to do this because we are the mm; do not copy
3023 * this code into drivers!
3024 */
3029 else
3031 pages;
3034 pgoff--;
3041 }
3044 }
3051 {
3080 out:
3083 }
3085 /*
3086 * Called with mm->mmap_sem held for writing.
3087 * Insert a new vma covering the given region, with the given flags.
3088 * Its pages are supplied by the given array of struct page *.
3089 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3090 * The region past the last page supplied will always produce SIGBUS.
3091 * The array pointer and the pages it points to are assumed to stay alive
3092 * for as long as this mapping might exist.
3093 */
3098 {
3101 }
3106 {
3112 }
3117 {
3119 /*
3120 * The LSB of head.next can't change from under us
3121 * because we hold the mm_all_locks_mutex.
3122 */
3124 /*
3125 * We can safely modify head.next after taking the
3126 * anon_vma->root->rwsem. If some other vma in this mm shares
3127 * the same anon_vma we won't take it again.
3128 *
3129 * No need of atomic instructions here, head.next
3130 * can't change from under us thanks to the
3131 * anon_vma->root->rwsem.
3132 */
3136 }
3137 }
3140 {
3142 /*
3143 * AS_MM_ALL_LOCKS can't change from under us because
3144 * we hold the mm_all_locks_mutex.
3145 *
3146 * Operations on ->flags have to be atomic because
3147 * even if AS_MM_ALL_LOCKS is stable thanks to the
3148 * mm_all_locks_mutex, there may be other cpus
3149 * changing other bitflags in parallel to us.
3150 */
3154 }
3155 }
3157 /*
3158 * This operation locks against the VM for all pte/vma/mm related
3159 * operations that could ever happen on a certain mm. This includes
3160 * vmtruncate, try_to_unmap, and all page faults.
3161 *
3162 * The caller must take the mmap_sem in write mode before calling
3163 * mm_take_all_locks(). The caller isn't allowed to release the
3164 * mmap_sem until mm_drop_all_locks() returns.
3165 *
3166 * mmap_sem in write mode is required in order to block all operations
3167 * that could modify pagetables and free pages without need of
3168 * altering the vma layout. It's also needed in write mode to avoid new
3169 * anon_vmas to be associated with existing vmas.
3170 *
3171 * A single task can't take more than one mm_take_all_locks() in a row
3172 * or it would deadlock.
3173 *
3174 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3175 * mapping->flags avoid to take the same lock twice, if more than one
3176 * vma in this mm is backed by the same anon_vma or address_space.
3177 *
3178 * We can take all the locks in random order because the VM code
3179 * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
3180 * takes more than one of them in a row. Secondly we're protected
3181 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3182 *
3183 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3184 * that may have to take thousand of locks.
3185 *
3186 * mm_take_all_locks() can fail if it's interrupted by signals.
3187 */
3189 {
3202 }
3210 }
3214 out_unlock:
3217 }
3220 {
3222 /*
3223 * The LSB of head.next can't change to 0 from under
3224 * us because we hold the mm_all_locks_mutex.
3225 *
3226 * We must however clear the bitflag before unlocking
3227 * the vma so the users using the anon_vma->rb_root will
3228 * never see our bitflag.
3229 *
3230 * No need of atomic instructions here, head.next
3231 * can't change from under us until we release the
3232 * anon_vma->root->rwsem.
3233 */
3238 }
3239 }
3242 {
3244 /*
3245 * AS_MM_ALL_LOCKS can't change to 0 from under us
3246 * because we hold the mm_all_locks_mutex.
3247 */
3252 }
3253 }
3255 /*
3256 * The mmap_sem cannot be released by the caller until
3257 * mm_drop_all_locks() returns.
3258 */
3260 {
3273 }
3276 }
3278 /*
3279 * initialise the VMA slab
3280 */
3282 {
3287 }
3289 /*
3290 * Initialise sysctl_user_reserve_kbytes.
3291 *
3292 * This is intended to prevent a user from starting a single memory hogging
3293 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3294 * mode.
3295 *
3296 * The default value is min(3% of free memory, 128MB)
3297 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3298 */
3300 {
3307 }
3310 /*
3311 * Initialise sysctl_admin_reserve_kbytes.
3312 *
3313 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3314 * to log in and kill a memory hogging process.
3315 *
3316 * Systems with more than 256MB will reserve 8MB, enough to recover
3317 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3318 * only reserve 3% of free pages by default.
3319 */
3321 {
3328 }
3331 /*
3332 * Reinititalise user and admin reserves if memory is added or removed.
3333 *
3334 * The default user reserve max is 128MB, and the default max for the
3335 * admin reserve is 8MB. These are usually, but not always, enough to
3336 * enable recovery from a memory hogging process using login/sshd, a shell,
3337 * and tools like top. It may make sense to increase or even disable the
3338 * reserve depending on the existence of swap or variations in the recovery
3339 * tools. So, the admin may have changed them.
3340 *
3341 * If memory is added and the reserves have been eliminated or increased above
3342 * the default max, then we'll trust the admin.
3343 *
3344 * If memory is removed and there isn't enough free memory, then we
3345 * need to reset the reserves.
3346 *
3347 * Otherwise keep the reserve set by the admin.
3348 */
3351 {
3356 /* Default max is 128MB. Leave alone if modified by operator. */
3361 /* Default max is 8MB. Leave alone if modified by operator. */
3373 sysctl_user_reserve_kbytes);
3374 }
3379 sysctl_admin_reserve_kbytes);
3380 }
3384 }
3386 }
3390 };
3393 {
3398 }