| blob | bd2e1a533bc182a113523bd16b1af09e81a3c8ea |
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/notifier.h>
41 #include <linux/memory.h>
42 #include <linux/printk.h>
43 #include <linux/userfaultfd_k.h>
44 #include <linux/moduleparam.h>
45 #include <linux/pkeys.h>
47 #include <asm/uaccess.h>
48 #include <asm/cacheflush.h>
49 #include <asm/tlb.h>
50 #include <asm/mmu_context.h>
54 #ifndef arch_mmap_check
55 #define arch_mmap_check(addr, len, flags) (0)
56 #endif
58 #ifndef arch_rebalance_pgtables
59 #define arch_rebalance_pgtables(addr, len) (addr)
60 #endif
62 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
66 #endif
67 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
71 #endif
80 /* description of effects of mapping type and prot in current implementation.
81 * this is due to the limited x86 page protection hardware. The expected
82 * behavior is in parens:
83 *
84 * map_type prot
85 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
86 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 * w: (no) no w: (no) no w: (yes) yes w: (no) no
88 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
89 *
90 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
91 * w: (no) no w: (no) no w: (copy) copy w: (no) no
92 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
93 *
94 */
98 };
101 {
105 }
109 {
111 }
113 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
115 {
122 vm_flags);
123 }
124 }
126 /*
127 * Requires inode->i_mapping->i_mmap_rwsem
128 */
131 {
140 }
142 /*
143 * Unlink a file-based vm structure from its interval tree, to hide
144 * vma from rmap and vmtruncate before freeing its page tables.
145 */
147 {
155 }
156 }
158 /*
159 * Close a vm structure and free it, returning the next.
160 */
162 {
173 }
178 {
187 #ifdef CONFIG_COMPAT_BRK
188 /*
189 * CONFIG_COMPAT_BRK can still be overridden by setting
190 * randomize_va_space to 2, which will still cause mm->start_brk
191 * to be arbitrarily shifted
192 */
195 else
197 #else
199 #endif
203 /*
204 * Check against rlimit here. If this check is done later after the test
205 * of oldbrk with newbrk then it can escape the test and let the data
206 * segment grow beyond its set limit the in case where the limit is
207 * not page aligned -Ram Gupta
208 */
218 /* Always allow shrinking brk. */
223 }
225 /* Check against existing mmap mappings. */
229 /* Ok, looks good - let it rip. */
233 set_brk:
241 out:
245 }
248 {
258 }
264 }
266 }
268 #ifdef CONFIG_DEBUG_VM_RB
270 {
283 }
288 }
293 }
300 }
302 i++;
306 }
309 j++;
313 }
315 }
318 {
326 vma);
327 }
328 }
331 {
346 }
350 i++;
351 }
355 }
360 }
366 }
368 }
369 #else
370 #define validate_mm_rb(root, ignore) do { } while (0)
371 #define validate_mm(mm) do { } while (0)
372 #endif
377 /*
378 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
379 * vma->vm_prev->vm_end values changed, without modifying the vma's position
380 * in the rbtree.
381 */
383 {
384 /*
385 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
386 * function that does exacltly what we want.
387 */
389 }
393 {
394 /* All rb_subtree_gap values must be consistent prior to insertion */
398 }
401 {
402 /*
403 * All rb_subtree_gap values must be consistent prior to erase,
404 * with the possible exception of the vma being erased.
405 */
408 /*
409 * Note rb_erase_augmented is a fairly large inline function,
410 * so make sure we instantiate it only once with our desired
411 * augmented rbtree callbacks.
412 */
414 }
416 /*
417 * vma has some anon_vma assigned, and is already inserted on that
418 * anon_vma's interval trees.
419 *
420 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
421 * vma must be removed from the anon_vma's interval trees using
422 * anon_vma_interval_tree_pre_update_vma().
423 *
424 * After the update, the vma will be reinserted using
425 * anon_vma_interval_tree_post_update_vma().
426 *
427 * The entire update must be protected by exclusive mmap_sem and by
428 * the root anon_vma's mutex.
429 */
432 {
437 }
441 {
446 }
451 {
464 /* Fail if an existing vma overlaps the area */
471 }
472 }
480 }
484 {
488 /* Find first overlaping mapping */
496 /* Iterate over the rest of the overlaps */
505 }
508 }
512 {
513 /* Update tracking information for the gap following the new vma. */
516 else
519 /*
520 * vma->vm_prev wasn't known when we followed the rbtree to find the
521 * correct insertion point for that vma. As a result, we could not
522 * update the vma vm_rb parents rb_subtree_gap values on the way down.
523 * So, we first insert the vma with a zero rb_subtree_gap value
524 * (to be consistent with what we did on the way down), and then
525 * immediately update the gap to the correct value. Finally we
526 * rebalance the rbtree after all augmented values have been set.
527 */
532 }
535 {
550 }
551 }
553 static void
557 {
560 }
565 {
571 }
581 }
583 /*
584 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
585 * mm's list and rbtree. It has already been inserted into the interval tree.
586 */
588 {
597 }
602 {
610 /* Kill the cache */
612 }
614 /*
615 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
616 * is already present in an i_mmap tree without adjusting the tree.
617 * The following helper function should be used when such adjustments
618 * are necessary. The "insert" vma (if any) is to be inserted
619 * before we drop the necessary locks.
620 */
623 {
639 /*
640 * vma expands, overlapping all the next, and
641 * perhaps the one after too (mprotect case 6).
642 */
648 /*
649 * vma expands, overlapping part of the next:
650 * mprotect case 5 shifting the boundary up.
651 */
656 /*
657 * vma shrinks, and !insert tells it's not
658 * split_vma inserting another: so it must be
659 * mprotect case 4 shifting the boundary down.
660 */
664 }
666 /*
667 * Easily overlooked: when mprotect shifts the boundary,
668 * make sure the expanding vma has anon_vma set if the
669 * shrinking vma had, to cover any anon pages imported.
670 */
678 }
679 }
691 /*
692 * Put into interval tree now, so instantiated pages
693 * are visible to arm/parisc __flush_dcache_page
694 * throughout; but we cannot insert into address
695 * space until vma start or end is updated.
696 */
698 }
699 }
713 }
720 }
725 }
729 }
734 }
741 }
744 /*
745 * vma_merge has merged next into vma, and needs
746 * us to remove next before dropping the locks.
747 */
752 /*
753 * split_vma has split insert from vma, and needs
754 * us to insert it before dropping the locks
755 * (it may either follow vma or precede it).
756 */
766 }
767 }
774 }
783 }
789 }
795 /*
796 * In mprotect's case 6 (see comments on vma_merge),
797 * we must remove another next too. It would clutter
798 * up the code too much to do both in one go.
799 */
805 else
807 }
814 }
816 /*
817 * If the vma has a ->close operation then the driver probably needs to release
818 * per-vma resources, so we don't attempt to merge those.
819 */
823 {
824 /*
825 * VM_SOFTDIRTY should not prevent from VMA merging, if we
826 * match the flags but dirty bit -- the caller should mark
827 * merged VMA as dirty. If dirty bit won't be excluded from
828 * comparison, we increase pressue on the memory system forcing
829 * the kernel to generate new VMAs when old one could be
830 * extended instead.
831 */
841 }
846 {
847 /*
848 * The list_is_singular() test is to avoid merging VMA cloned from
849 * parents. This can improve scalability caused by anon_vma lock.
850 */
855 }
857 /*
858 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
859 * in front of (at a lower virtual address and file offset than) the vma.
860 *
861 * We cannot merge two vmas if they have differently assigned (non-NULL)
862 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
863 *
864 * We don't check here for the merged mmap wrapping around the end of pagecache
865 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
866 * wrap, nor mmaps which cover the final page at index -1UL.
867 */
868 static int
871 pgoff_t vm_pgoff,
873 {
878 }
880 }
882 /*
883 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
884 * beyond (at a higher virtual address and file offset than) the vma.
885 *
886 * We cannot merge two vmas if they have differently assigned (non-NULL)
887 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
888 */
889 static int
892 pgoff_t vm_pgoff,
894 {
897 pgoff_t vm_pglen;
901 }
903 }
905 /*
906 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
907 * whether that can be merged with its predecessor or its successor.
908 * Or both (it neatly fills a hole).
909 *
910 * In most cases - when called for mmap, brk or mremap - [addr,end) is
911 * certain not to be mapped by the time vma_merge is called; but when
912 * called for mprotect, it is certain to be already mapped (either at
913 * an offset within prev, or at the start of next), and the flags of
914 * this area are about to be changed to vm_flags - and the no-change
915 * case has already been eliminated.
916 *
917 * The following mprotect cases have to be considered, where AAAA is
918 * the area passed down from mprotect_fixup, never extending beyond one
919 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
920 *
921 * AAAA AAAA AAAA AAAA
922 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
923 * cannot merge might become might become might become
924 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
925 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
926 * mremap move: PPPPNNNNNNNN 8
927 * AAAA
928 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
929 * might become case 1 below case 2 below case 3 below
930 *
931 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
932 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
933 */
940 {
945 /*
946 * We later require that vma->vm_flags == vm_flags,
947 * so this tests vma->vm_flags & VM_SPECIAL, too.
948 */
954 else
960 /*
961 * Can it merge with the predecessor?
962 */
967 vm_userfaultfd_ctx)) {
968 /*
969 * OK, it can. Can we now merge in the successor as well?
970 */
976 vm_userfaultfd_ctx) &&
979 /* cases 1, 6 */
989 }
991 /*
992 * Can this new request be merged in front of next?
993 */
998 vm_userfaultfd_ctx)) {
1009 }
1012 }
1014 /*
1015 * Rough compatbility check to quickly see if it's even worth looking
1016 * at sharing an anon_vma.
1017 *
1018 * They need to have the same vm_file, and the flags can only differ
1019 * in things that mprotect may change.
1020 *
1021 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1022 * we can merge the two vma's. For example, we refuse to merge a vma if
1023 * there is a vm_ops->close() function, because that indicates that the
1024 * driver is doing some kind of reference counting. But that doesn't
1025 * really matter for the anon_vma sharing case.
1026 */
1028 {
1034 }
1036 /*
1037 * Do some basic sanity checking to see if we can re-use the anon_vma
1038 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1039 * the same as 'old', the other will be the new one that is trying
1040 * to share the anon_vma.
1041 *
1042 * NOTE! This runs with mm_sem held for reading, so it is possible that
1043 * the anon_vma of 'old' is concurrently in the process of being set up
1044 * by another page fault trying to merge _that_. But that's ok: if it
1045 * is being set up, that automatically means that it will be a singleton
1046 * acceptable for merging, so we can do all of this optimistically. But
1047 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1048 *
1049 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1050 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1051 * is to return an anon_vma that is "complex" due to having gone through
1052 * a fork).
1053 *
1054 * We also make sure that the two vma's are compatible (adjacent,
1055 * and with the same memory policies). That's all stable, even with just
1056 * a read lock on the mm_sem.
1057 */
1058 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1059 {
1065 }
1067 }
1069 /*
1070 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1071 * neighbouring vmas for a suitable anon_vma, before it goes off
1072 * to allocate a new anon_vma. It checks because a repetitive
1073 * sequence of mprotects and faults may otherwise lead to distinct
1074 * anon_vmas being allocated, preventing vma merge in subsequent
1075 * mprotect.
1076 */
1078 {
1089 try_prev:
1097 none:
1098 /*
1099 * There's no absolute need to look only at touching neighbours:
1100 * we could search further afield for "compatible" anon_vmas.
1101 * But it would probably just be a waste of time searching,
1102 * or lead to too many vmas hanging off the same anon_vma.
1103 * We're trying to allow mprotect remerging later on,
1104 * not trying to minimize memory used for anon_vmas.
1105 */
1107 }
1109 /*
1110 * If a hint addr is less than mmap_min_addr change hint to be as
1111 * low as possible but still greater than mmap_min_addr
1112 */
1114 {
1120 }
1125 {
1128 /* mlock MCL_FUTURE? */
1136 }
1138 }
1140 /*
1141 * The caller must hold down_write(¤t->mm->mmap_sem).
1142 */
1147 {
1156 /*
1157 * Does the application expect PROT_READ to imply PROT_EXEC?
1158 *
1159 * (the exception is when the underlying filesystem is noexec
1160 * mounted, in which case we dont add PROT_EXEC.)
1161 */
1169 /* Careful about overflows.. */
1174 /* offset overflow? */
1178 /* Too many mappings? */
1182 /* Obtain the address to map to. we verify (or select) it and ensure
1183 * that it represents a valid section of the address space.
1184 */
1193 }
1195 /* Do simple checking here so the lower-level routines won't have
1196 * to. we assume access permissions have been handled by the open
1197 * of the memory object, so we don't do any here.
1198 */
1217 /*
1218 * Make sure we don't allow writing to an append-only
1219 * file..
1220 */
1224 /*
1225 * Make sure there are no mandatory locks on the file.
1226 */
1234 /* fall through */
1242 }
1252 }
1258 /*
1259 * Ignore pgoff.
1260 */
1265 /*
1266 * Set pgoff according to addr for anon_vma.
1267 */
1272 }
1273 }
1275 /*
1276 * Set 'VM_NORESERVE' if we should not account for the
1277 * memory use of this mapping.
1278 */
1280 /* We honor MAP_NORESERVE if allowed to overcommit */
1284 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1287 }
1295 }
1300 {
1323 /*
1324 * VM_NORESERVE is used because the reservations will be
1325 * taken when vm_ops->mmap() is called
1326 * A dummy user value is used because we are not locking
1327 * memory so no accounting is necessary
1328 */
1330 VM_NORESERVE,
1335 }
1340 out_fput:
1344 }
1346 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1354 };
1357 {
1367 }
1370 /*
1371 * Some shared mappigns will want the pages marked read-only
1372 * to track write events. If so, we'll downgrade vm_page_prot
1373 * to the private version (using protection_map[] without the
1374 * VM_SHARED bit).
1375 */
1377 {
1381 /* If it was private or non-writable, the write bit is already clear */
1385 /* The backer wishes to know when pages are first written to? */
1389 /* The open routine did something to the protections that pgprot_modify
1390 * won't preserve? */
1395 /* Do we need to track softdirty? */
1399 /* Specialty mapping? */
1403 /* Can the mapping track the dirty pages? */
1406 }
1408 /*
1409 * We account for memory if it's a private writeable mapping,
1410 * not hugepages and VM_NORESERVE wasn't set.
1411 */
1413 {
1414 /*
1415 * hugetlb has its own accounting separate from the core VM
1416 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1417 */
1422 }
1426 {
1433 /* Check against address space limit. */
1437 /*
1438 * MAP_FIXED may remove pages of mappings that intersects with
1439 * requested mapping. Account for the pages it would unmap.
1440 */
1446 }
1448 /* Clear old maps */
1453 }
1455 /*
1456 * Private writable mapping: check memory availability
1457 */
1463 }
1465 /*
1466 * Can we just expand an old mapping?
1467 */
1473 /*
1474 * Determine the object being mapped and call the appropriate
1475 * specific mapper. the address has already been validated, but
1476 * not unmapped, but the maps are removed from the list.
1477 */
1482 }
1497 }
1502 }
1504 /* ->mmap() can change vma->vm_file, but must guarantee that
1505 * vma_link() below can deny write-access if VM_DENYWRITE is set
1506 * and map writably if VM_SHARED is set. This usually means the
1507 * new file must not have been exposed to user-space, yet.
1508 */
1514 /* Can addr have changed??
1515 *
1516 * Answer: Yes, several device drivers can do it in their
1517 * f_op->mmap method. -DaveM
1518 * Bug: If addr is changed, prev, rb_link, rb_parent should
1519 * be updated for vma_link()
1520 */
1529 }
1532 /* Once vma denies write, undo our temporary denial count */
1538 }
1540 out:
1548 else
1550 }
1555 /*
1556 * New (or expanded) vma always get soft dirty status.
1557 * Otherwise user-space soft-dirty page tracker won't
1558 * be able to distinguish situation when vma area unmapped,
1559 * then new mapped in-place (which must be aimed as
1560 * a completely new data area).
1561 */
1568 unmap_and_free_vma:
1572 /* Undo any partial mapping done by a device driver. */
1577 allow_write_and_free_vma:
1580 free_vma:
1582 unacct_error:
1586 }
1589 {
1590 /*
1591 * We implement the search by looking for an rbtree node that
1592 * immediately follows a suitable gap. That is,
1593 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1594 * - gap_end = vma->vm_start >= info->low_limit + length;
1595 * - gap_end - gap_start >= length
1596 */
1602 /* Adjust search length to account for worst case alignment overhead */
1607 /* Adjust search limits by the desired length */
1616 /* Check if rbtree root looks promising */
1624 /* Visit left subtree if it looks promising */
1633 }
1634 }
1637 check_current:
1638 /* Check if current node has a suitable gap */
1644 /* Visit right subtree if it looks promising */
1652 }
1653 }
1655 /* Go back up the rbtree to find next candidate node */
1666 }
1667 }
1668 }
1670 check_highest:
1671 /* Check highest gap, which does not precede any rbtree node */
1677 found:
1678 /* We found a suitable gap. Clip it with the original low_limit. */
1682 /* Adjust gap address to the desired alignment */
1688 }
1691 {
1696 /* Adjust search length to account for worst case alignment overhead */
1701 /*
1702 * Adjust search limits by the desired length.
1703 * See implementation comment at top of unmapped_area().
1704 */
1714 /* Check highest gap, which does not precede any rbtree node */
1719 /* Check if rbtree root looks promising */
1727 /* Visit right subtree if it looks promising */
1736 }
1737 }
1739 check_current:
1740 /* Check if current node has a suitable gap */
1747 /* Visit left subtree if it looks promising */
1755 }
1756 }
1758 /* Go back up the rbtree to find next candidate node */
1769 }
1770 }
1771 }
1773 found:
1774 /* We found a suitable gap. Clip it with the original high_limit. */
1778 found_highest:
1779 /* Compute highest gap address at the desired alignment */
1786 }
1788 /* Get an address range which is currently unmapped.
1789 * For shmat() with addr=0.
1790 *
1791 * Ugly calling convention alert:
1792 * Return value with the low bits set means error value,
1793 * ie
1794 * if (ret & ~PAGE_MASK)
1795 * error = ret;
1796 *
1797 * This function "knows" that -ENOMEM has the bits set.
1798 */
1799 #ifndef HAVE_ARCH_UNMAPPED_AREA
1800 unsigned long
1803 {
1820 }
1828 }
1829 #endif
1831 /*
1832 * This mmap-allocator allocates new areas top-down from below the
1833 * stack's low limit (the base):
1834 */
1835 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1836 unsigned long
1840 {
1846 /* requested length too big for entire address space */
1853 /* requesting a specific address */
1860 }
1869 /*
1870 * A failed mmap() very likely causes application failure,
1871 * so fall back to the bottom-up function here. This scenario
1872 * can happen with large stack limits and large mmap()
1873 * allocations.
1874 */
1881 }
1884 }
1885 #endif
1887 unsigned long
1890 {
1898 /* Careful about overflows.. */
1917 }
1921 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1923 {
1927 /* Check the cache first. */
1946 }
1951 }
1955 /*
1956 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1957 */
1961 {
1973 }
1974 }
1976 }
1978 /*
1979 * Verify that the stack growth is acceptable and
1980 * update accounting. This is shared with both the
1981 * grow-up and grow-down cases.
1982 */
1983 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1984 {
1989 /* address space limit tests */
1993 /* Stack limit test */
2000 /* mlock limit tests */
2009 }
2011 /* Check to ensure the stack will not grow into a hugetlb-only region */
2017 /*
2018 * Overcommit.. This must be the final test, as it will
2019 * update security statistics.
2020 */
2025 }
2027 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2028 /*
2029 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2030 * vma is the last one with address > vma->vm_end. Have to extend vma.
2031 */
2033 {
2040 /* Guard against wrapping around to address 0. */
2043 else
2046 /* We must make sure the anon_vma is allocated. */
2050 /*
2051 * vma->vm_start/vm_end cannot change under us because the caller
2052 * is required to hold the mmap_sem in read mode. We need the
2053 * anon_vma lock to serialize against concurrent expand_stacks.
2054 */
2057 /* Somebody else might have raced and expanded it already */
2068 /*
2069 * vma_gap_update() doesn't support concurrent
2070 * updates, but we only hold a shared mmap_sem
2071 * lock here, so we need to protect against
2072 * concurrent vma expansions.
2073 * anon_vma_lock_write() doesn't help here, as
2074 * we don't guarantee that all growable vmas
2075 * in a mm share the same root anon vma.
2076 * So, we reuse mm->page_table_lock to guard
2077 * against concurrent vma expansions.
2078 */
2088 else
2093 }
2094 }
2095 }
2100 }
2103 /*
2104 * vma is the first one with address < vma->vm_start. Have to extend vma.
2105 */
2108 {
2117 /* We must make sure the anon_vma is allocated. */
2121 /*
2122 * vma->vm_start/vm_end cannot change under us because the caller
2123 * is required to hold the mmap_sem in read mode. We need the
2124 * anon_vma lock to serialize against concurrent expand_stacks.
2125 */
2128 /* Somebody else might have raced and expanded it already */
2139 /*
2140 * vma_gap_update() doesn't support concurrent
2141 * updates, but we only hold a shared mmap_sem
2142 * lock here, so we need to protect against
2143 * concurrent vma expansions.
2144 * anon_vma_lock_write() doesn't help here, as
2145 * we don't guarantee that all growable vmas
2146 * in a mm share the same root anon vma.
2147 * So, we reuse mm->page_table_lock to guard
2148 * against concurrent vma expansions.
2149 */
2162 }
2163 }
2164 }
2169 }
2171 /*
2172 * Note how expand_stack() refuses to expand the stack all the way to
2173 * abut the next virtual mapping, *unless* that mapping itself is also
2174 * a stack mapping. We want to leave room for a guard page, after all
2175 * (the guard page itself is not added here, that is done by the
2176 * actual page faulting logic)
2177 *
2178 * This matches the behavior of the guard page logic (see mm/memory.c:
2179 * check_stack_guard_page()), which only allows the guard page to be
2180 * removed under these circumstances.
2181 */
2182 #ifdef CONFIG_STACK_GROWSUP
2184 {
2192 }
2194 }
2198 {
2210 }
2211 #else
2213 {
2221 }
2223 }
2227 {
2245 }
2246 #endif
2250 /*
2251 * Ok - we have the memory areas we should free on the vma list,
2252 * so release them, and do the vma updates.
2253 *
2254 * Called with the mm semaphore held.
2255 */
2257 {
2260 /* Update high watermark before we lower total_vm */
2272 }
2274 /*
2275 * Get rid of page table information in the indicated region.
2276 *
2277 * Called with the mm semaphore held.
2278 */
2282 {
2293 }
2295 /*
2296 * Create a list of vma's touched by the unmap, removing them from the mm's
2297 * vma list as we go..
2298 */
2299 static void
2302 {
2322 /* Kill the cache */
2324 }
2326 /*
2327 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2328 * munmap path where it doesn't make sense to fail.
2329 */
2332 {
2344 /* most fields are the same, copy all, and then fixup */
2354 }
2373 else
2376 /* Success. */
2380 /* Clean everything up if vma_adjust failed. */
2386 out_free_mpol:
2388 out_free_vma:
2391 }
2393 /*
2394 * Split a vma into two pieces at address 'addr', a new vma is allocated
2395 * either for the first part or the tail.
2396 */
2399 {
2404 }
2406 /* Munmap is split into 2 main parts -- this part which finds
2407 * what needs doing, and the areas themselves, which do the
2408 * work. This now handles partial unmappings.
2409 * Jeremy Fitzhardinge <jeremy@goop.org>
2410 */
2412 {
2423 /* Find the first overlapping VMA */
2428 /* we have start < vma->vm_end */
2430 /* if it doesn't overlap, we have nothing.. */
2435 /*
2436 * If we need to split any vma, do it now to save pain later.
2437 *
2438 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2439 * unmapped vm_area_struct will remain in use: so lower split_vma
2440 * places tmp vma above, and higher split_vma places tmp vma below.
2441 */
2445 /*
2446 * Make sure that map_count on return from munmap() will
2447 * not exceed its limit; but let map_count go just above
2448 * its limit temporarily, to help free resources as expected.
2449 */
2457 }
2459 /* Does it split the last one? */
2465 }
2468 /*
2469 * unlock any mlock()ed ranges before detaching vmas
2470 */
2477 }
2479 }
2480 }
2482 /*
2483 * Remove the vma's, and unmap the actual pages
2484 */
2490 /* Fix up all other VM information */
2494 }
2497 {
2505 }
2509 {
2512 }
2515 /*
2516 * Emulation of deprecated remap_file_pages() syscall.
2517 */
2520 {
2528 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2539 /* Does pgoff wrap? */
2556 /* hole between vmas ? */
2568 }
2572 }
2584 /* drop PG_Mlocked flag for over-mapped range */
2590 }
2591 }
2597 out:
2604 }
2607 {
2608 #ifdef CONFIG_DEBUG_VM
2612 }
2613 #endif
2614 }
2616 /*
2617 * this is really a simplified "do_mmap". it only handles
2618 * anonymous maps. eventually we may be able to do some
2619 * brk-specific accounting here.
2620 */
2622 {
2644 /*
2645 * mm->mmap_sem is required to protect against another thread
2646 * changing the mappings in case we sleep.
2647 */
2650 /*
2651 * Clear old maps. this also does some error checking for us
2652 */
2657 }
2659 /* Check against address space limits *after* clearing old maps... */
2669 /* Can we just expand an old private anonymous mapping? */
2675 /*
2676 * create a vma struct for an anonymous mapping
2677 */
2682 }
2692 out:
2700 }
2703 {
2715 }
2718 /* Release all mmaps. */
2720 {
2725 /* mm's last user has gone, and its about to be pulled down */
2734 }
2735 }
2746 /* update_hiwater_rss(mm) here? but nobody should be looking */
2747 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2753 /*
2754 * Walk the list again, actually closing and freeing it,
2755 * with preemption enabled, without holding any MM locks.
2756 */
2761 }
2763 }
2765 /* Insert vm structure into process list sorted by address
2766 * and into the inode's i_mmap tree. If vm_file is non-NULL
2767 * then i_mmap_rwsem is taken here.
2768 */
2770 {
2781 /*
2782 * The vm_pgoff of a purely anonymous vma should be irrelevant
2783 * until its first write fault, when page's anon_vma and index
2784 * are set. But now set the vm_pgoff it will almost certainly
2785 * end up with (unless mremap moves it elsewhere before that
2786 * first wfault), so /proc/pid/maps tells a consistent story.
2787 *
2788 * By setting it to reflect the virtual start address of the
2789 * vma, merges and splits can happen in a seamless way, just
2790 * using the existing file pgoff checks and manipulations.
2791 * Similarly in do_mmap_pgoff and in do_brk.
2792 */
2796 }
2800 }
2802 /*
2803 * Copy the vma structure to a new location in the same mm,
2804 * prior to moving page table entries, to effect an mremap move.
2805 */
2809 {
2817 /*
2818 * If anonymous vma has not yet been faulted, update new pgoff
2819 * to match new location, to increase its chance of merging.
2820 */
2824 }
2832 /*
2833 * Source vma may have been merged into new_vma
2834 */
2837 /*
2838 * The only way we can get a vma_merge with
2839 * self during an mremap is if the vma hasn't
2840 * been faulted in yet and we were allowed to
2841 * reset the dst vma->vm_pgoff to the
2842 * destination address of the mremap to allow
2843 * the merge to happen. mremap must change the
2844 * vm_pgoff linearity between src and dst vmas
2845 * (in turn preventing a vma_merge) to be
2846 * safe. It is only safe to keep the vm_pgoff
2847 * linear if there are no pages mapped yet.
2848 */
2851 }
2872 }
2875 out_free_mempol:
2877 out_free_vma:
2879 out:
2881 }
2883 /*
2884 * Return true if the calling process may expand its vm space by the passed
2885 * number of pages
2886 */
2888 {
2899 else
2901 }
2904 }
2907 {
2916 }
2921 /*
2922 * Having a close hook prevents vma merging regardless of flags.
2923 */
2925 {
2926 }
2929 {
2931 }
2937 };
2942 };
2946 {
2947 pgoff_t pgoff;
2959 }
2962 pgoff--;
2969 }
2972 }
2979 {
3008 out:
3011 }
3013 /*
3014 * Called with mm->mmap_sem held for writing.
3015 * Insert a new vma covering the given region, with the given flags.
3016 * Its pages are supplied by the given array of struct page *.
3017 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3018 * The region past the last page supplied will always produce SIGBUS.
3019 * The array pointer and the pages it points to are assumed to stay alive
3020 * for as long as this mapping might exist.
3021 */
3026 {
3029 }
3034 {
3040 }
3045 {
3047 /*
3048 * The LSB of head.next can't change from under us
3049 * because we hold the mm_all_locks_mutex.
3050 */
3052 /*
3053 * We can safely modify head.next after taking the
3054 * anon_vma->root->rwsem. If some other vma in this mm shares
3055 * the same anon_vma we won't take it again.
3056 *
3057 * No need of atomic instructions here, head.next
3058 * can't change from under us thanks to the
3059 * anon_vma->root->rwsem.
3060 */
3064 }
3065 }
3068 {
3070 /*
3071 * AS_MM_ALL_LOCKS can't change from under us because
3072 * we hold the mm_all_locks_mutex.
3073 *
3074 * Operations on ->flags have to be atomic because
3075 * even if AS_MM_ALL_LOCKS is stable thanks to the
3076 * mm_all_locks_mutex, there may be other cpus
3077 * changing other bitflags in parallel to us.
3078 */
3082 }
3083 }
3085 /*
3086 * This operation locks against the VM for all pte/vma/mm related
3087 * operations that could ever happen on a certain mm. This includes
3088 * vmtruncate, try_to_unmap, and all page faults.
3089 *
3090 * The caller must take the mmap_sem in write mode before calling
3091 * mm_take_all_locks(). The caller isn't allowed to release the
3092 * mmap_sem until mm_drop_all_locks() returns.
3093 *
3094 * mmap_sem in write mode is required in order to block all operations
3095 * that could modify pagetables and free pages without need of
3096 * altering the vma layout. It's also needed in write mode to avoid new
3097 * anon_vmas to be associated with existing vmas.
3098 *
3099 * A single task can't take more than one mm_take_all_locks() in a row
3100 * or it would deadlock.
3101 *
3102 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3103 * mapping->flags avoid to take the same lock twice, if more than one
3104 * vma in this mm is backed by the same anon_vma or address_space.
3105 *
3106 * We take locks in following order, accordingly to comment at beginning
3107 * of mm/rmap.c:
3108 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3109 * hugetlb mapping);
3110 * - all i_mmap_rwsem locks;
3111 * - all anon_vma->rwseml
3112 *
3113 * We can take all locks within these types randomly because the VM code
3114 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3115 * mm_all_locks_mutex.
3116 *
3117 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3118 * that may have to take thousand of locks.
3119 *
3120 * mm_take_all_locks() can fail if it's interrupted by signals.
3121 */
3123 {
3137 }
3145 }
3153 }
3157 out_unlock:
3160 }
3163 {
3165 /*
3166 * The LSB of head.next can't change to 0 from under
3167 * us because we hold the mm_all_locks_mutex.
3168 *
3169 * We must however clear the bitflag before unlocking
3170 * the vma so the users using the anon_vma->rb_root will
3171 * never see our bitflag.
3172 *
3173 * No need of atomic instructions here, head.next
3174 * can't change from under us until we release the
3175 * anon_vma->root->rwsem.
3176 */
3181 }
3182 }
3185 {
3187 /*
3188 * AS_MM_ALL_LOCKS can't change to 0 from under us
3189 * because we hold the mm_all_locks_mutex.
3190 */
3195 }
3196 }
3198 /*
3199 * The mmap_sem cannot be released by the caller until
3200 * mm_drop_all_locks() returns.
3201 */
3203 {
3216 }
3219 }
3221 /*
3222 * initialise the VMA slab
3223 */
3225 {
3230 }
3232 /*
3233 * Initialise sysctl_user_reserve_kbytes.
3234 *
3235 * This is intended to prevent a user from starting a single memory hogging
3236 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3237 * mode.
3238 *
3239 * The default value is min(3% of free memory, 128MB)
3240 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3241 */
3243 {
3250 }
3253 /*
3254 * Initialise sysctl_admin_reserve_kbytes.
3255 *
3256 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3257 * to log in and kill a memory hogging process.
3258 *
3259 * Systems with more than 256MB will reserve 8MB, enough to recover
3260 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3261 * only reserve 3% of free pages by default.
3262 */
3264 {
3271 }
3274 /*
3275 * Reinititalise user and admin reserves if memory is added or removed.
3276 *
3277 * The default user reserve max is 128MB, and the default max for the
3278 * admin reserve is 8MB. These are usually, but not always, enough to
3279 * enable recovery from a memory hogging process using login/sshd, a shell,
3280 * and tools like top. It may make sense to increase or even disable the
3281 * reserve depending on the existence of swap or variations in the recovery
3282 * tools. So, the admin may have changed them.
3283 *
3284 * If memory is added and the reserves have been eliminated or increased above
3285 * the default max, then we'll trust the admin.
3286 *
3287 * If memory is removed and there isn't enough free memory, then we
3288 * need to reset the reserves.
3289 *
3290 * Otherwise keep the reserve set by the admin.
3291 */
3294 {
3299 /* Default max is 128MB. Leave alone if modified by operator. */
3304 /* Default max is 8MB. Leave alone if modified by operator. */
3316 sysctl_user_reserve_kbytes);
3317 }
3322 sysctl_admin_reserve_kbytes);
3323 }
3327 }
3329 }
3333 };
3336 {
3341 }