This patch fixes corner case for MAP_FIXED when requested mapping length
is larger than rlimit for virtual memory. In such case any overlapping
mappings are unmapped before we check for the limit and return ENOMEM.
The check is moved before the loop that unmaps overlapping parts of
existing mappings. When we are about to hit the limit (currently mapped
pages + len > limit) we scan for overlapping pages and check again
accounting for them.
This fixes situation when userspace program expects that the previous
mappings are preserved after the mmap() syscall has returned with error.
(POSIX clearly states that successfull mapping shall replace any
previous mappings.)
This corner case was found and can be tested with LTP testcase:
testcases/open_posix_testsuite/conformance/interfaces/mmap/24-2.c
In this case the mmap, which is clearly over current limit, unmaps
dynamic libraries and the testcase segfaults right after returning into
userspace.
I've also looked at the second instance of the unmapping loop in the
do_brk(). The do_brk() is called from brk() syscall and from vm_brk().
The brk() syscall checks for overlapping mappings and bails out when
there are any (so it can't be triggered from the brk syscall). The
vm_brk() is called only from binmft handlers so it shouldn't be
triggered unless binmft handler created overlapping mappings.
Signed-off-by: Cyril Hrubis <[email protected]>
---
mm/mmap.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++----
1 file changed, 46 insertions(+), 4 deletions(-)
diff --git a/mm/mmap.c b/mm/mmap.c
index 2664a47..e755080 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -33,6 +33,7 @@
#include <linux/uprobes.h>
#include <linux/rbtree_augmented.h>
#include <linux/sched/sysctl.h>
+#include <linux/kernel.h>
#include <asm/uaccess.h>
#include <asm/cacheflush.h>
@@ -543,6 +544,34 @@ static int find_vma_links(struct mm_struct *mm, unsigned long addr,
return 0;
}
+static unsigned long count_vma_pages_range(struct mm_struct *mm,
+ unsigned long addr, unsigned long end)
+{
+ unsigned long nr_pages = 0;
+ struct vm_area_struct *vma;
+
+ /* Find first overlaping mapping */
+ vma = find_vma_intersection(mm, addr, end);
+ if (!vma)
+ return 0;
+
+ nr_pages = (min(end, vma->vm_end) -
+ max(addr, vma->vm_start)) >> PAGE_SHIFT;
+
+ /* Iterate over the rest of the overlaps */
+ for (vma = vma->vm_next; vma; vma = vma->vm_next) {
+ unsigned long overlap_len;
+
+ if (vma->vm_start > end)
+ break;
+
+ overlap_len = min(end, vma->vm_end) - vma->vm_start;
+ nr_pages += overlap_len >> PAGE_SHIFT;
+ }
+
+ return nr_pages;
+}
+
void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
struct rb_node **rb_link, struct rb_node *rb_parent)
{
@@ -1433,6 +1462,23 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
unsigned long charged = 0;
struct inode *inode = file ? file_inode(file) : NULL;
+ /* Check against address space limit. */
+ if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
+ unsigned long nr_pages;
+
+ /*
+ * MAP_FIXED may remove pages of mappings that intersects with
+ * requested mapping. Account for the pages it would unmap.
+ */
+ if (!(vm_flags & MAP_FIXED))
+ return -ENOMEM;
+
+ nr_pages = count_vma_pages_range(mm, addr, addr + len);
+
+ if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
+ return -ENOMEM;
+ }
+
/* Clear old maps */
error = -ENOMEM;
munmap_back:
@@ -1442,10 +1488,6 @@ munmap_back:
goto munmap_back;
}
- /* Check against address space limit. */
- if (!may_expand_vm(mm, len >> PAGE_SHIFT))
- return -ENOMEM;
-
/*
* Private writable mapping: check memory availability
*/
--
1.8.1.5
See also a testsuite that exercies the newly added codepaths which is
attached as a tarball (All testcases minus the second that tests
that this patch works succeeds both before and after this patch).
--
Cyril Hrubis
[email protected]
On Tue, Apr 02, 2013 at 11:54:03AM +0200, Cyril Hrubis wrote:
> This patch fixes corner case for MAP_FIXED when requested mapping length
> is larger than rlimit for virtual memory. In such case any overlapping
> mappings are unmapped before we check for the limit and return ENOMEM.
>
> The check is moved before the loop that unmaps overlapping parts of
> existing mappings. When we are about to hit the limit (currently mapped
> pages + len > limit) we scan for overlapping pages and check again
> accounting for them.
>
> This fixes situation when userspace program expects that the previous
> mappings are preserved after the mmap() syscall has returned with error.
> (POSIX clearly states that successfull mapping shall replace any
> previous mappings.)
>
> This corner case was found and can be tested with LTP testcase:
>
> testcases/open_posix_testsuite/conformance/interfaces/mmap/24-2.c
>
> In this case the mmap, which is clearly over current limit, unmaps
> dynamic libraries and the testcase segfaults right after returning into
> userspace.
>
> I've also looked at the second instance of the unmapping loop in the
> do_brk(). The do_brk() is called from brk() syscall and from vm_brk().
> The brk() syscall checks for overlapping mappings and bails out when
> there are any (so it can't be triggered from the brk syscall). The
> vm_brk() is called only from binmft handlers so it shouldn't be
> triggered unless binmft handler created overlapping mappings.
>
> Signed-off-by: Cyril Hrubis <[email protected]>
Reviewed-by: Mel Gorman <[email protected]>
--
Mel Gorman
SUSE Labs
On Tue, 2 Apr 2013 11:54:03 +0200 Cyril Hrubis <[email protected]> wrote:
> This patch fixes corner case for MAP_FIXED when requested mapping length
> is larger than rlimit for virtual memory. In such case any overlapping
> mappings are unmapped before we check for the limit and return ENOMEM.
>
> The check is moved before the loop that unmaps overlapping parts of
> existing mappings. When we are about to hit the limit (currently mapped
> pages + len > limit) we scan for overlapping pages and check again
> accounting for them.
>
> This fixes situation when userspace program expects that the previous
> mappings are preserved after the mmap() syscall has returned with error.
> (POSIX clearly states that successfull mapping shall replace any
> previous mappings.)
>
> This corner case was found and can be tested with LTP testcase:
>
> testcases/open_posix_testsuite/conformance/interfaces/mmap/24-2.c
>
> In this case the mmap, which is clearly over current limit, unmaps
> dynamic libraries and the testcase segfaults right after returning into
> userspace.
>
> I've also looked at the second instance of the unmapping loop in the
> do_brk(). The do_brk() is called from brk() syscall and from vm_brk().
> The brk() syscall checks for overlapping mappings and bails out when
> there are any (so it can't be triggered from the brk syscall). The
> vm_brk() is called only from binmft handlers so it shouldn't be
> triggered unless binmft handler created overlapping mappings.
>
> ...
>
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -33,6 +33,7 @@
> #include <linux/uprobes.h>
> #include <linux/rbtree_augmented.h>
> #include <linux/sched/sysctl.h>
> +#include <linux/kernel.h>
>
> #include <asm/uaccess.h>
> #include <asm/cacheflush.h>
> @@ -543,6 +544,34 @@ static int find_vma_links(struct mm_struct *mm, unsigned long addr,
> return 0;
> }
>
> +static unsigned long count_vma_pages_range(struct mm_struct *mm,
> + unsigned long addr, unsigned long end)
> +{
> + unsigned long nr_pages = 0;
> + struct vm_area_struct *vma;
> +
> + /* Find first overlaping mapping */
> + vma = find_vma_intersection(mm, addr, end);
> + if (!vma)
> + return 0;
> +
> + nr_pages = (min(end, vma->vm_end) -
> + max(addr, vma->vm_start)) >> PAGE_SHIFT;
urgh, these things always make my head spin. Is it guaranteed that
end, vm_end, addr and vm_start are all multiples of PAGE_SIZE? If not,
we have a problem don't we?
Hi!
> > +static unsigned long count_vma_pages_range(struct mm_struct *mm,
> > + unsigned long addr, unsigned long end)
> > +{
> > + unsigned long nr_pages = 0;
> > + struct vm_area_struct *vma;
> > +
> > + /* Find first overlaping mapping */
> > + vma = find_vma_intersection(mm, addr, end);
> > + if (!vma)
> > + return 0;
> > +
> > + nr_pages = (min(end, vma->vm_end) -
> > + max(addr, vma->vm_start)) >> PAGE_SHIFT;
>
> urgh, these things always make my head spin. Is it guaranteed that
> end, vm_end, addr and vm_start are all multiples of PAGE_SIZE? If not,
> we have a problem don't we?
Yes, it takes a little of concentration before one can say what the code
does, unfortunatelly this is the most readable variant I've came up
with.
The len is page aligned right at the start of the do_mmap_pgoff() (end
is addr + len). The addr should be aligned in the get_unmapped_area()
although the codepath is more complicated to follow, but it seems to end
up in one of the arch_get_unmapped_area* and these makes sure the
address is aligned.
Moreover mmap() manual page says that the addr passed to mmap() is page
aligned (although I tend to check the code rather than the docs).
And for the vmas I belive these are page aligned by definition, correct
me if I'm wrong.
--
Cyril Hrubis
[email protected]