Hi,
On 2022/10/27 15:30, Liam Howlett wrote:
* Li Zetao <lizetao1@xxxxxxxxxx> [221026 21:55]:
There is a memory leak reported by kmemleak:
unreferenced object 0xffff88817231ce40 (size 224):
comm "mount.cifs", pid 19308, jiffies 4295917571 (age 405.880s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
60 c0 b2 00 81 88 ff ff 98 83 01 42 81 88 ff ff `..........B....
backtrace:
[<ffffffff81936171>] __alloc_file+0x21/0x250
[<ffffffff81937051>] alloc_empty_file+0x41/0xf0
[<ffffffff81937159>] alloc_file+0x59/0x710
[<ffffffff81937964>] alloc_file_pseudo+0x154/0x210
[<ffffffff81741dbf>] __shmem_file_setup+0xff/0x2a0
[<ffffffff817502cd>] shmem_zero_setup+0x8d/0x160
[<ffffffff817cc1d5>] mmap_region+0x1075/0x19d0
[<ffffffff817cd257>] do_mmap+0x727/0x1110
[<ffffffff817518b2>] vm_mmap_pgoff+0x112/0x1e0
[<ffffffff83adf955>] do_syscall_64+0x35/0x80
[<ffffffff83c0006a>] entry_SYSCALL_64_after_hwframe+0x46/0xb0
The root cause was traced to an error handing path in mmap_region()
when arch_validate_flags() or mas_preallocate() fails. In the shared
anonymous mapping sence, vma will be setuped and mapped with a new
shared anonymous file via shmem_zero_setup(). So in this case, the
file resource needs to be released.
Fix it by calling fput(vma->vm_file) when arch_validate_flags() or
mas_preallocate() returns an error. And for the beauty of the code,
put fput() under mapping_unmap_writable().
It does look like the unrolling is in the wrong order in that section.
Fixes: d4af56c5c7c6 ("mm: start tracking VMAs with maple tree")
Fixes: c462ac288f2c ("mm: Introduce arch_validate_flags()")
Signed-off-by: Li Zetao <lizetao1@xxxxxxxxxx>
---
mm/mmap.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git a/mm/mmap.c b/mm/mmap.c
index e270057ed04e..8530195b3ec5 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2674,6 +2674,8 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
error = -EINVAL;
if (file)
goto close_and_free_vma;I will modify it in the second version
+ else if (vm_flags & VM_SHARED)
+ goto put_vma_file;
else
goto free_vma;
}
@@ -2682,6 +2684,8 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
error = -ENOMEM;
if (file)
goto close_and_free_vma;
+ else if (vm_flags & VM_SHARED)
+ goto put_vma_file;
else
goto free_vma;
}
I am not happy about this getting more complex as it already duplicates
too much code.
It's true that my modification would cause the code to further
complicate. I will rework it in the
next version.
@@ -2746,13 +2750,13 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
if (vma->vm_ops && vma->vm_ops->close)
vma->vm_ops->close(vma);
unmap_and_free_vma:
- fput(vma->vm_file);
- vma->vm_file = NULL;
-
/* Undo any partial mapping done by a device driver. */
unmap_region(mm, mas.tree, vma, prev, next, vma->vm_start, vma->vm_end);
if (vm_flags & VM_SHARED)
Could we change the above if to "if (file && vm_flags & VM_SHARED)" and
jump to unmap_and_free_vma "if (vma->vm_file)"? We could then drop your
new goto label. I still think the reodering is correct and worth while.
Or am I missing something?
Thanks for your constructive comments.
mapping_unmap_writable(file->f_mapping);
+put_vma_file:
+ fput(vma->vm_file);
+ vma->vm_file = NULL;
free_vma:
vm_area_free(vma);
unacct_error:
--
2.31.1
Regards,
Li Zetao
