Re: [PATCH] kvm tools:Fix memory leakage in open all disks

[Alexandru Elisei <alexandru.elisei@arm.com>]

Hi,

On Wed, Jul 10, 2024 at 04:12:37PM +0800, leixiang wrote:
> Dear Alex,
> Thank you for your reply and suggestions.
> 
> On 2024/7/9 18:12, Alexandru Elisei wrote:
> > Hi,
> > 
> > Adding the kvmtool maintainers (you can find them in the README file).
> > 
> > On Tue, Jun 18, 2024 at 03:52:47PM +0800, leixiang wrote:
> >> Fix memory leakage in disk/core disk_image__open_all when malloc disk failed,
> >> should free the disks that already malloced.
> >>
> >> Signed-off-by: Lei Xiang <leixiang@kylinos.cn>
> >> Suggested-by: Xie Ming <xieming@kylinos.cn>
> >> ---
> >>  disk/core.c | 6 ++++--
> >>  1 file changed, 4 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/disk/core.c b/disk/core.c
> >> index dd2f258..affeece 100644
> >> --- a/disk/core.c
> >> +++ b/disk/core.c
> >> @@ -195,8 +195,10 @@ static struct disk_image **disk_image__open_all(struct kvm *kvm)
> >>  
> >>  		if (wwpn) {
> >>  			disks[i] = malloc(sizeof(struct disk_image));
> >> -			if (!disks[i])
> >> -				return ERR_PTR(-ENOMEM);
> >> +			if (!disks[i]) {
> >> +				err = ERR_PTR(-ENOMEM);
> >> +				goto error;
> >> +			}
> >>  			disks[i]->wwpn = wwpn;
> >>  			disks[i]->tpgt = tpgt;
> > 
> > Currently, the latest patch on branch master is ca31abf5d9c3 ("arm64: Allow
> > the user to select the max SVE vector length"), and struct disk_image
> > doesn't have a tpgt field. Did you write this patch on a local branch?
> > 
> >>  			continue;
> > 
> There is no doubt that you are correct, I had realize that I git clone a wrong repo.
> > This is what the 'error' label does:
> > 
> > error:
> >         for (i = 0; i < count; i++)
> >                 if (!IS_ERR_OR_NULL(disks[i]))
> >                         disk_image__close(disks[i]);
> > 
> >         free(disks);
> >         return err;
> > 
> > And disk_image__close() ends up poking all sort of fields from struct
> > disk_image, including dereferencing pointers embedded in the struct. If
> > WWPN is specified for a disk, struct disk_image is allocated using malloc
> > as above, the field wwwpn is set and the rest of the fields are left
> > uninitialized. Because of this, calling disk_image__close() on a struct
> > disk_image with wwpn can lead to all sorts of nasty things happening.
> > 
> > May I suggest allocating disks[i] using calloc in the wwpn case to fix
> > this? Ideally, you would have two patches:
> > 
> > 1. A patch that changes the disk[i] allocation to calloc(), to prevent
> > disk_image__close() accessing unitialized fields when disk_image__open()
> > fails after initialized a WWPN disk.
> > 
> > 2. This patch.
> > 

> When the new disk_image is allocated successfully, 
> the fields will eventually be initialized by disk_image__new().
> And disk_image__close() accessing fields also checked before use.
> So I don't think it's necessary to replace malloc with calloc.

When and where is disk_image__new() called?

Thanks,
Alex