When p9pdu_readf is called with "s?d" attribute, it allocates a pointer
that will store a string. But when p9pdu_readf() fails while handling "d"
then this pointer will not be freed in p9_check_errors.
Fixes: ca41bb3e21d7 ("[net/9p] Handle Zero Copy TREAD/RERROR case in !dotl case.")
Signed-off-by: Hangyu Hua <[email protected]>
---
net/9p/client.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/net/9p/client.c b/net/9p/client.c
index 86bbc7147fc1..6c7cd765b714 100644
--- a/net/9p/client.c
+++ b/net/9p/client.c
@@ -540,12 +540,15 @@ static int p9_check_errors(struct p9_client *c, struct p9_req_t *req)
return 0;
if (!p9_is_proto_dotl(c)) {
- char *ename;
+ char *ename = NULL;
err = p9pdu_readf(&req->rc, c->proto_version, "s?d",
&ename, &ecode);
- if (err)
+ if (err) {
+ if (ename != NULL)
+ kfree(ename);
goto out_err;
+ }
if (p9_is_proto_dotu(c) && ecode < 512)
err = -ecode;
--
2.34.1
Hangyu Hua wrote on Thu, Oct 26, 2023 at 05:23:51PM +0800:
> When p9pdu_readf is called with "s?d" attribute, it allocates a pointer
> that will store a string. But when p9pdu_readf() fails while handling "d"
> then this pointer will not be freed in p9_check_errors.
Right, that sounds correct to me.
Out of curiosity how did you notice this? The leak shouldn't happen with
any valid server.
This cannot break anything so I'll push this to -next tomorrow and
submit to Linus next week
> Fixes: ca41bb3e21d7 ("[net/9p] Handle Zero Copy TREAD/RERROR case in !dotl case.")
This commit moves this code a bit, but the p9pdu_readf call predates
it -- in this case the Fixes tag is probably not useful; this affects
all maintained kernels.
> Signed-off-by: Hangyu Hua <[email protected]>
> ---
> net/9p/client.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/net/9p/client.c b/net/9p/client.c
> index 86bbc7147fc1..6c7cd765b714 100644
> --- a/net/9p/client.c
> +++ b/net/9p/client.c
> @@ -540,12 +540,15 @@ static int p9_check_errors(struct p9_client *c, struct p9_req_t *req)
> return 0;
>
> if (!p9_is_proto_dotl(c)) {
> - char *ename;
> + char *ename = NULL;
>
> err = p9pdu_readf(&req->rc, c->proto_version, "s?d",
> &ename, &ecode);
> - if (err)
> + if (err) {
> + if (ename != NULL)
> + kfree(ename);
Don't check for NULL before kfree - kfree does it.
If that's the only remark you get I can fix it when applying the commit
on my side.
> goto out_err;
> + }
>
> if (p9_is_proto_dotu(c) && ecode < 512)
> err = -ecode;
--
Dominique Martinet | Asmadeus
On Thursday, October 26, 2023 1:53:55 PM CEST [email protected] wrote:
>
> Hangyu Hua wrote on Thu, Oct 26, 2023 at 05:23:51PM +0800:
> > When p9pdu_readf is called with "s?d" attribute, it allocates a pointer
> > that will store a string. But when p9pdu_readf() fails while handling "d"
> > then this pointer will not be freed in p9_check_errors.
>
> Right, that sounds correct to me.
>
> Out of curiosity how did you notice this? The leak shouldn't happen with
> any valid server.
>
> This cannot break anything so I'll push this to -next tomorrow and
> submit to Linus next week
>
> > Fixes: ca41bb3e21d7 ("[net/9p] Handle Zero Copy TREAD/RERROR case in !dotl case.")
>
> This commit moves this code a bit, but the p9pdu_readf call predates
> it -- in this case the Fixes tag is probably not useful; this affects
> all maintained kernels.
Looks like it exists since introduction of p9_check_errors(), therefore:
Fixes: 51a87c552dfd ("9p: rework client code to use new protocol support functions")
> > Signed-off-by: Hangyu Hua <[email protected]>
> > ---
> > net/9p/client.c | 7 +++++--
> > 1 file changed, 5 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/9p/client.c b/net/9p/client.c
> > index 86bbc7147fc1..6c7cd765b714 100644
> > --- a/net/9p/client.c
> > +++ b/net/9p/client.c
> > @@ -540,12 +540,15 @@ static int p9_check_errors(struct p9_client *c, struct p9_req_t *req)
> > return 0;
> >
> > if (!p9_is_proto_dotl(c)) {
> > - char *ename;
> > + char *ename = NULL;
> >
> > err = p9pdu_readf(&req->rc, c->proto_version, "s?d",
> > &ename, &ecode);
> > - if (err)
> > + if (err) {
> > + if (ename != NULL)
> > + kfree(ename);
>
> Don't check for NULL before kfree - kfree does it.
> If that's the only remark you get I can fix it when applying the commit
> on my side.
With those two remarks addressed:
Reviewed-by: Christian Schoenebeck <[email protected]>
>
>
> > goto out_err;
> > + }
> >
> > if (p9_is_proto_dotu(c) && ecode < 512)
> > err = -ecode;
>
>
On 26/10/2023 19:53, [email protected] wrote:
>
> Hangyu Hua wrote on Thu, Oct 26, 2023 at 05:23:51PM +0800:
>> When p9pdu_readf is called with "s?d" attribute, it allocates a pointer
>> that will store a string. But when p9pdu_readf() fails while handling "d"
>> then this pointer will not be freed in p9_check_errors.
>
> Right, that sounds correct to me.
>
> Out of curiosity how did you notice this? The leak shouldn't happen with
> any valid server.
I just found that any attributes that require memory allocation are
prone to errors when mixed with ordinary attributes.
>
> This cannot break anything so I'll push this to -next tomorrow and
> submit to Linus next week
Agreed.
>
>> Fixes: ca41bb3e21d7 ("[net/9p] Handle Zero Copy TREAD/RERROR case in !dotl case.")
>
> This commit moves this code a bit, but the p9pdu_readf call predates
> it -- in this case the Fixes tag is probably not useful; this affects
> all maintained kernels.
>
>> Signed-off-by: Hangyu Hua <[email protected]>
>> ---
>> net/9p/client.c | 7 +++++--
>> 1 file changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/net/9p/client.c b/net/9p/client.c
>> index 86bbc7147fc1..6c7cd765b714 100644
>> --- a/net/9p/client.c
>> +++ b/net/9p/client.c
>> @@ -540,12 +540,15 @@ static int p9_check_errors(struct p9_client *c, struct p9_req_t *req)
>> return 0;
>>
>> if (!p9_is_proto_dotl(c)) {
>> - char *ename;
>> + char *ename = NULL;
>>
>> err = p9pdu_readf(&req->rc, c->proto_version, "s?d",
>> &ename, &ecode);
>> - if (err)
>> + if (err) {
>> + if (ename != NULL)
>> + kfree(ename);
>
> Don't check for NULL before kfree - kfree does it.
> If that's the only remark you get I can fix it when applying the commit
> on my side.
I get it. I will revise it based on your and Christian's comments and
send a v2.
Thanks,
Hangyu
>
>
>> goto out_err;
>> + }
>>
>> if (p9_is_proto_dotu(c) && ecode < 512)
>> err = -ecode;
>