linux-user/syscall.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
The `event` parameter is ignored by the kernel if `op` is EPOLL_CTL_DEL,
do the same and avoid returning EFAULT if garbage is passed instead of a
valid pointer.
Signed-off-by: Giuseppe Musacchio <thatlemon@gmail.com>
---
linux-user/syscall.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 5e29e675e9..32d463d58d 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -11329,7 +11329,7 @@ static abi_long do_syscall1(void *cpu_env, int
num, abi_long arg1,
{
struct epoll_event ep;
struct epoll_event *epp = 0;
- if (arg4) {
+ if (arg2 != EPOLL_CTL_DEL && arg4) {
struct target_epoll_event *target_ep;
if (!lock_user_struct(VERIFY_READ, target_ep, arg4, 1)) {
return -TARGET_EFAULT;
--
2.20.1
Le 30/05/2019 à 17:25, Giuseppe Musacchio a écrit :
> The `event` parameter is ignored by the kernel if `op` is EPOLL_CTL_DEL,
> do the same and avoid returning EFAULT if garbage is passed instead of a
> valid pointer.
>
> Signed-off-by: Giuseppe Musacchio <thatlemon@gmail.com>
> ---
> linux-user/syscall.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index 5e29e675e9..32d463d58d 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -11329,7 +11329,7 @@ static abi_long do_syscall1(void *cpu_env, int
> num, abi_long arg1,
> {
> struct epoll_event ep;
> struct epoll_event *epp = 0;
> - if (arg4) {
> + if (arg2 != EPOLL_CTL_DEL && arg4) {
> struct target_epoll_event *target_ep;
> if (!lock_user_struct(VERIFY_READ, target_ep, arg4, 1)) {
> return -TARGET_EFAULT;
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Le 30/05/2019 à 18:00, Laurent Vivier a écrit :
> Le 30/05/2019 à 17:25, Giuseppe Musacchio a écrit :
>> The `event` parameter is ignored by the kernel if `op` is EPOLL_CTL_DEL,
>> do the same and avoid returning EFAULT if garbage is passed instead of a
>> valid pointer.
>>
>> Signed-off-by: Giuseppe Musacchio <thatlemon@gmail.com>
>> ---
>> linux-user/syscall.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
>> index 5e29e675e9..32d463d58d 100644
>> --- a/linux-user/syscall.c
>> +++ b/linux-user/syscall.c
>> @@ -11329,7 +11329,7 @@ static abi_long do_syscall1(void *cpu_env, int
>> num, abi_long arg1,
>> {
>> struct epoll_event ep;
>> struct epoll_event *epp = 0;
>> - if (arg4) {
>> + if (arg2 != EPOLL_CTL_DEL && arg4) {
>> struct target_epoll_event *target_ep;
>> if (!lock_user_struct(VERIFY_READ, target_ep, arg4, 1)) {
>> return -TARGET_EFAULT;
>
> Reviewed-by: Laurent Vivier <laurent@vivier.eu>
>
in fact, the BUGS section of epoll_ctl(2) says:
"In kernel versions before 2.6.9, the EPOLL_CTL_DEL operation required a
non-null pointer in event, even though this argument is ignored. Since
Linux 2.6.9, event can be specified as NULL when using EPOLL_CTL_DEL.
Applications that need to be portable to kernels before 2.6.9 should
specify a non-null pointer in event."
So something like this would be more portable:
@@ -11329,6 +11329,7 @@ static abi_long do_syscall1(void *cpu_env, int
num, abi_long arg1,
struct epoll_event ep;
struct epoll_event *epp = 0;
if (arg4) {
+ if (arg2 != EPOLL_CTL_DEL) {
struct target_epoll_event *target_ep;
if (!lock_user_struct(VERIFY_READ, target_ep, arg4, 1)) {
return -TARGET_EFAULT;
@@ -11340,6 +11341,11 @@ static abi_long do_syscall1(void *cpu_env, int
num, abi_long arg1,
*/
ep.data.u64 = tswap64(target_ep->data.u64);
unlock_user_struct(target_ep, arg4, 0);
+ }
+ /*
+ * before kernel 2.6.9, EPOLL_CTL_DEL operation required a
+ * non-null pointer, even though this argument is ignored.
+ * */
epp = &ep;
}
Thanks,
Laurent
Yes, that's much better if compatibility with such an old kernel version
is wanted.
I suppose there's no need for me to re-send the patch.
On 30/05/19 18:12, Laurent Vivier wrote:
> Le 30/05/2019 à 18:00, Laurent Vivier a écrit :
>> Le 30/05/2019 à 17:25, Giuseppe Musacchio a écrit :
>>> The `event` parameter is ignored by the kernel if `op` is EPOLL_CTL_DEL,
>>> do the same and avoid returning EFAULT if garbage is passed instead of a
>>> valid pointer.
>>>
>>> Signed-off-by: Giuseppe Musacchio <thatlemon@gmail.com>
>>> ---
>>> linux-user/syscall.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
>>> index 5e29e675e9..32d463d58d 100644
>>> --- a/linux-user/syscall.c
>>> +++ b/linux-user/syscall.c
>>> @@ -11329,7 +11329,7 @@ static abi_long do_syscall1(void *cpu_env, int
>>> num, abi_long arg1,
>>> {
>>> struct epoll_event ep;
>>> struct epoll_event *epp = 0;
>>> - if (arg4) {
>>> + if (arg2 != EPOLL_CTL_DEL && arg4) {
>>> struct target_epoll_event *target_ep;
>>> if (!lock_user_struct(VERIFY_READ, target_ep, arg4, 1)) {
>>> return -TARGET_EFAULT;
>>
>> Reviewed-by: Laurent Vivier <laurent@vivier.eu>
>>
>
> in fact, the BUGS section of epoll_ctl(2) says:
>
> "In kernel versions before 2.6.9, the EPOLL_CTL_DEL operation required a
> non-null pointer in event, even though this argument is ignored. Since
> Linux 2.6.9, event can be specified as NULL when using EPOLL_CTL_DEL.
> Applications that need to be portable to kernels before 2.6.9 should
> specify a non-null pointer in event."
>
> So something like this would be more portable:
>
> @@ -11329,6 +11329,7 @@ static abi_long do_syscall1(void *cpu_env, int
> num, abi_long arg1,
> struct epoll_event ep;
> struct epoll_event *epp = 0;
> if (arg4) {
> + if (arg2 != EPOLL_CTL_DEL) {
> struct target_epoll_event *target_ep;
> if (!lock_user_struct(VERIFY_READ, target_ep, arg4, 1)) {
> return -TARGET_EFAULT;
> @@ -11340,6 +11341,11 @@ static abi_long do_syscall1(void *cpu_env, int
> num, abi_long arg1,
> */
> ep.data.u64 = tswap64(target_ep->data.u64);
> unlock_user_struct(target_ep, arg4, 0);
> + }
> + /*
> + * before kernel 2.6.9, EPOLL_CTL_DEL operation required a
> + * non-null pointer, even though this argument is ignored.
> + * */
> epp = &ep;
> }
>
> Thanks,
> Laurent
>
© 2016 - 2026 Red Hat, Inc.