| Message ID | 20230807062345.490-1-xuewen.yan@unisoc.com (mailing list archive) |
|---|---|
| State | RFC, archived |
| Headers | show |
| Series | PM: QOS: Always use "kstrtos32_from_user()" in cpu_latency_qos_write() | expand |
Hi rafael What do you think of this patch？ # echo 500 > /dev/cpu_dma_latency [T4893] write: qos value=170930229 [T4893] write: count value=4 the value is 170930229 is because it's hex is 0x0A303035, It is exactly the "500\0" # echo 200> /dev/cpu_dma_latency [T4893] write: qos value=170930226 [T4893] write: count value=4 the value is 170930226 is because it's hex is 0x0A303032, It is exactly the "200\0" Thanks! On Mon, Aug 7, 2023 at 3:17 PM Xuewen Yan <xuewen.yan@unisoc.com> wrote: > > In cpu_latency_qos_write, there is a judgment on whether the count > value is equal to sizeof(s32). This means that when user write 100~999, > it would get error value because it would call the "copy_from_user()" > instead of "kstrtos32". > Just like: > > # echo 500 > /dev/cpu_dma_latency > [T4893] write: qos value=170930229 > [T4893] write: count value=4 > > [T4893] write: qos value=170930226 > [T4893] write: count value=4 > > To prevent this happening, delete the "copy_from_user()" and > always use "kstrtos32_from_user()". > > Signed-off-by: Xuewen Yan <xuewen.yan@unisoc.com> > --- > kernel/power/qos.c | 14 ++++---------- > 1 file changed, 4 insertions(+), 10 deletions(-) > > diff --git a/kernel/power/qos.c b/kernel/power/qos.c > index 782d3b41c1f3..21a2f873e921 100644 > --- a/kernel/power/qos.c > +++ b/kernel/power/qos.c > @@ -379,17 +379,11 @@ static ssize_t cpu_latency_qos_write(struct file *filp, const char __user *buf, > size_t count, loff_t *f_pos) > { > s32 value; > + int ret; > > - if (count == sizeof(s32)) { > - if (copy_from_user(&value, buf, sizeof(s32))) > - return -EFAULT; > - } else { > - int ret; > - > - ret = kstrtos32_from_user(buf, count, 16, &value); > - if (ret) > - return ret; > - } > + ret = kstrtos32_from_user(buf, count, 16, &value); > + if (ret) > + return ret; > > cpu_latency_qos_update_request(filp->private_data, value); > > -- > 2.25.1 >
On 08/07/23 14:23, Xuewen Yan wrote: > In cpu_latency_qos_write, there is a judgment on whether the count > value is equal to sizeof(s32). This means that when user write 100~999, > it would get error value because it would call the "copy_from_user()" > instead of "kstrtos32". > Just like: > > # echo 500 > /dev/cpu_dma_latency > [T4893] write: qos value=170930229 > [T4893] write: count value=4 > > [T4893] write: qos value=170930226 > [T4893] write: count value=4 > > To prevent this happening, delete the "copy_from_user()" and > always use "kstrtos32_from_user()". > > Signed-off-by: Xuewen Yan <xuewen.yan@unisoc.com> > --- > kernel/power/qos.c | 14 ++++---------- > 1 file changed, 4 insertions(+), 10 deletions(-) > > diff --git a/kernel/power/qos.c b/kernel/power/qos.c > index 782d3b41c1f3..21a2f873e921 100644 > --- a/kernel/power/qos.c > +++ b/kernel/power/qos.c > @@ -379,17 +379,11 @@ static ssize_t cpu_latency_qos_write(struct file *filp, const char __user *buf, > size_t count, loff_t *f_pos) > { > s32 value; > + int ret; > > - if (count == sizeof(s32)) { > - if (copy_from_user(&value, buf, sizeof(s32))) > - return -EFAULT; In your test you use `echo` which produces a string. But from C, someone can write a program where buf contains s32 and not a string, and this will break those programs. If this expectations is no longer valid, then it's fine to remove it. Maybe we can change the order to call kstrtos32() first, but this is not bullet proof though.. Cheers -- Qais Yousef > - } else { > - int ret; > - > - ret = kstrtos32_from_user(buf, count, 16, &value); > - if (ret) > - return ret; > - } > + ret = kstrtos32_from_user(buf, count, 16, &value); > + if (ret) > + return ret; > > cpu_latency_qos_update_request(filp->private_data, value); > > -- > 2.25.1 >
diff --git a/kernel/power/qos.c b/kernel/power/qos.c index 782d3b41c1f3..21a2f873e921 100644 --- a/kernel/power/qos.c +++ b/kernel/power/qos.c @@ -379,17 +379,11 @@ static ssize_t cpu_latency_qos_write(struct file *filp, const char __user *buf, size_t count, loff_t *f_pos) { s32 value; + int ret; - if (count == sizeof(s32)) { - if (copy_from_user(&value, buf, sizeof(s32))) - return -EFAULT; - } else { - int ret; - - ret = kstrtos32_from_user(buf, count, 16, &value); - if (ret) - return ret; - } + ret = kstrtos32_from_user(buf, count, 16, &value); + if (ret) + return ret; cpu_latency_qos_update_request(filp->private_data, value);
In cpu_latency_qos_write, there is a judgment on whether the count value is equal to sizeof(s32). This means that when user write 100~999, it would get error value because it would call the "copy_from_user()" instead of "kstrtos32". Just like: # echo 500 > /dev/cpu_dma_latency [T4893] write: qos value=170930229 [T4893] write: count value=4 [T4893] write: qos value=170930226 [T4893] write: count value=4 To prevent this happening, delete the "copy_from_user()" and always use "kstrtos32_from_user()". Signed-off-by: Xuewen Yan <xuewen.yan@unisoc.com> --- kernel/power/qos.c | 14 ++++---------- 1 file changed, 4 insertions(+), 10 deletions(-)