The simple attribute files do not accept a negative value since the
commit 488dac0c9237 ("libfs: fix error cast of negative value in
simple_attr_write()"), so we can no longer use DEFINE_SIMPLE_ATTRIBUTE() to
define a file operations for errno value.
Fixes: 488dac0c9237 ("libfs: fix error cast of negative value in simple_attr_write()")
Reported-by: Zhao Gongyi <[email protected]>
Signed-off-by: Akinobu Mita <[email protected]>
---
v2: Fix Reported-by line
lib/notifier-error-inject.c | 38 ++++++++++++++++++++++++++++++-------
1 file changed, 31 insertions(+), 7 deletions(-)
diff --git a/lib/notifier-error-inject.c b/lib/notifier-error-inject.c
index 21016b32d313..30ec41f58d53 100644
--- a/lib/notifier-error-inject.c
+++ b/lib/notifier-error-inject.c
@@ -3,20 +3,44 @@
#include "notifier-error-inject.h"
-static int debugfs_errno_set(void *data, u64 val)
+static int notifier_err_errno_show(struct seq_file *m, void *data)
{
- *(int *)data = clamp_t(int, val, -MAX_ERRNO, 0);
+ int *value = m->private;
+
+ seq_printf(m, "%d\n", *value);
+
return 0;
}
-static int debugfs_errno_get(void *data, u64 *val)
+static int notifier_err_errno_open(struct inode *inode, struct file *file)
{
- *val = *(int *)data;
- return 0;
+ return single_open(file, notifier_err_errno_show, inode->i_private);
+}
+
+static ssize_t notifier_err_errno_write(struct file *file, const char __user *ubuf, size_t len,
+ loff_t *offp)
+{
+ struct seq_file *m = file->private_data;
+ int *value = m->private;
+ int ret;
+
+ ret = kstrtoint_from_user(ubuf, len, 0, value);
+ if (ret)
+ return ret;
+
+ *value = clamp(*value, -MAX_ERRNO, 0);
+
+ return len;
}
-DEFINE_SIMPLE_ATTRIBUTE(fops_errno, debugfs_errno_get, debugfs_errno_set,
- "%lld\n");
+static const struct file_operations fops_errno = {
+ .owner = THIS_MODULE,
+ .open = notifier_err_errno_open,
+ .read = seq_read,
+ .write = notifier_err_errno_write,
+ .llseek = seq_lseek,
+ .release = single_release,
+};
static struct dentry *debugfs_create_errno(const char *name, umode_t mode,
struct dentry *parent, int *value)
--
2.34.1
On 17.09.22 09:14, Akinobu Mita wrote:
> The simple attribute files do not accept a negative value since the
> commit 488dac0c9237 ("libfs: fix error cast of negative value in
> simple_attr_write()"), so we can no longer use DEFINE_SIMPLE_ATTRIBUTE() to
> define a file operations for errno value.
>
> Fixes: 488dac0c9237 ("libfs: fix error cast of negative value in simple_attr_write()")
> Reported-by: Zhao Gongyi <[email protected]>
> Signed-off-by: Akinobu Mita <[email protected]>
But shouldn't we fix simple_attr_write() instead?
I mean, simple_attr_read() might use attr->fmt to print a signed value,
but simple_attr_write() fails on signed values now?
I might be wrong, but there is a disconnect. I feel like
simple_attr_write() should similarly make decisions based on attr->fmt.
> ---
> v2: Fix Reported-by line
>
> lib/notifier-error-inject.c | 38 ++++++++++++++++++++++++++++++-------
> 1 file changed, 31 insertions(+), 7 deletions(-)
>
> diff --git a/lib/notifier-error-inject.c b/lib/notifier-error-inject.c
> index 21016b32d313..30ec41f58d53 100644
> --- a/lib/notifier-error-inject.c
> +++ b/lib/notifier-error-inject.c
> @@ -3,20 +3,44 @@
>
> #include "notifier-error-inject.h"
>
> -static int debugfs_errno_set(void *data, u64 val)
> +static int notifier_err_errno_show(struct seq_file *m, void *data)
> {
> - *(int *)data = clamp_t(int, val, -MAX_ERRNO, 0);
> + int *value = m->private;
> +
> + seq_printf(m, "%d\n", *value);
> +
> return 0;
> }
>
> -static int debugfs_errno_get(void *data, u64 *val)
> +static int notifier_err_errno_open(struct inode *inode, struct file *file)
> {
> - *val = *(int *)data;
> - return 0;
> + return single_open(file, notifier_err_errno_show, inode->i_private);
> +}
> +
> +static ssize_t notifier_err_errno_write(struct file *file, const char __user *ubuf, size_t len,
> + loff_t *offp)
> +{
> + struct seq_file *m = file->private_data;
> + int *value = m->private;
> + int ret;
> +
> + ret = kstrtoint_from_user(ubuf, len, 0, value);
> + if (ret)
> + return ret;
> +
> + *value = clamp(*value, -MAX_ERRNO, 0);
> +
> + return len;
> }
>
> -DEFINE_SIMPLE_ATTRIBUTE(fops_errno, debugfs_errno_get, debugfs_errno_set,
> - "%lld\n");
> +static const struct file_operations fops_errno = {
> + .owner = THIS_MODULE,
> + .open = notifier_err_errno_open,
> + .read = seq_read,
> + .write = notifier_err_errno_write,
> + .llseek = seq_lseek,
> + .release = single_release,
> +};
>
> static struct dentry *debugfs_create_errno(const char *name, umode_t mode,
> struct dentry *parent, int *value)
--
Thanks,
David / dhildenb
2022年9月19日(月) 18:20 David Hildenbrand <[email protected]>:
>
> On 17.09.22 09:14, Akinobu Mita wrote:
> > The simple attribute files do not accept a negative value since the
> > commit 488dac0c9237 ("libfs: fix error cast of negative value in
> > simple_attr_write()"), so we can no longer use DEFINE_SIMPLE_ATTRIBUTE() to
> > define a file operations for errno value.
> >
> > Fixes: 488dac0c9237 ("libfs: fix error cast of negative value in simple_attr_write()")
> > Reported-by: Zhao Gongyi <[email protected]>
> > Signed-off-by: Akinobu Mita <[email protected]>
>
> But shouldn't we fix simple_attr_write() instead?
>
> I mean, simple_attr_read() might use attr->fmt to print a signed value,
> but simple_attr_write() fails on signed values now?
>
> I might be wrong, but there is a disconnect. I feel like
> simple_attr_write() should similarly make decisions based on attr->fmt.
I agree there is a disconnect, but I have no idea how to fix
simple_attr_write().
(strcmp(attr->fmt, "%%lld\n") is ugly)
If no one seems to come up with a good idea, I'll fix the similar problems
in fault-injection the same way I did here.
On 19.09.22 13:40, Akinobu Mita wrote:
> 2022年9月19日(月) 18:20 David Hildenbrand <[email protected]>:
>>
>> On 17.09.22 09:14, Akinobu Mita wrote:
>>> The simple attribute files do not accept a negative value since the
>>> commit 488dac0c9237 ("libfs: fix error cast of negative value in
>>> simple_attr_write()"), so we can no longer use DEFINE_SIMPLE_ATTRIBUTE() to
>>> define a file operations for errno value.
>>>
>>> Fixes: 488dac0c9237 ("libfs: fix error cast of negative value in simple_attr_write()")
>>> Reported-by: Zhao Gongyi <[email protected]>
>>> Signed-off-by: Akinobu Mita <[email protected]>
>>
>> But shouldn't we fix simple_attr_write() instead?
>>
>> I mean, simple_attr_read() might use attr->fmt to print a signed value,
>> but simple_attr_write() fails on signed values now?
>>
>> I might be wrong, but there is a disconnect. I feel like
>> simple_attr_write() should similarly make decisions based on attr->fmt.
>
> I agree there is a disconnect, but I have no idea how to fix
> simple_attr_write().
> (strcmp(attr->fmt, "%%lld\n") is ugly)
> If no one seems to come up with a good idea, I'll fix the similar problems
> in fault-injection the same way I did here.
>
Maybe we simply want another interface for the handful of %lld users. Or
another way to allow DEFINE_SIMPLE_ATTRIBUTE users to specify it.
Might be good enough to specify instead of/in addition to "%llu" an enum
that expresses what we want.
$ git grep -C2 DEFINE_SIMPLE | grep "\%"
tells me that we have
a) %llu
b) %llx
c) %lld
d) 0x%llx
e) 0x%08llx
Maybe we can adjust the debugfs cases to be more similar, to eventually
get only a, b and c.
--
Thanks,
David / dhildenb