Fix retrieve user value in IOCTL_VALSET_NUM case

[applepwc]

The test program is:

    arg.val = 0xAA;
    ioctl(fd, IOCTL_VALSET_NUM, &arg);

    arg.val = 0;
    ioctl(fd, IOCTL_VALGET_NUM, &arg);
    printf("Read val: 0x%x\n", arg.val);

---

Summary by cubic

Fix IOCTL_VALSET_NUM to read the integer via get_user from the user pointer instead of using arg directly. Returns -EFAULT on bad pointers, prevents invalid memory access, and makes the set/get test pass.

^{Written for commit a507630. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 1 file

[jserv]

Remove "fix: " in the subject of git commit messages.
Refine git commit message, making it more informative. See https://chris.beams.io/git-commit
Don't close this pull request. Use git force-push instead.

[linD026]

To be compatible with the old version, which might still use the C89 standard.
Don't declare the variable inside the switch statement.
Or, you should declare the variable within the block scope.

Also, if you want to declare the variable inside the switch statement, please fix this:

ioctl.c:78:9: error: typename in expression
ioctl.c:78:13: error: Expected ; at end of statement
ioctl.c:78:13: error: got user_val
ioctl.c:78:9: error: undefined identifier 'int'
ioctl.c:79:13: error: undefined identifier 'user_val'
ioctl.c:81:21: error: undefined identifier 'user_val'

[applepwc]

Hello, I force-pushed a new commit. Please review it.

[visitorckw]

Hello, I force-pushed a new commit. Please review it.

Github notifies the relevant people automatically, so there's no need to leave a comment like this.

[visitorckw]

Thanks for the patch.

While this addresses the pointer-to-value mismatch, should we consider whether IOCTL_VALSET_NUM is better suited for "pass-by-value" rather than "pass-by-reference"?

Currently, the macro is defined as:

#define IOCTL_VALSET_NUM _IOW(IOC_MAGIC, 3, int)

Using _IOW implies that arg is a user-space pointer. However, for a simple integer assignment, would it be more idiomatic to use pass-by-value by redefining the macro with _IO? This would allow us to keep the logic lightweight without the overhead of memory copying.

[jserv]

Ensure consistent code style with 'clang-format' in examples directory.

[applepwc]

Yes, using _IO with pass-by-value is more idiomatic and avoids unnecessary memory copying.
But this would be an ABI change: existing user-space code calling ioctl(fd, IOCTL_VALSET_NUM, &value) would break.

[visitorckw]

Wouldn't the proposed patch also change the existing ABI?

[applepwc]

Hello, I'm uncertain whether to add more code as shown below or simply rewrite the original.

#define IOCTL_VALSET_NUM2 _IO(IOC_MAGIC, 4)
#define IOCTL_VAL_MAXNR 4
...
case IOCTL_VALSET_NUM2:
    ioctl_num = (int)arg; 
    break;