ioctl is not called if cmd = 2

Question

I am developing a kernel module that uses unlocked_ioctl. I tested it with kernel version 2.6.24-23-generic and it works perfectly. Now I tried it with kernel version 3.3.1-1-ARCH and something weird happens: the ioctl function is not executed when the request value (cmd) is 2. It returns 0, but the function is not executed. In order to check that it is not executed I did the following:

static long midriver_ioctl(struct file *file,
    unsigned int cmd, unsigned long arg) {

printk("Called with cmd = %d\n", cmd);

I wrote a test program that calls ioctl for this device from 0 to 4096, and I can see in dmesg the message "Called with cmd = n" for all those values, except of "2", the only one that is not shown.

Any clues about what I am doing wrong?

Thank you in advance,

Some code up in the chain is intercepting cmd==2 and handling it? — zvrba, Apr 09 '12 at 09:32

score 11 · Accepted Answer · answered Apr 09 '12 at 09:32

Take a look on this:

 546 int do_vfs_ioctl(struct file *filp, unsigned int fd, unsigned int cmd,
 547             unsigned long arg)
 548 {
 549        int error = 0;
 550        int __user *argp = (int __user *)arg;
 551        struct inode *inode = filp->f_path.dentry->d_inode;
 552
 553        switch (cmd) {
 554        case FIOCLEX:
 555                set_close_on_exec(fd, 1);
 556                break;
 557
 558        case FIONCLEX:
 559                set_close_on_exec(fd, 0);
 560                break;
 561
 562        case FIONBIO:
 563                error = ioctl_fionbio(filp, argp);
 564                break;
 565
 566        case FIOASYNC:
 567                error = ioctl_fioasync(fd, filp, argp);
 568                break;
 569
 570        case FIOQSIZE:
 571                if (S_ISDIR(inode->i_mode) || S_ISREG(inode->i_mode) ||
 572                    S_ISLNK(inode->i_mode)) {
 573                        loff_t res = inode_get_bytes(inode);
 574                        error = copy_to_user(argp, &res, sizeof(res)) ?
 575                                        -EFAULT : 0;
 576                } else
 577                        error = -ENOTTY;
 578                break;
 579
 580        case FIFREEZE:
 581                error = ioctl_fsfreeze(filp);
 582                break;
 583
 584        case FITHAW:
 585                error = ioctl_fsthaw(filp);
 586                break;
 587
 588        case FS_IOC_FIEMAP:
 589                return ioctl_fiemap(filp, arg);
 590
 591        case FIGETBSZ:
 592                return put_user(inode->i_sb->s_blocksize, argp);
 593
 594        default:
 595                if (S_ISREG(inode->i_mode))
 596                        error = file_ioctl(filp, cmd, arg);
 597                else
 598                        error = vfs_ioctl(filp, cmd, arg);
 599                break;
 600        }
 601        return error;
 602

As you can see, there is some of the switch-cases before vfs_ioctl or file_ioctl call.

I fail to see the relevance, none of these cases apply – Hasturkun Apr 09 '12 at 09:40 — Hasturkun, Apr 09 '12 at 09:40

score 4 · Answer 2 · answered Apr 09 '12 at 09:45

4

As pointed out by @Ilya, You are hitting a generic case (FIGETBSZ in this case).

In general, you want to compose your ioctl commands using the _IO family of macros, with a unique type, to avoid collisions.

I suggest reading ioctl-number.txt from the kernel documentation for more information, including a list of most used types

answered Apr 09 '12 at 09:45

Hasturkun

35,395
6
71
104

Here is a minimal `ioctl` example using `_IOx`: https://stackoverflow.com/a/44613896/895245 – Ciro Santilli OurBigBook.com Jun 18 '17 at 10:11

ioctl is not called if cmd = 2

2 Answers2

Linked