Re: [RFC 1/3] x86/vdso: create vdso file, use it for mapping

From: Dmitry Safonov
Date: Sat Sep 03 2016 - 03:33:10 EST


Big thanks on review, Al!

2016-09-03 3:20 GMT+03:00 Al Viro <viro@xxxxxxxxxxxxxxxxxx>:
> On Thu, Aug 25, 2016 at 06:21:08PM +0300, Dmitry Safonov wrote:
>> + unsigned long n_addr = mmap_region(vdso_file_64, text_start,
>> + image->size, VM_READ|VM_EXEC|
>> + VM_DONTEXPAND|VM_SOFTDIRTY|
>> + VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC, 0);
>> + if (text_start != n_addr) {
>> + pr_err("Failed to mmap vdso file at %lx, mmap_region returned %lx\n",
>> + text_start, n_addr);
>> + goto old_way;
>> + }
>> + vma = find_vma(mm, text_start);
>> + if (IS_ERR(vma) || vma->vm_start != text_start) {
>> + pr_err("Failed to find vdso mapped vma at %lx\n",
>> + text_start);
>> + goto old_way;
>
> Umm... Since when can find_vma() return ERR_PTR()?

Will fix

>> + d_set_d_op(path.dentry, &vdso_dops);
>
> Nope. Set ->s_d_op to &vdso_dops and be done with that.

Ok

>> +static struct file_system_type vdso_fs_type = {
>> + .name = "vdsofs",
>> + .mount = ramfs_mount,
>
> Probably the wrong thing here. Just use a simple wrapper using mount_pseudo()
> for all work; see fs/aio.c:aio_mount().

Will check, thanks!

>> + ret = register_filesystem(&vdso_fs_type);
>
> Do you really want it user-mountable? If not, no need to register...

Of course, I don't want to, will omit that, thanks.

--
Dmitry