Skip to contentSkip to conversation

Nonotify 0.3.2 (A simple dnotify replacement)

DISCUSSION Pascal Schmidt 2 participants 3 messages Aug 22, 2004 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

On Sun, 22 Aug 2004 05:40:06 +0200, you wrote in linux.kernel:

Quoted 7 lines

2) The /dev/nonotify device:

/dev/nonotify has the only purpose to offer a special stat() call via ioctl to read the contents_mtime field of directories (together with atime, mtime, ctime). The client has to set the 'filename' field of the 'nonotify_stat' structure and receives the four timespec fields updated via ioctl.

A lot of people here (Linus, for instance) frown on ioctl() interfaces. They're hard to do right in 32/64bit compat layers, for example. How about using a syscall interface instead?

Signature · 2 lines
Ciao,
Pascal
#2 lore raw
Quoted 10 lines · Pascal Schmidt

On Sun, 22 Aug 2004 05:40:06 +0200, you wrote in linux.kernel:

Quoted 7 lines

2) The /dev/nonotify device:

/dev/nonotify has the only purpose to offer a special stat() call via ioctl to read the contents_mtime field of directories (together with atime, mtime, ctime). The client has to set the 'filename' field of the 'nonotify_stat' structure and receives the four timespec fields updated via ioctl.

Quoted 3 lines

A lot of people here (Linus, for instance) frown on ioctl() interfaces. They're hard to do right in 32/64bit compat layers, for example. How about using a syscall interface instead?

Nonotify uses ioctl mainly for 'pragmatical' reasons. A syscall would be technically better - that's for sure, actually it was my initial idea to use one (or to change the stat-call).

But i didn't want to bother people with asking to assign me a syscall-number before even knowing if they like my idea. And changing it to use a syscall lateron would be no problem at all from the concept of nonotify.

Also - as a kernel newbie - i didn't find any good documentation how to add my own syscall into the kernel. I just wanted to get nonotify working as an 'optional patch', without changing tons of files.

But you could help me if you have a look at my ioctl function. I'm using a structure which contains a char* pointer and four timespec fields. Do you know if this causes problems with 32/64bit compatibility.

Thanks,

Norbert

#3 lore raw
Quoted 4 lines · nf

But i didn't want to bother people with asking to assign me a syscall-number before even knowing if they like my idea. And changing it to use a syscall lateron would be no problem at all from the concept of nonotify.

Right. It's just something to consider if you aim for inclusion in the mainline tree later.

Quoted 3 lines

Also - as a kernel newbie - i didn't find any good documentation how to add my own syscall into the kernel. I just wanted to get nonotify working as an 'optional patch', without changing tons of files.

Adding a syscall should roughly work like this:

- write your sys_foo() function
- add your syscall number to include/asm-i386/unistd.h, don't forget to change the "#define NR_syscalls" accordingly
- add ".long sys_foo" to sys_call_table in arch/i386/kernel/entry.S, in the right place (i.e. syscall 290 would need to be element 290 in the list)
- arrange for the object file with sys_foo in it to be linked into the kernel

You're right, your going to have to touch a couple of files for this approach.

Quoted 3 lines

But you could help me if you have a look at my ioctl function. I'm using a structure which contains a char* pointer and four timespec fields. Do you know if this causes problems with 32/64bit compatibility.

I don't think the timespecs will be a problem. However, the char pointer will have different sizes and alignment requirements on 32bit and 64bit system. I'm no expert in how that can be handled in the compat layer, though.

Signature · 2 lines
Ciao,
Pascal