Skip to contentSkip to conversation

fanotify and pidfd?

DISCUSSION Marko Rauhamaa 2 participants 3 messages Jun 7, 2019 – Jun 8, 2019 VFS

lore
Share & tools
lore
#0 lore raw

As it stands, fanotify reports the process ID of the file that is being operated on:

           struct fanotify_event_metadata {
               __u32 event_len;
               __u8 vers;
               __u8 reserved;
               __u16 metadata_len;
               __aligned_u64 mask;
               __s32 fd;
               __s32 pid;
           };

One nasty problem with this is that the process often is long gone by the time the notification arrives.

Would it be possible to amend this format with:

               __s32 pidfd;

It would hold the pid still for the duration of notification processing and allow for the fanotify monitor to safely use the pid field to inspect /proc/<pid>.

And the possibility of sending signals to the monitored process might come in handy as well.

Thinking about this a bit more, could the fd field take on the dual role of allowing you to read the file in question as well as acting as a pidfd?

Marko

#1 lore raw
Quoted 15 lines · Marko Rauhamaa

As it stands, fanotify reports the process ID of the file that is being operated on:

           struct fanotify_event_metadata {
               __u32 event_len;
               __u8 vers;
               __u8 reserved;
               __u16 metadata_len;
               __aligned_u64 mask;
               __s32 fd;
               __s32 pid;
           };
Quoted 7 lines

One nasty problem with this is that the process often is long gone by the time the notification arrives.

Would it be possible to amend this format with:

               __s32 pidfd;

It's possible to report pidfd instead of pid with user opt-in (i.e. FAN_REPORT_PIDFD)

If you want to implement this, follow the footsteps of code, test and man page for FAN_REPORT_TID.

Quoted 4 lines

It would hold the pid still for the duration of notification processing and allow for the fanotify monitor to safely use the pid field to inspect /proc/<pid>.

It will hold pid until someone closes pidfd.
That also needs to be documented.

Quoted 7 lines

And the possibility of sending signals to the monitored process might come in handy as well.

Thinking about this a bit more, could the fd field take on the dual role of allowing you to read the file in question as well as acting as a pidfd?

Please no.

Thanks,
Amir.

#2 lore raw

Amir Goldstein <[email protected]>:

Quoted 4 lines
Quoted 2 lines · Marko Rauhamaa

Would it be possible to amend this format with:

Quoted 8 lines
Quoted 2 lines
               __s32 pidfd;

It's possible to report pidfd instead of pid with user opt-in (i.e. FAN_REPORT_PIDFD)

If you want to implement this, follow the footsteps of code, test and man page for FAN_REPORT_TID.

I hear you.

Marko