GSOC: Introduction

[Eugene Syromyatnikov]

> Thank you for the patch acceptance.
> Is the following list of open system calls complete ? It is based on
> reply that function calls that returns new file descriptor
>
> 1) mq_open
> 2) openat
> 3) perf_event_open
> 4) open_by_handle_at
> 5) memfd_create
> 6) epoll_create1
> 7) dup
> 8) dup3
> 9) fcntl64
> 10) timerfd_create
> 11) socket
> 12) accept
> 13) pipe2
As Dmitry reminded me, there are two interpretations; the one is quite
narrow — different architectures have both open/openat syscalls or
only one of them; %open may be used in order to track both (as old
code tends to use old one, moder uses openat and you ca never be
sure). The other one is quite wide — all syscalls that return new fd
(as I previously suggested), and in my mind the idea was to track all
new descriptors. But in this latter case, the question arises, whether
some unortodox (but quite utilised) ways of obtaining new fds, like
recvfrom or ioctl(NS_GET_PARENT, ...) or bpf(BPF_MAP_CREATE, ...)
should be covered. In my opinion, it makes the filter too complicated,
but event not considering them, there are also at least
timerfd_create, eventfd/eventfd2, inotify_init, signalfd,
fanotify_init.


-- 
Eugene Syromyatnikov
mailto:evgsyr at gmail.com
xmpp:esyr at jabber.{ru|org}