linux user signals leak file descriptors

[arr2036]

Under some conditions the user signal tests leak exactly one file descriptor. Seems like the issue is in the code itself not the tests.

The tests are broken too however, ./test/libkqueue-test user -n 2 results in

[/home/arr2036/Repositories/dependencies/libkqueue/test/user.c:109]: kevent_cmp() failed:
  expected [ident=2, filter=-11, flags = 16 (EV_ONESHOT), fflags = 0, data=0, udata=(nil)]
  but got  [ident=2, filter=-11, flags = 16 (EV_ONESHOT), fflags = 0, data=0, udata=(nil)]

which is interesting...

[arr2036]

Also if you change the tests to use their own ident numbers they start failing too.

[jpereira]

The repeated output looks to be due to the static buffer in kevent_to_str(). I did that test https://github.com/mheily/libkqueue/compare/master...jpereira:fix5?expand=1 and now is printing out correctly the difference between the buffers.

[arr2036]

Confirmed, thanks.

The actual error is:

[/home/arr2036/Documents/Repositories/dependencies/libkqueue/test/user.c:113]: kevent_cmp() failed:
expected [ident=2, filter=-11 (EVFILT_USER), flags = 16 (EV_ONESHOT), fflags=0x0000 (), data=0, udata=(nil)]
but got  [ident=2, filter=-11 (EVFILT_USER), flags = 32 (EV_CLEAR), fflags=0x0000 (), data=0, udata=(nil)]
***** ERROR: Program received signal 6 *****

[arr2036]

On run 1

KQ [22802]: kevent_copyin(): nchanges=1 nevents=0
KQ [22802]: kevent_copyin_one(): src={ ident=2, filter=-11 (EVFILT_USER), flags=0x0011 (EV_ADD EV_ONESHOT), fflags=0x0000 (), data=0, udata=(nil) }
KQ [22802]: linux_evfilt_user_knote_create(): event_fd=7 - created
KQ [22802]: kevent_copyin_one(): kn=0x5626ae3812f0 - created knote { ident=2, filter=-11 (EVFILT_USER), flags=0x0011 (EV_ADD EV_ONESHOT), fflags=0x0000 (), data=0, udata=(nil) }
KQ [22802]: kevent(): unlocked &(kq)->kq_mtx

Run 2

2802]: kevent_copyin(): nchanges=1 nevents=0
KQ [22802]: kevent_copyin_one(): src={ ident=2, filter=-11 (EVFILT_USER), flags=0x0011 (EV_ADD EV_ONESHOT), fflags=0x0000 (), data=0, udata=(nil) }
KQ [22802]: kevent_copyin_one(): kn=0x5626ae3813e0 - resolved ident=2 to knote
KQ [22802]: kevent_copyin_one(): kn=0x5626ae3813e0 - kn_modify rv=0
KQ [22802]: kevent(): unlocked &(kq)->kq_mtx

So I guess EV_ONESHOT isn't working correctly here.