[Bug 2037595] [NEW] _gpgme_io_select breaks when more than 1024 file descriptors are open

[Colin Watson]

Public bug reported:

[Impact]

Launchpad's buildd-manager is a complex event-driven daemon that manages
several hundred builders on production.  As such, it can often end up
with more than 1024 file descriptors open.  It also uses gpgme from time
to time to fetch signing keys for PPAs when dispatching builds that
involve PPAs.

On focal (and also on jammy), this breaks because `_gpgme_io_select` is
implemented using select(2), whose manual page says: "WARNING: select()
can monitor only file descriptors numbers that are less than FD_SETSIZE
(1024)—an unreasonably low limit for many modern applications—and this
limitation will not change.  All modern applications should instead use
poll(2) or epoll(7), which do not suffer this limitation."  This is
exactly the limitation we're running into, resulting in mysterious "Too
many open files" errors despite resource limits being set to more than
comfortable values.

Fortunately, this was fixed upstream in https://git.gnupg.org/cgi-
bin/gitweb.cgi?p=gpgme.git;a=commitdiff;h=8148237cb4ae20755c06a44d71761c7030973c3d,
and lunar has that fix.  focal and jammy do not (there may be a similar
problem with older LTSes too, but I only care about it back to focal).
buildd-manager currently runs on focal.

We've been running buildd-manager with a cherry-picked fix from upstream
for a few weeks now, and I'd like to get that SRUed back into Ubuntu so
that we don't accidentally lose this patch in the event of a security
update or similar.

[Test Plan]

William Grant provided this minimal reproducer:

```
root at tmp:~# python3
Python 3.8.10 (default, May 26 2023, 14:05:08)
[GCC 9.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import gpg
>>> c = gpg.Context()
>>> list(c.keylist())
[]
>>> fs = []
>>> for i in range(1024):
...     fs.append(open('/dev/null'))
...
>>> list(c.keylist())
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3/dist-packages/gpg/core.py", line 751, in keylist
    key = self.op_keylist_next()
  File "/usr/lib/python3/dist-packages/gpg/core.py", line 1195, in op_keylist_next
    raise excp
  File "/usr/lib/python3/dist-packages/gpg/core.py", line 1190, in op_keylist_next
    errorcheck(gpgme.gpgme_op_keylist_next(self.wrapped, ptr))
  File "/usr/lib/python3/dist-packages/gpg/errors.py", line 129, in errorcheck
    raise GPGMEError(retval, extradata)
gpg.errors.GPGMEError: GPGME: Too many open files
```

[Where problems could occur]

`_gpgme_io_select` is pretty central to gpgme, so if something is wrong
the package will be pretty broken.  On the other hand, that also means
any glaring problems are likely to be caught by the package's test
suite.

** Affects: gpgme1.0 (Ubuntu)
     Importance: Undecided
         Status: Fix Released

** Affects: gpgme1.0 (Ubuntu Focal)
     Importance: High
     Assignee: Colin Watson (cjwatson)
         Status: In Progress

** Affects: gpgme1.0 (Ubuntu Jammy)
     Importance: High
     Assignee: Colin Watson (cjwatson)
         Status: In Progress

** Also affects: gpgme1.0 (Ubuntu Jammy)
   Importance: Undecided
       Status: New

** Also affects: gpgme1.0 (Ubuntu Focal)
   Importance: Undecided
       Status: New

** Changed in: gpgme1.0 (Ubuntu)
       Status: New => Fix Released

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to gpgme1.0 in Ubuntu.
https://bugs.launchpad.net/bugs/2037595

Title:
  _gpgme_io_select breaks when more than 1024 file descriptors are open

Status in gpgme1.0 package in Ubuntu:
  Fix Released
Status in gpgme1.0 source package in Focal:
  In Progress
Status in gpgme1.0 source package in Jammy:
  In Progress

Bug description:
  [Impact]

  Launchpad's buildd-manager is a complex event-driven daemon that
  manages several hundred builders on production.  As such, it can often
  end up with more than 1024 file descriptors open.  It also uses gpgme
  from time to time to fetch signing keys for PPAs when dispatching
  builds that involve PPAs.

  On focal (and also on jammy), this breaks because `_gpgme_io_select`
  is implemented using select(2), whose manual page says: "WARNING:
  select() can monitor only file descriptors numbers that are less than
  FD_SETSIZE (1024)—an unreasonably low limit for many modern
  applications—and this limitation will not change.  All modern
  applications should instead use poll(2) or epoll(7), which do not
  suffer this limitation."  This is exactly the limitation we're running
  into, resulting in mysterious "Too many open files" errors despite
  resource limits being set to more than comfortable values.

  Fortunately, this was fixed upstream in https://git.gnupg.org/cgi-
  bin/gitweb.cgi?p=gpgme.git;a=commitdiff;h=8148237cb4ae20755c06a44d71761c7030973c3d,
  and lunar has that fix.  focal and jammy do not (there may be a
  similar problem with older LTSes too, but I only care about it back to
  focal).  buildd-manager currently runs on focal.

  We've been running buildd-manager with a cherry-picked fix from
  upstream for a few weeks now, and I'd like to get that SRUed back into
  Ubuntu so that we don't accidentally lose this patch in the event of a
  security update or similar.

  [Test Plan]

  William Grant provided this minimal reproducer:

  ```
  root at tmp:~# python3
  Python 3.8.10 (default, May 26 2023, 14:05:08)
  [GCC 9.4.0] on linux
  Type "help", "copyright", "credits" or "license" for more information.
  >>> import gpg
  >>> c = gpg.Context()
  >>> list(c.keylist())
  []
  >>> fs = []
  >>> for i in range(1024):
  ...     fs.append(open('/dev/null'))
  ...
  >>> list(c.keylist())
  Traceback (most recent call last):
    File "<stdin>", line 1, in <module>
    File "/usr/lib/python3/dist-packages/gpg/core.py", line 751, in keylist
      key = self.op_keylist_next()
    File "/usr/lib/python3/dist-packages/gpg/core.py", line 1195, in op_keylist_next
      raise excp
    File "/usr/lib/python3/dist-packages/gpg/core.py", line 1190, in op_keylist_next
      errorcheck(gpgme.gpgme_op_keylist_next(self.wrapped, ptr))
    File "/usr/lib/python3/dist-packages/gpg/errors.py", line 129, in errorcheck
      raise GPGMEError(retval, extradata)
  gpg.errors.GPGMEError: GPGME: Too many open files
  ```

  [Where problems could occur]

  `_gpgme_io_select` is pretty central to gpgme, so if something is
  wrong the package will be pretty broken.  On the other hand, that also
  means any glaring problems are likely to be caught by the package's
  test suite.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/gpgme1.0/+bug/2037595/+subscriptions