[CRIU] [PATCH v2 00/24] exec: Move unshare_files and guarantee files_struct.count is correct

[Eric W. Biederman]

A while ago it was reported that posix file locking goes wrong when a
multi-threaded process calls exec.  I looked into the history and this
is definitely a regression, that should be fixed if we can.

This set of changes cleanups of the code in exec so hopefully this code
will not regress again.  Then it adds helpers and fixes the users of
files_struct so the reference count is only incremented if COPY_FILES is
passed to clone (or if io_uring takes a reference).  Then it removes
helpers (get_files_struct, __install_fd, __alloc_fd, __close_fd) that
are no longer needed and if used would encourage code that increments
the count of files_struct somewhere besides in clone when COPY_FILES is
passed.

In addition to fixing the bug in exec and simplifing the code this set
of changes by virtue of getting files_struct.count correct it optimizes
fdget.  With proc and other places not temporarily increasing the count
on files_struct __fget_light should succeed more often in being able to
return a struct file without touching it's reference count.

Fixing the count in files_struct was suggested by Oleg[1].

For those that are interested in the history of this issue I have
included as much of it as I could find in the first change.

Since v1:

- Renamed the functions
  __fcheck_files      -> files_lookup_fd_raw
  fcheck_files        -> files_lookup_fd_locked
  fcheck_files        -> files_lookup_fd_rcu
  fcheck_files        -> lookup_fd_rcu
  fcheck_task         -> task_lookup_fd_rcu
  fnext_task          -> task_lookup_next_fd_rcu
  __close_fd_get_file -> close_fd_get_file

- Simplified get_file_raw_ptr

- Removed ksys_close

- Examined the penalty for taking task_lock.  The helper
  task_lookup_next_fd_rcu takes task_lock each iteration.  Concern was
  expressed that this might be a problem.  The function tid_fd_mode
  isn called from tid_fd_revalidate which is called when ever a file
  descriptor file is stat'ed, opened, or otherwise accessed.  The
  function tid_fd_mode histrocally called get_files_struct which took
  and dropped task_lock.  So the volume of task_lock calls is already
  proportional to the number of file descriptors.  A micro benchmark
  did not see the move to task_lookup_next_fd_rcu making a difference
  in performance.  Which suggests that the change to taking the task
  lock for every file descriptor found in task_lookup_next_fd will not
  be a problem.

- Reviewed the code for conflicts with io_uring (especially the
  removal of get_files_struct).  To my surprise no conflicts were
  found as io_uring does not use standard helpers but instead rolls
  it's own version of get_files_struct by hand.

 Documentation/filesystems/files.rst          |   8 +-
 arch/powerpc/platforms/cell/spufs/coredump.c |   2 +-
 drivers/android/binder.c                     |   2 +-
 fs/autofs/dev-ioctl.c                        |   5 +-
 fs/coredump.c                                |   5 +-
 fs/exec.c                                    |  26 +++---
 fs/file.c                                    | 124 +++++++++++++--------------
 fs/io_uring.c                                |   2 +-
 fs/locks.c                                   |  14 +--
 fs/notify/dnotify/dnotify.c                  |   2 +-
 fs/open.c                                    |   2 +-
 fs/proc/fd.c                                 |  48 ++++-------
 include/linux/fdtable.h                      |  40 +++++----
 include/linux/syscalls.h                     |  12 ---
 kernel/bpf/syscall.c                         |  20 +----
 kernel/bpf/task_iter.c                       |  44 +++-------
 kernel/fork.c                                |  12 +--
 kernel/kcmp.c                                |  29 ++-----
 18 files changed, 152 insertions(+), 245 deletions(-)

Eric W. Biederman (24):
      exec: Move unshare_files to fix posix file locking during exec
      exec: Simplify unshare_files
      exec: Remove reset_files_struct
      kcmp: In kcmp_epoll_target use fget_task
      bpf: In bpf_task_fd_query use fget_task
      proc/fd: In proc_fd_link use fget_task
      file: Rename __fcheck_files to files_lookup_fd_raw
      file: Factor files_lookup_fd_locked out of fcheck_files
      file: Replace fcheck_files with files_lookup_fd_rcu
      file: Rename fcheck lookup_fd_rcu
      file: Implement task_lookup_fd_rcu
      proc/fd: In tid_fd_mode use task_lookup_fd_rcu
      kcmp: In get_file_raw_ptr use task_lookup_fd_rcu
      file: Implement task_lookup_next_fd_rcu
      proc/fd: In proc_readfd_common use task_lookup_next_fd_rcu
      bpf/task_iter: In task_file_seq_get_next use task_lookup_next_fd_rcu
      proc/fd: In fdinfo seq_show don't use get_files_struct
      file: Merge __fd_install into fd_install
      file: In f_dupfd read RLIMIT_NOFILE once.
      file: Merge __alloc_fd into alloc_fd
      file: Rename __close_fd to close_fd and remove the files parameter
      file: Replace ksys_close with close_fd
      file: Rename __close_fd_get_file close_fd_get_file
      file: Remove get_files_struct

[1] https://lkml.kernel.org/r/20180915160423.GA31461@redhat.com
v1: https://lkml.kernel.org/r/87ft8l6ic3.fsf@x220.int.ebiederm.org
Reported-by: Jeff Layton <jlayton at redhat.com>
Reported-by: Daniel P. Berrangé <berrange at redhat.com>
Suggested-by: Oleg Nesterov <oleg at redhat.com>
Signed-off-by: "Eric W. Biederman" <ebiederm at xmission.com>

p.s. I had hoped to publish this revised patchset a while ago but it is
a big one and life kept interrupting.  I have looked over everything and
I think I have addressed all of the review comments and have a good set
of patches that finally sort out posix file locking after exec.