This is the final piece of the series to establish io_uring support on
Linux.
MPD doesn't need io_uring for its efficient bulk I/O support, but to
allow file I/O to be cancelled. This is a big problem on CIFS/NFS
mounts where processes sleep uninterruptable if the file server
disappears, deadlocking MPD.
With io_uring, a flaky NFS connection allows MPD to continue to work
(even though there are still deadlocks inside MPD which need to be
addressed).
This plugin does not yet use cancellable `open()` using
`IORING_OP_OPENAT`. This will be implemented later.
Lots of other optimization opportunities for io_uring are still
missing as well - for example the database update could benefit a lot,
but unfortunately, io_uring doesn't have `readdir()` support just yet.
Reduces unstripped size. stripped size is the same.
Also took the time to remove using std::placeholders.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
The command is used to configure the client's connection, and this
shouldn't require any permissions. The client should be able to do
that before sending a password.
The parser implemented in libmpdclient requires the first key-value pair
of the server response to be the file pair. This is due to the fact that
libmpdclient scan pairs sequentially and first attempts to extract the
file pair before parsing the currentsong response further. See:
https://github.com/MusicPlayerDaemon/libmpdclient/blob/5c751a761ec9f71a43ee0e41cfff3208f31a58b8/src/song.c#L559-L563
Meta data encoded as pairs in the currentsong response will be ignored
if they are placed before the file pair in the response.
libnfs is compiled with `-D_FILE_OFFSET_BITS=64`, but Meson decides
not to enable this mode. We could force this mode, but then again,
these days, nobody should be using 32-bit Windows ... so this is a
kludge only for debugging with 32-bit WINE.
This fixes a freeze bug in the NFS input/storage plugins: when libnfs
auto-reconnets after a failure, it installs the new socket on the same
file descriptor number. MPD's attempt to unregister the old socket by
calling SocketMonitor::Steal() from NfsConnection::ScheduleSocket()
fails because the new/old socket number is not registered in epoll, so
epoll_ctl() returns ENOENT. The problem is that it left
`scheduled_flags`, and so subsequent Schedule() calls will use
`EPOLL_CTL_MOD`, which will fail again and again. Instead, we need to
use `EPOLL_CTL_ADD` to register the new socket.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/806
Closes https://github.com/MusicPlayerDaemon/MPD/issues/756
Max Kellermann