From: Luca Boccassi <bluca@xxxxxxxxxx> Date: Tue, 13 May 2025 02:09:24 +0100 > On Tue, 13 May 2025 at 01:18, Kuniyuki Iwashima <kuniyu@xxxxxxxxxx> wrote: > > > > From: Luca Boccassi <bluca@xxxxxxxxxx> > > Date: Mon, 12 May 2025 11:58:54 +0100 > > > On Mon, 12 May 2025 at 09:56, Christian Brauner <brauner@xxxxxxxxxx> wrote: > > > > > > > > Coredumping currently supports two modes: > > > > > > > > (1) Dumping directly into a file somewhere on the filesystem. > > > > (2) Dumping into a pipe connected to a usermode helper process > > > > spawned as a child of the system_unbound_wq or kthreadd. > > > > > > > > For simplicity I'm mostly ignoring (1). There's probably still some > > > > users of (1) out there but processing coredumps in this way can be > > > > considered adventurous especially in the face of set*id binaries. > > > > > > > > The most common option should be (2) by now. It works by allowing > > > > userspace to put a string into /proc/sys/kernel/core_pattern like: > > > > > > > > |/usr/lib/systemd/systemd-coredump %P %u %g %s %t %c %h > > > > > > > > The "|" at the beginning indicates to the kernel that a pipe must be > > > > used. The path following the pipe indicator is a path to a binary that > > > > will be spawned as a usermode helper process. Any additional parameters > > > > pass information about the task that is generating the coredump to the > > > > binary that processes the coredump. > > > > > > > > In the example core_pattern shown above systemd-coredump is spawned as a > > > > usermode helper. There's various conceptual consequences of this > > > > (non-exhaustive list): > > > > > > > > - systemd-coredump is spawned with file descriptor number 0 (stdin) > > > > connected to the read-end of the pipe. All other file descriptors are > > > > closed. That specifically includes 1 (stdout) and 2 (stderr). This has > > > > already caused bugs because userspace assumed that this cannot happen > > > > (Whether or not this is a sane assumption is irrelevant.). > > > > > > > > - systemd-coredump will be spawned as a child of system_unbound_wq. So > > > > it is not a child of any userspace process and specifically not a > > > > child of PID 1. It cannot be waited upon and is in a weird hybrid > > > > upcall which are difficult for userspace to control correctly. > > > > > > > > - systemd-coredump is spawned with full kernel privileges. This > > > > necessitates all kinds of weird privilege dropping excercises in > > > > userspace to make this safe. > > > > > > > > - A new usermode helper has to be spawned for each crashing process. > > > > > > > > This series adds a new mode: > > > > > > > > (3) Dumping into an abstract AF_UNIX socket. > > > > > > > > Userspace can set /proc/sys/kernel/core_pattern to: > > > > > > > > @address SO_COOKIE > > > > > > > > The "@" at the beginning indicates to the kernel that the abstract > > > > AF_UNIX coredump socket will be used to process coredumps. The address > > > > is given by @address and must be followed by the socket cookie of the > > > > coredump listening socket. > > > > > > > > The socket cookie is used to verify the socket connection. If the > > > > coredump server restarts or crashes and someone recycles the socket > > > > address the kernel will detect that the address has been recycled as the > > > > socket cookie will have necessarily changed and refuse to connect. > > > > > > This dynamic/cookie prefix makes it impossible to use this with socket > > > activation units. The way systemd-coredump works is that every > > > instance is an independent templated unit, spawned when there's a > > > connection to the private socket. If the path was fixed, we could just > > > reuse the same mechanism, it would fit very nicely with minimal > > > changes. > > > > Note this version does not use prefix. Now it requires users to > > just pass the socket cookie via core_pattern so that the kernel > > can verify the peer. > > Exactly - this means the pattern cannot be static in a sysctl.d early > on boot anymore, and has to be set dynamically by <something>. You missed the socket has to be created dynamically by <something>. > This is > a severe degradation over the status quo. > > > > But because you need a "server" to be permanently running, this means > > > socket-based activation can no longer work, and systemd-coredump must > > > switch to a persistently-running mode. > > > > The only thing for systemd to do is assign a cookie after socket creation. > > > > As long as systemd hold the file descriptor of the socket, you don't need > > a dedicated "server" running permanently, and the fd can be passed around > > to a spawned/activated process. > > There is no such facility, a socket is just a socket and there's no > infrastructure to randomly extract random information from one and > write it to some other random file in procfs, As only one socket can be registered to core_pattern, the socket must not be a random. > and I don't see why we > should add some super-special-case just for this, Because this is a new special use case. > it sounds really > messy. > Also sockets can be and in fact are routinely restarted (eg: on > package upgrades), which would invalidate this whole scheme, and > result in a very racy setup. When packages are upgraded it's one of > the most complex workflows in modern distros, and it's very likely > that things start crashing exactly at that point, and with this > workflow it would mean we'll lose core files due to the race between > restarting the socket unit and <something> updating the pattern > accordingly. Looks like you misunderstood the series. As you need to specify the socket in core_pattern, there must be only one socket that can receive core data, so the problem statement is always true throughout the series. kernel_connect() does not connect() to a random one out of sockets that have the common prefix. That's why the BPF was mentioned in the previous cover letter: - Since unix_stream_connect() runs bpf programs during connect it's possible to even redirect or multiplex coredumps to other sockets. > Also we very much want to be able to spawn as many core handlers at > the same time as needed, which I don't see how can work with a cookie > that has to be unique per socket. As said, you can just pass the fd of the coredump listener or a fd accept()ed from the listener, depending on how you want to handle this in userspace.
