kernel/linux.git/fs/pidfs.c, branch v6.19.11

pidfs: return -EREMOTE when PIDFD_GET_INFO is called on another ns

2026-02-26T23:01:26+00:00

[ Upstream commit ab89060fbc92edd6e852bf0f533f29140afabe0e ] Currently it is not possible to distinguish between the case where a process has already exited and the case where a process is in a different namespace, as both return -ESRCH. glibc's pidfd_getpid() procfs-based implementation returns -EREMOTE in the latter, so that distinguishing the two is possible, as the fdinfo in procfs will list '0' as the PID in that case: https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/pidfd_getpid.c;h=860829cf07da2267484299ccb02861822c0d07b4;hb=HEAD#l121 Change the error code so that the kernel also returns -EREMOTE in that case. Fixes: 7477d7dce48a ("pidfs: allow to retrieve exit information") Signed-off-by: Luca Boccassi Link: https://patch.msgid.link/20260127225209.2293342-1-luca.boccassi@gmail.com Signed-off-by: Christian Brauner Signed-off-by: Sasha Levin

pidfs: protect PIDFD_GET_* ioctls() via ifdef

2026-01-06T22:08:12+00:00

We originally protected PIDFD_GET__NAMESPACE ioctls() through ifdefs and recent rework made it possible to drop them. There was an oversight though. When the relevant namespace is turned off ns->ops will be NULL so even though opening a file descriptor is perfectly legitimate it would fail during inode eviction when the file was closed. The simple fix would be to check ns->ops for NULL and continue allow to retrieve namespace fds from pidfds but we don't allow retrieving them when the relevant namespace type is turned off. So keep the simplification but add the ifdefs back in. Link: https://lore.kernel.org/20251222214907.GA189632@quark Link: https://patch.msgid.link/20251224-ununterbrochen-gagen-ea949b83f8f2@brauner Fixes: a71e4f103aed ("pidfs: simplify PIDFD_GET__NAMESPACE ioctls") Tested-by: Brendan Jackman Tested-by: Eric Biggers Reported-by: Eric Biggers Signed-off-by: Christian Brauner

Merge tag 'vfs-6.19-rc1.coredump' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs

2025-12-01T18:17:39+00:00

Pull pidfd and coredump updates from Christian Brauner: "Features: - Expose coredump signal via pidfd Expose the signal that caused the coredump through the pidfd interface. The recent changes to rework coredump handling to rely on unix sockets are in the process of being used in systemd. The previous systemd coredump container interface requires the coredump file descriptor and basic information including the signal number to be sent to the container. This means the signal number needs to be available before sending the coredump to the container. - Add supported_mask field to pidfd Add a new supported_mask field to struct pidfd_info that indicates which information fields are supported by the running kernel. This allows userspace to detect feature availability without relying on error codes or kernel version checks. Cleanups: - Drop struct pidfs_exit_info and prepare to drop exit_info pointer, simplifying the internal publication mechanism for exit and coredump information retrievable via the pidfd ioctl - Use guard() for task_lock in pidfs - Reduce wait_pidfd lock scope - Add missing PIDFD_INFO_SIZE_VER1 constant - Add missing BUILD_BUG_ON() assert on struct pidfd_info Fixes: - Fix PIDFD_INFO_COREDUMP handling Selftests: - Split out coredump socket tests and common helpers into separate files for better organization - Fix userspace coredump client detection issues - Handle edge-triggered epoll correctly - Ignore ENOSPC errors in tests - Add debug logging to coredump socket tests, socket protocol tests, and test helpers - Add tests for PIDFD_INFO_COREDUMP_SIGNAL - Add tests for supported_mask field - Update pidfd header for selftests" * tag 'vfs-6.19-rc1.coredump' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: (23 commits) pidfs: reduce wait_pidfd lock scope selftests/coredump: add second PIDFD_INFO_COREDUMP_SIGNAL test selftests/coredump: add first PIDFD_INFO_COREDUMP_SIGNAL test selftests/coredump: ignore ENOSPC errors selftests/coredump: add debug logging to coredump socket protocol tests selftests/coredump: add debug logging to coredump socket tests selftests/coredump: add debug logging to test helpers selftests/coredump: handle edge-triggered epoll correctly selftests/coredump: fix userspace coredump client detection selftests/coredump: fix userspace client detection selftests/coredump: split out coredump socket tests selftests/coredump: split out common helpers selftests/pidfd: add second supported_mask test selftests/pidfd: add first supported_mask test selftests/pidfd: update pidfd header pidfs: expose coredump signal pidfs: drop struct pidfs_exit_info pidfs: prepare to drop exit_info pointer pidfd: add a new supported_mask field pidfs: add missing BUILD_BUG_ON() assert on struct pidfd_info ...

pidfs: simplify PIDFD_GET__NAMESPACE ioctls

2025-11-17T15:23:13+00:00

We have reworked namespaces sufficiently that all this special-casing shouldn't be needed anymore Link: https://patch.msgid.link/20251117-eidesstattlich-apotheke-36d2e644079f@brauner Signed-off-by: Christian Brauner

pidfs: reduce wait_pidfd lock scope

2025-11-04T23:09:06+00:00

There's no need to hold the lock after we realized that pid->attr is set. We're holding a reference to struct pid so it won't go away and pidfs_exit() is called once per struct pid. Link: https://patch.msgid.link/20251105-work-pidfs-wait_pidfd-lock-v1-1-02638783be07@kernel.org Signed-off-by: Christian Brauner

pidfs: raise DCACHE_DONTCACHE explicitly

2025-10-31T09:16:24+00:00

While pidfs dentries are never hashed and thus retain_dentry() will never consider them for placing them on the LRU it isn't great to always have to go and remember that. Raise DCACHE_DONTCACHE explicitly as a visual marker that dentries aren't kept but freed immediately instead. Link: https://patch.msgid.link/20251029-work-namespace-nstree-listns-v4-4-2e6f823ebdc0@kernel.org Tested-by: syzbot@syzkaller.appspotmail.com Reviewed-by: Jeff Layton Signed-off-by: Christian Brauner

pidfs: expose coredump signal

2025-10-30T13:25:14+00:00

Userspace needs access to the signal that caused the coredump before the coredumping process has been reaped. Expose it as part of the coredump information in struct pidfd_info. After the process has been reaped that info is also available as part of PIDFD_INFO_EXIT's exit_code field. Link: https://patch.msgid.link/20251028-work-coredump-signal-v1-8-ca449b7b7aa0@kernel.org Reviewed-by: Alexander Mikhalitsyn Reviewed-by: Oleg Nesterov Signed-off-by: Christian Brauner

pidfs: drop struct pidfs_exit_info

2025-10-30T13:25:14+00:00

This is not needed anymore now that we have the new scheme to guarantee all-or-nothing information exposure. Link: https://patch.msgid.link/20251028-work-coredump-signal-v1-7-ca449b7b7aa0@kernel.org Reviewed-by: Alexander Mikhalitsyn Reviewed-by: Oleg Nesterov Signed-off-by: Christian Brauner

pidfs: prepare to drop exit_info pointer

2025-10-30T13:25:14+00:00

There will likely be more info that we need to store in struct pidfs_attr. We need to make sure that some of the information such as exit info or coredump info that consists of multiple bits is either available completely or not at all, but never partially. Currently we use a pointer that we assign to. That doesn't scale. We can't waste a pointer for each mulit-part information struct we want to expose. Use a bitmask instead. Link: https://patch.msgid.link/20251028-work-coredump-signal-v1-6-ca449b7b7aa0@kernel.org Reviewed-by: Alexander Mikhalitsyn Reviewed-by: Oleg Nesterov Signed-off-by: Christian Brauner

pidfd: add a new supported_mask field

2025-10-30T13:25:13+00:00

Some of the future fields in struct pidfd_info can be optional. If the kernel has nothing to emit in that field, then it doesn't set the flag in the reply. This presents a problem: There is currently no way to know what mask flags the kernel supports since one can't always count on them being in the reply. Add a new PIDFD_INFO_SUPPORTED_MASK flag and field that the kernel can set in the reply. Userspace can use this to determine if the fields it requires from the kernel are supported. This also gives us a way to deprecate fields in the future, if that should become necessary. Link: https://patch.msgid.link/20251028-work-coredump-signal-v1-5-ca449b7b7aa0@kernel.org Reviewed-by: Alexander Mikhalitsyn Reviewed-by: Oleg Nesterov Signed-off-by: Christian Brauner