[PATCH v5 0/9] coredump: add coredump socket

[PATCH v5 0/9] coredump: add coredump socket

[Christian Brauner]

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:

        @linuxafsk/coredump_socket

The "@" at the beginning indicates to the kernel that the abstract
AF_UNIX coredump socket will be used to process coredumps.

The coredump socket uses the fixed address "linuxafsk/coredump.socket"
for now.

The coredump socket is located in the initial network namespace. To bind
the coredump socket userspace must hold CAP_SYS_ADMIN in the initial
user namespace. Listening and reading can happen from whatever
unprivileged context is necessary to safely process coredumps.

When a task coredumps it opens a client socket in the initial network
namespace and connects to the coredump socket.

- The coredump server should use SO_PEERPIDFD to get a stable handle on
  the connected crashing task. The retrieved pidfd will provide a stable
  reference even if the crashing task gets SIGKILLed while generating
  the coredump.

- When a coredump connection is initiated use the socket cookie as the
  coredump cookie and store it in the pidfd. The receiver can now easily
  authenticate that the connection is coming from the kernel.

  Unless the coredump server expects to handle connection from
  non-crashing task it can validate that the connection has been made from
  a crashing task:

     fd_coredump = accept4(fd_socket, NULL, NULL, SOCK_CLOEXEC);
     getsockopt(fd_coredump, SOL_SOCKET, SO_PEERPIDFD, &fd_peer_pidfd, &fd_peer_pidfd_len);

     struct pidfd_info info = {
             info.mask = PIDFD_INFO_EXIT | PIDFD_INFO_COREDUMP,
     };

     ioctl(pidfd, PIDFD_GET_INFO, &info);
     /* Refuse connections that aren't from a crashing task. */
     if (!(info.mask & PIDFD_INFO_COREDUMP) || !(info.coredump_mask & PIDFD_COREDUMPED) )
             close(fd_coredump);

     /*
      * Make sure that the coredump cookie matches the connection cookie.
      * If they don't it's not the coredump connection from the kernel.
      * We'll get another connection request in a bit.
      */
     getsocketop(fd_coredump, SOL_SOCKET, SO_COOKIE, &peer_cookie, &peer_cookie_len);
     if (!info.coredump_cookie || (info.coredump_cookie != peer_cookie))
             close(fd_coredump);

  The kernel guarantees that by the time the connection is made the
  coredump info is available.

- By setting core_pipe_limit non-zero userspace can guarantee that the
  crashing task cannot be reaped behind it's back and thus process all
  necessary information in /proc/<pid>. The SO_PEERPIDFD can be used to
  detect whether /proc/<pid> still refers to the same process.

  The core_pipe_limit isn't used to rate-limit connections to the
  socket. This can simply be done via AF_UNIX socket directly.

- The pidfd for the crashing task will contain information how the task
  coredumps. The PIDFD_GET_INFO ioctl gained a new flag
  PIDFD_INFO_COREDUMP which can be used to retreive the coredump
  information.

  If the coredump gets a new coredump client connection the kernel
  guarantees that PIDFD_INFO_COREDUMP information is available.
  Currently the following information is provided in the new
  @coredump_mask extension to struct pidfd_info:

  * PIDFD_COREDUMPED is raised if the task did actually coredump.
  * PIDFD_COREDUMP_SKIP	is raised if the task skipped coredumping (e.g.,
    undumpable).
  * PIDFD_COREDUMP_USER	is raised if this is a regular coredump and
    doesn't need special care by the coredump server.
  * IDFD_COREDUMP_ROOT is raised if the generated coredump should be
    treated as sensitive and the coredump server should restrict to the
    generated coredump to sufficiently privileged users.

- Since unix_stream_connect() runs bpf programs during connect it's
  possible to even redirect or multiplex coredumps to other sockets.

- The coredump server should mark itself as non-dumpable.
  To capture coredumps for the coredump server itself a bpf program
  should be run at connect to redirect it to another socket in
  userspace. This can be useful for debugging crashing coredump servers.

- A container coredump server in a separate network namespace can simply
  bind to linuxafsk/coredump.socket and systemd-coredump fowards
  coredumps to the container.

- Fwiw, one idea is to handle coredumps via per-user/session coredump
  servers that run with that users privileges.

  The coredump server listens on the coredump socket and accepts a
  new coredump connection. It then retrieves SO_PEERPIDFD for the
  client, inspects uid/gid and hands the accepted client to the users
  own coredump handler which runs with the users privileges only.

The new coredump socket will allow userspace to not have to rely on
usermode helpers for processing coredumps and provides a safer way to
handle them instead of relying on super privileged coredumping helpers.

This will also be significantly more lightweight since no fork()+exec()
for the usermodehelper is required for each crashing process. The
coredump server in userspace can just keep a worker pool.

This is easy to test:

(a) coredump processing (we're using socat):

    > cat coredump_socket.sh
    #!/bin/bash

    set -x

    sudo bash -c "echo '@linuxafsk/coredump.socket' > /proc/sys/kernel/core_pattern"
    sudo socat --statistics abstract-listen:linuxafsk/coredump.socket,fork FILE:core_file,create,append,trunc

(b) trigger a coredump:

    user1@localhost:~/data/scripts$ cat crash.c
    #include <stdio.h>
    #include <unistd.h>

    int main(int argc, char *argv[])
    {
            fprintf(stderr, "%u\n", (1 / 0));
            _exit(0);
    }

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
Changes in v5:
- Don't use a prefix just the specific address.
- Link to v4: https://lore.kernel.org/20250507-work-coredump-socket-v4-0-af0ef317b2d0@kernel.org

Changes in v4:
- Expose the coredump socket cookie through the pidfd. This allows the
  coredump server to easily recognize coredump socket connections.
- Link to v3: https://lore.kernel.org/20250505-work-coredump-socket-v3-0-e1832f0e1eae@kernel.org

Changes in v3:
- Use an abstract unix socket.
- Add documentation.
- Add selftests.
- Link to v2: https://lore.kernel.org/20250502-work-coredump-socket-v2-0-43259042ffc7@kernel.org

Changes in v2:
- Expose dumpability via PIDFD_GET_INFO.
- Place COREDUMP_SOCK handling under CONFIG_UNIX.
- Link to v1: https://lore.kernel.org/20250430-work-coredump-socket-v1-0-2faf027dbb47@kernel.org

---
Christian Brauner (9):
      coredump: massage format_corname()
      coredump: massage do_coredump()
      coredump: reflow dump helpers a little
      coredump: add coredump socket
      pidfs, coredump: add PIDFD_INFO_COREDUMP
      coredump: show supported coredump modes
      coredump: validate socket name as it is written
      selftests/pidfd: add PIDFD_INFO_COREDUMP infrastructure
      selftests/coredump: add tests for AF_UNIX coredumps

 fs/coredump.c                                     | 395 +++++++++++++++++-----
 fs/pidfs.c                                        |  75 ++++
 include/linux/coredump.h                          |  14 +
 include/linux/net.h                               |   1 +
 include/linux/pidfs.h                             |  10 +
 include/uapi/linux/pidfd.h                        |  22 ++
 net/unix/af_unix.c                                |  28 +-
 tools/testing/selftests/coredump/stackdump_test.c | 231 ++++++++++++-
 tools/testing/selftests/pidfd/pidfd.h             |  23 ++
 9 files changed, 703 insertions(+), 96 deletions(-)
---
base-commit: 4dd6566b5a8ca1e8c9ff2652c2249715d6c64217
change-id: 20250429-work-coredump-socket-87cc0f17729c

[PATCH v5 1/9] coredump: massage format_corname()

[Christian Brauner]

We're going to extend the coredump code in follow-up patches.
Clean it up so we can do this more easily.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c | 41 ++++++++++++++++++++++++-----------------
 1 file changed, 24 insertions(+), 17 deletions(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index d740a0411266..281320ea351f 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -76,9 +76,15 @@ static char core_pattern[CORENAME_MAX_SIZE] = "core";
 static int core_name_size = CORENAME_MAX_SIZE;
 unsigned int core_file_note_size_limit = CORE_FILE_NOTE_SIZE_DEFAULT;
 
+enum coredump_type_t {
+	COREDUMP_FILE = 1,
+	COREDUMP_PIPE = 2,
+};
+
 struct core_name {
 	char *corename;
 	int used, size;
+	enum coredump_type_t core_type;
 };
 
 static int expand_corename(struct core_name *cn, int size)
@@ -218,18 +224,21 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 {
 	const struct cred *cred = current_cred();
 	const char *pat_ptr = core_pattern;
-	int ispipe = (*pat_ptr == '|');
 	bool was_space = false;
 	int pid_in_pattern = 0;
 	int err = 0;
 
 	cn->used = 0;
 	cn->corename = NULL;
+	if (*pat_ptr == '|')
+		cn->core_type = COREDUMP_PIPE;
+	else
+		cn->core_type = COREDUMP_FILE;
 	if (expand_corename(cn, core_name_size))
 		return -ENOMEM;
 	cn->corename[0] = '\0';
 
-	if (ispipe) {
+	if (cn->core_type == COREDUMP_PIPE) {
 		int argvs = sizeof(core_pattern) / 2;
 		(*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
 		if (!(*argv))
@@ -247,7 +256,7 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 		 * Split on spaces before doing template expansion so that
 		 * %e and %E don't get split if they have spaces in them
 		 */
-		if (ispipe) {
+		if (cn->core_type == COREDUMP_PIPE) {
 			if (isspace(*pat_ptr)) {
 				if (cn->used != 0)
 					was_space = true;
@@ -353,7 +362,7 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 				 * Installing a pidfd only makes sense if
 				 * we actually spawn a usermode helper.
 				 */
-				if (!ispipe)
+				if (!(cn->core_type != COREDUMP_PIPE))
 					break;
 
 				/*
@@ -384,12 +393,12 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 	 * If core_pattern does not include a %p (as is the default)
 	 * and core_uses_pid is set, then .%pid will be appended to
 	 * the filename. Do not do this for piped commands. */
-	if (!ispipe && !pid_in_pattern && core_uses_pid) {
+	if (!(cn->core_type == COREDUMP_PIPE) && !pid_in_pattern && core_uses_pid) {
 		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
 		if (err)
 			return err;
 	}
-	return ispipe;
+	return 0;
 }
 
 static int zap_process(struct signal_struct *signal, int exit_code)
@@ -583,7 +592,6 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 	const struct cred *old_cred;
 	struct cred *cred;
 	int retval = 0;
-	int ispipe;
 	size_t *argv = NULL;
 	int argc = 0;
 	/* require nonrelative corefile path and be extra careful */
@@ -632,19 +640,18 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 
 	old_cred = override_creds(cred);
 
-	ispipe = format_corename(&cn, &cprm, &argv, &argc);
+	retval = format_corename(&cn, &cprm, &argv, &argc);
+	if (retval < 0) {
+		coredump_report_failure("format_corename failed, aborting core");
+		goto fail_unlock;
+	}
 
-	if (ispipe) {
+	if (cn.core_type == COREDUMP_PIPE) {
 		int argi;
 		int dump_count;
 		char **helper_argv;
 		struct subprocess_info *sub_info;
 
-		if (ispipe < 0) {
-			coredump_report_failure("format_corename failed, aborting core");
-			goto fail_unlock;
-		}
-
 		if (cprm.limit == 1) {
 			/* See umh_coredump_setup() which sets RLIMIT_CORE = 1.
 			 *
@@ -695,7 +702,7 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 			coredump_report_failure("|%s pipe failed", cn.corename);
 			goto close_fail;
 		}
-	} else {
+	} else if (cn.core_type == COREDUMP_FILE) {
 		struct mnt_idmap *idmap;
 		struct inode *inode;
 		int open_flags = O_CREAT | O_WRONLY | O_NOFOLLOW |
@@ -823,13 +830,13 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 		file_end_write(cprm.file);
 		free_vma_snapshot(&cprm);
 	}
-	if (ispipe && core_pipe_limit)
+	if ((cn.core_type == COREDUMP_PIPE) && core_pipe_limit)
 		wait_for_dump_helpers(cprm.file);
 close_fail:
 	if (cprm.file)
 		filp_close(cprm.file, NULL);
 fail_dropcount:
-	if (ispipe)
+	if (cn.core_type == COREDUMP_PIPE)
 		atomic_dec(&core_dump_count);
 fail_unlock:
 	kfree(argv);

-- 
2.47.2

[PATCH v5 2/9] coredump: massage do_coredump()

[Christian Brauner]

We're going to extend the coredump code in follow-up patches.
Clean it up so we can do this more easily.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c | 122 +++++++++++++++++++++++++++++++---------------------------
 1 file changed, 65 insertions(+), 57 deletions(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index 281320ea351f..41491dbfafdf 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -646,63 +646,8 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 		goto fail_unlock;
 	}
 
-	if (cn.core_type == COREDUMP_PIPE) {
-		int argi;
-		int dump_count;
-		char **helper_argv;
-		struct subprocess_info *sub_info;
-
-		if (cprm.limit == 1) {
-			/* See umh_coredump_setup() which sets RLIMIT_CORE = 1.
-			 *
-			 * Normally core limits are irrelevant to pipes, since
-			 * we're not writing to the file system, but we use
-			 * cprm.limit of 1 here as a special value, this is a
-			 * consistent way to catch recursive crashes.
-			 * We can still crash if the core_pattern binary sets
-			 * RLIM_CORE = !1, but it runs as root, and can do
-			 * lots of stupid things.
-			 *
-			 * Note that we use task_tgid_vnr here to grab the pid
-			 * of the process group leader.  That way we get the
-			 * right pid if a thread in a multi-threaded
-			 * core_pattern process dies.
-			 */
-			coredump_report_failure("RLIMIT_CORE is set to 1, aborting core");
-			goto fail_unlock;
-		}
-		cprm.limit = RLIM_INFINITY;
-
-		dump_count = atomic_inc_return(&core_dump_count);
-		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
-			coredump_report_failure("over core_pipe_limit, skipping core dump");
-			goto fail_dropcount;
-		}
-
-		helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
-					    GFP_KERNEL);
-		if (!helper_argv) {
-			coredump_report_failure("%s failed to allocate memory", __func__);
-			goto fail_dropcount;
-		}
-		for (argi = 0; argi < argc; argi++)
-			helper_argv[argi] = cn.corename + argv[argi];
-		helper_argv[argi] = NULL;
-
-		retval = -ENOMEM;
-		sub_info = call_usermodehelper_setup(helper_argv[0],
-						helper_argv, NULL, GFP_KERNEL,
-						umh_coredump_setup, NULL, &cprm);
-		if (sub_info)
-			retval = call_usermodehelper_exec(sub_info,
-							  UMH_WAIT_EXEC);
-
-		kfree(helper_argv);
-		if (retval) {
-			coredump_report_failure("|%s pipe failed", cn.corename);
-			goto close_fail;
-		}
-	} else if (cn.core_type == COREDUMP_FILE) {
+	switch (cn.core_type) {
+	case COREDUMP_FILE: {
 		struct mnt_idmap *idmap;
 		struct inode *inode;
 		int open_flags = O_CREAT | O_WRONLY | O_NOFOLLOW |
@@ -796,6 +741,69 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 		if (do_truncate(idmap, cprm.file->f_path.dentry,
 				0, 0, cprm.file))
 			goto close_fail;
+		break;
+	}
+	case COREDUMP_PIPE: {
+		int argi;
+		int dump_count;
+		char **helper_argv;
+		struct subprocess_info *sub_info;
+
+		if (cprm.limit == 1) {
+			/* See umh_coredump_setup() which sets RLIMIT_CORE = 1.
+			 *
+			 * Normally core limits are irrelevant to pipes, since
+			 * we're not writing to the file system, but we use
+			 * cprm.limit of 1 here as a special value, this is a
+			 * consistent way to catch recursive crashes.
+			 * We can still crash if the core_pattern binary sets
+			 * RLIM_CORE = !1, but it runs as root, and can do
+			 * lots of stupid things.
+			 *
+			 * Note that we use task_tgid_vnr here to grab the pid
+			 * of the process group leader.  That way we get the
+			 * right pid if a thread in a multi-threaded
+			 * core_pattern process dies.
+			 */
+			coredump_report_failure("RLIMIT_CORE is set to 1, aborting core");
+			goto fail_unlock;
+		}
+		cprm.limit = RLIM_INFINITY;
+
+		dump_count = atomic_inc_return(&core_dump_count);
+		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
+			coredump_report_failure("over core_pipe_limit, skipping core dump");
+			goto fail_dropcount;
+		}
+
+		helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
+					    GFP_KERNEL);
+		if (!helper_argv) {
+			coredump_report_failure("%s failed to allocate memory", __func__);
+			goto fail_dropcount;
+		}
+		for (argi = 0; argi < argc; argi++)
+			helper_argv[argi] = cn.corename + argv[argi];
+		helper_argv[argi] = NULL;
+
+		retval = -ENOMEM;
+		sub_info = call_usermodehelper_setup(helper_argv[0],
+						helper_argv, NULL, GFP_KERNEL,
+						umh_coredump_setup, NULL, &cprm);
+		if (sub_info)
+			retval = call_usermodehelper_exec(sub_info,
+							  UMH_WAIT_EXEC);
+
+		kfree(helper_argv);
+		if (retval) {
+			coredump_report_failure("|%s pipe failed", cn.corename);
+			goto close_fail;
+		}
+		break;
+	}
+	default:
+		WARN_ON_ONCE(true);
+		goto close_fail;
 	}
 
 	/* get us an unshared descriptor table; almost always a no-op */

-- 
2.47.2

[PATCH v5 3/9] coredump: reflow dump helpers a little

[Christian Brauner]

They look rather messy right now.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index 41491dbfafdf..b2eda7b176e4 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -867,10 +867,9 @@ static int __dump_emit(struct coredump_params *cprm, const void *addr, int nr)
 	struct file *file = cprm->file;
 	loff_t pos = file->f_pos;
 	ssize_t n;
+
 	if (cprm->written + nr > cprm->limit)
 		return 0;
-
-
 	if (dump_interrupted())
 		return 0;
 	n = __kernel_write(file, addr, nr, &pos);
@@ -887,20 +886,21 @@ static int __dump_skip(struct coredump_params *cprm, size_t nr)
 {
 	static char zeroes[PAGE_SIZE];
 	struct file *file = cprm->file;
+
 	if (file->f_mode & FMODE_LSEEK) {
-		if (dump_interrupted() ||
-		    vfs_llseek(file, nr, SEEK_CUR) < 0)
+		if (dump_interrupted() || vfs_llseek(file, nr, SEEK_CUR) < 0)
 			return 0;
 		cprm->pos += nr;
 		return 1;
-	} else {
-		while (nr > PAGE_SIZE) {
-			if (!__dump_emit(cprm, zeroes, PAGE_SIZE))
-				return 0;
-			nr -= PAGE_SIZE;
-		}
-		return __dump_emit(cprm, zeroes, nr);
 	}
+
+	while (nr > PAGE_SIZE) {
+		if (!__dump_emit(cprm, zeroes, PAGE_SIZE))
+			return 0;
+		nr -= PAGE_SIZE;
+	}
+
+	return __dump_emit(cprm, zeroes, nr);
 }
 
 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)

-- 
2.47.2

[PATCH v5 4/9] coredump: add coredump socket

[Christian Brauner]

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:

        @linuxafsk/coredump_socket

The "@" at the beginning indicates to the kernel that the abstract
AF_UNIX coredump socket will be used to process coredumps.

The coredump socket uses the fixed address "linuxafsk/coredump.socket"
for now.

The coredump socket is located in the initial network namespace. To bind
the coredump socket userspace must hold CAP_SYS_ADMIN in the initial
user namespace. Listening and reading can happen from whatever
unprivileged context is necessary to safely process coredumps.

When a task coredumps it opens a client socket in the initial network
namespace and connects to the coredump socket.

- The coredump server uses SO_PEERPIDFD to get a stable handle on the
  connected crashing task. The retrieved pidfd will provide a stable
  reference even if the crashing task gets SIGKILLed while generating
  the coredump.

- By setting core_pipe_limit non-zero userspace can guarantee that the
  crashing task cannot be reaped behind it's back and thus process all
  necessary information in /proc/<pid>. The SO_PEERPIDFD can be used to
  detect whether /proc/<pid> still refers to the same process.

  The core_pipe_limit isn't used to rate-limit connections to the
  socket. This can simply be done via AF_UNIX sockets directly.

- The pidfd for the crashing task will grow new information how the task
  coredumps.

- The coredump server should mark itself as non-dumpable.
  To capture coredumps for the coredump server itself a bpf program
  should be run at connect to redirect it to another socket in
  userspace. This can be useful for debugging crashing coredump servers.

- A container coredump server in a separate network namespace can simply
  bind to another well-know address and systemd-coredump fowards
  coredumps to the container.

- Coredumps could in the future also be handled via per-user/session
  coredump servers that run only with that users privileges.

  The coredump server listens on the coredump socket and accepts a
  new coredump connection. It then retrieves SO_PEERPIDFD for the
  client, inspects uid/gid and hands the accepted client to the users
  own coredump handler which runs with the users privileges only
  (It must of coure pay close attention to not forward crashing suid
  binaries.).

The new coredump socket will allow userspace to not have to rely on
usermode helpers for processing coredumps and provides a safer way to
handle them instead of relying on super privileged coredumping helpers
that have and continue to cause significant CVEs.

This will also be significantly more lightweight since no fork()+exec()
for the usermodehelper is required for each crashing process. The
coredump server in userspace can e.g., just keep a worker pool.

This is easy to test:

(a) coredump processing (we're using socat):

    > cat coredump_socket.sh
    #!/bin/bash

    set -x

    sudo bash -c "echo '@linuxafsk/coredump.socket' > /proc/sys/kernel/core_pattern"
    sudo socat --statistics abstract-listen:linuxafsk/coredump.socket,fork FILE:core_file,create,append,trunc

(b) trigger a coredump:

    user1@localhost:~/data/scripts$ cat crash.c
    #include <stdio.h>
    #include <unistd.h>

    int main(int argc, char *argv[])
    {
            fprintf(stderr, "%u\n", (1 / 0));
            _exit(0);
    }

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c            | 155 ++++++++++++++++++++++++++++++++++++++++++++---
 include/linux/coredump.h |  14 +++++
 net/unix/af_unix.c       |  21 +++++--
 3 files changed, 178 insertions(+), 12 deletions(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index b2eda7b176e4..d3599d671c51 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -44,7 +44,11 @@
 #include <linux/sysctl.h>
 #include <linux/elf.h>
 #include <linux/pidfs.h>
+#include <linux/net.h>
+#include <linux/socket.h>
+#include <net/net_namespace.h>
 #include <uapi/linux/pidfd.h>
+#include <uapi/linux/un.h>
 
 #include <linux/uaccess.h>
 #include <asm/mmu_context.h>
@@ -79,6 +83,7 @@ unsigned int core_file_note_size_limit = CORE_FILE_NOTE_SIZE_DEFAULT;
 enum coredump_type_t {
 	COREDUMP_FILE = 1,
 	COREDUMP_PIPE = 2,
+	COREDUMP_SOCK = 3,
 };
 
 struct core_name {
@@ -232,13 +237,16 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 	cn->corename = NULL;
 	if (*pat_ptr == '|')
 		cn->core_type = COREDUMP_PIPE;
+	else if (*pat_ptr == '@')
+		cn->core_type = COREDUMP_SOCK;
 	else
 		cn->core_type = COREDUMP_FILE;
 	if (expand_corename(cn, core_name_size))
 		return -ENOMEM;
 	cn->corename[0] = '\0';
 
-	if (cn->core_type == COREDUMP_PIPE) {
+	switch (cn->core_type) {
+	case COREDUMP_PIPE: {
 		int argvs = sizeof(core_pattern) / 2;
 		(*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
 		if (!(*argv))
@@ -247,6 +255,34 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 		++pat_ptr;
 		if (!(*pat_ptr))
 			return -ENOMEM;
+		break;
+	}
+	case COREDUMP_SOCK: {
+		err = cn_printf(cn, "%s", pat_ptr);
+		if (err)
+			return err;
+
+		/*
+		 * We can potentially allow this to be changed later but
+		 * I currently see no reason to.
+		 */
+		if (strcmp(cn->corename, "@linuxafsk/coredump.socket"))
+			return -EINVAL;
+
+		/*
+		 * Currently no need to parse any other options.
+		 * Relevant information can be retrieved from the peer
+		 * pidfd retrievable via SO_PEERPIDFD by the receiver or
+		 * via /proc/<pid>, using the SO_PEERPIDFD to guard
+		 * against pid recycling when opening /proc/<pid>.
+		 */
+		return 0;
+	}
+	case COREDUMP_FILE:
+		break;
+	default:
+		WARN_ON_ONCE(true);
+		return -EINVAL;
 	}
 
 	/* Repeat as long as we have more pattern to process and more output
@@ -393,11 +429,20 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 	 * If core_pattern does not include a %p (as is the default)
 	 * and core_uses_pid is set, then .%pid will be appended to
 	 * the filename. Do not do this for piped commands. */
-	if (!(cn->core_type == COREDUMP_PIPE) && !pid_in_pattern && core_uses_pid) {
-		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
-		if (err)
-			return err;
+	if (!pid_in_pattern && core_uses_pid) {
+		switch (cn->core_type) {
+		case COREDUMP_FILE:
+			return cn_printf(cn, ".%d", task_tgid_vnr(current));
+		case COREDUMP_PIPE:
+			break;
+		case COREDUMP_SOCK:
+			break;
+		default:
+			WARN_ON_ONCE(true);
+			return -EINVAL;
+		}
 	}
+
 	return 0;
 }
 
@@ -583,6 +628,37 @@ static int umh_coredump_setup(struct subprocess_info *info, struct cred *new)
 	return 0;
 }
 
+#ifdef CONFIG_UNIX
+static const struct sockaddr_un coredump_unix_socket = {
+	.sun_family = AF_UNIX,
+	.sun_path = "\0linuxafsk/coredump.socket",
+};
+/* Without trailing NUL byte. */
+#define COREDUMP_UNIX_SOCKET_ADDR_SIZE            \
+	(offsetof(struct sockaddr_un, sun_path) + \
+	 sizeof("\0linuxafsk/coredump.socket") - 1)
+
+/*
+ * Ensure that only a process privileged over the initial network
+ * namespace can bind the coredump server address. Protect us against
+ * the coredump server crashing, or intentionally restarting and an
+ * unprivileged process binding the coredump server address to receive
+ * coredumps.
+ */
+int unix_may_bind_coredump_addr(struct net *net, struct sockaddr_un *sunname, int len)
+{
+	if (net != &init_net)
+		return 0;
+	if (COREDUMP_UNIX_SOCKET_ADDR_SIZE != len)
+		return 0;
+	if (memcmp(&coredump_unix_socket, sunname, len))
+		return 0;
+	if (ns_capable(net->user_ns, CAP_NET_ADMIN))
+		return 0;
+	return -ECONNREFUSED;
+}
+#endif
+
 void do_coredump(const kernel_siginfo_t *siginfo)
 {
 	struct core_state core_state;
@@ -801,6 +877,45 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 		}
 		break;
 	}
+	case COREDUMP_SOCK: {
+#ifdef CONFIG_UNIX
+		struct file *file __free(fput) = NULL;
+		struct socket *socket;
+
+		/*
+		 * It is possible that the userspace process which is
+		 * supposed to handle the coredump and is listening on
+		 * the AF_UNIX socket coredumps. Userspace should just
+		 * mark itself non dumpable.
+		 */
+
+		retval = sock_create_kern(&init_net, AF_UNIX, SOCK_STREAM, 0, &socket);
+		if (retval < 0)
+			goto close_fail;
+
+		file = sock_alloc_file(socket, 0, NULL);
+		if (IS_ERR(file)) {
+			sock_release(socket);
+			goto close_fail;
+		}
+
+		retval = kernel_connect(socket,
+					(struct sockaddr *)(&coredump_unix_socket),
+					COREDUMP_UNIX_SOCKET_ADDR_SIZE, O_NONBLOCK);
+		if (retval) {
+			if (retval == -EAGAIN)
+				coredump_report_failure("Skipping as coredump socket connection %s couldn't complete immediately", cn.corename);
+			goto close_fail;
+		}
+
+		cprm.limit = RLIM_INFINITY;
+		cprm.file = no_free_ptr(file);
+#else
+		coredump_report_failure("Core dump socket support %s disabled", cn.corename);
+		goto close_fail;
+#endif
+		break;
+	}
 	default:
 		WARN_ON_ONCE(true);
 		goto close_fail;
@@ -838,8 +953,32 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 		file_end_write(cprm.file);
 		free_vma_snapshot(&cprm);
 	}
-	if ((cn.core_type == COREDUMP_PIPE) && core_pipe_limit)
-		wait_for_dump_helpers(cprm.file);
+
+	/*
+	 * When core_pipe_limit is set we wait for the coredump server
+	 * or usermodehelper to finish before exiting so it can e.g.,
+	 * inspect /proc/<pid>.
+	 */
+	if (core_pipe_limit) {
+		switch (cn.core_type) {
+		case COREDUMP_PIPE:
+			wait_for_dump_helpers(cprm.file);
+			break;
+		case COREDUMP_SOCK: {
+			/*
+			 * We use a simple read to wait for the coredump
+			 * processing to finish. Either the socket is
+			 * closed or we get sent unexpected data. In
+			 * both cases, we're done.
+			 */
+			__kernel_read(cprm.file, &(char){}, 1, NULL);
+			break;
+		}
+		default:
+			break;
+		}
+	}
+
 close_fail:
 	if (cprm.file)
 		filp_close(cprm.file, NULL);
@@ -1069,7 +1208,7 @@ EXPORT_SYMBOL(dump_align);
 void validate_coredump_safety(void)
 {
 	if (suid_dumpable == SUID_DUMP_ROOT &&
-	    core_pattern[0] != '/' && core_pattern[0] != '|') {
+	    core_pattern[0] != '/' && core_pattern[0] != '|' && core_pattern[0] != '@') {
 
 		coredump_report_failure("Unsafe core_pattern used with fs.suid_dumpable=2: "
 			"pipe handler or fully qualified core dump path required. "
diff --git a/include/linux/coredump.h b/include/linux/coredump.h
index 76e41805b92d..aa6820df916d 100644
--- a/include/linux/coredump.h
+++ b/include/linux/coredump.h
@@ -7,6 +7,8 @@
 #include <linux/fs.h>
 #include <asm/siginfo.h>
 
+struct sockaddr_un;
+
 #ifdef CONFIG_COREDUMP
 struct core_vma_metadata {
 	unsigned long start, end;
@@ -76,4 +78,16 @@ extern void validate_coredump_safety(void);
 static inline void validate_coredump_safety(void) {}
 #endif
 
+#if defined(CONFIG_COREDUMP) && defined(CONFIG_UNIX)
+int unix_may_bind_coredump_addr(struct net *net, struct sockaddr_un *sunname,
+				int len);
+#elif defined(CONFIG_UNIX)
+static inline int unix_may_bind_coredump_addr(struct net *net,
+					      struct sockaddr_un *sunname,
+					      int len)
+{
+	return 0;
+}
+#endif
+
 #endif /* _LINUX_COREDUMP_H */
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 472f8aa9ea15..05e5a4737333 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -101,6 +101,7 @@
 #include <linux/string.h>
 #include <linux/uaccess.h>
 #include <linux/pidfs.h>
+#include <linux/coredump.h>
 #include <net/af_unix.h>
 #include <net/net_namespace.h>
 #include <net/scm.h>
@@ -436,6 +437,18 @@ static struct sock *__unix_find_socket_byname(struct net *net,
 	return NULL;
 }
 
+static int unix_may_bind_name(struct net *net, struct sockaddr_un *sunname,
+			      int len, unsigned int hash)
+{
+	struct sock *s;
+
+	s = __unix_find_socket_byname(net, sunname, len, hash);
+	if (s)
+		return -EADDRINUSE;
+
+	return unix_may_bind_coredump_addr(net, sunname, len);
+}
+
 static inline struct sock *unix_find_socket_byname(struct net *net,
 						   struct sockaddr_un *sunname,
 						   int len, unsigned int hash)
@@ -1258,10 +1271,10 @@ static int unix_autobind(struct sock *sk)
 	new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
 	unix_table_double_lock(net, old_hash, new_hash);
 
-	if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
+	if (unix_may_bind_name(net, addr->name, addr->len, new_hash)) {
 		unix_table_double_unlock(net, old_hash, new_hash);
 
-		/* __unix_find_socket_byname() may take long time if many names
+		/* unix_may_bind_name() may take long time if many names
 		 * are already in use.
 		 */
 		cond_resched();
@@ -1379,7 +1392,8 @@ static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
 	new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
 	unix_table_double_lock(net, old_hash, new_hash);
 
-	if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
+	err = unix_may_bind_name(net, addr->name, addr->len, new_hash);
+	if (err)
 		goto out_spin;
 
 	__unix_set_addr_hash(net, sk, addr, new_hash);
@@ -1389,7 +1403,6 @@ static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
 
 out_spin:
 	unix_table_double_unlock(net, old_hash, new_hash);
-	err = -EADDRINUSE;
 out_mutex:
 	mutex_unlock(&u->bindlock);
 out:

-- 
2.47.2

[PATCH v5 5/9] pidfs, coredump: add PIDFD_INFO_COREDUMP

[Christian Brauner]

Extend the PIDFD_INFO_COREDUMP ioctl() with the new PIDFD_INFO_COREDUMP
mask flag. This adds the fields @coredump_mask and @coredump_cookie to
struct pidfd_info.

When a task coredumps the kernel will provide the following information
to userspace in @coredump_mask:

* PIDFD_COREDUMPED is raised if the task did actually coredump.
* PIDFD_COREDUMP_SKIP is raised if the task skipped coredumping (e.g.,
  undumpable).
* PIDFD_COREDUMP_USER is raised if this is a regular coredump and
  doesn't need special care by the coredump server.
* PIDFD_COREDUMP_ROOT is raised if the generated coredump should be
  treated as sensitive and the coredump server should restrict to the
  generated coredump to sufficiently privileged users.

If userspace uses the coredump socket to process coredumps it needs to
be able to discern connection from the kernel from connects from
userspace (e.g., Python generating it's own coredumps and forwarding
them to systemd). The @coredump_cookie extension uses the SO_COOKIE of
the new connection. This allows userspace to validate that the
connection has been made from the kernel by a crashing task:

   fd_coredump = accept4(fd_socket, NULL, NULL, SOCK_CLOEXEC);
   getsockopt(fd_coredump, SOL_SOCKET, SO_PEERPIDFD, &fd_peer_pidfd, &fd_peer_pidfd_len);

   struct pidfd_info info = {
           info.mask = PIDFD_INFO_EXIT | PIDFD_INFO_COREDUMP,
   };

   ioctl(pidfd, PIDFD_GET_INFO, &info);
   /* Refuse connections that aren't from a crashing task. */
   if (!(info.mask & PIDFD_INFO_COREDUMP) || !(info.coredump_mask & PIDFD_COREDUMPED) )
           close(fd_coredump);

   /*
    * Make sure that the coredump cookie matches the connection cookie.
    * If they don't it's not the coredump connection from the kernel.
    * We'll get another connection request in a bit.
    */
   getsocketop(fd_coredump, SOL_SOCKET, SO_COOKIE, &peer_cookie, &peer_cookie_len);
   if (!info.coredump_cookie || (info.coredump_cookie != peer_cookie))
           close(fd_coredump);

The kernel guarantees that by the time the connection is made the all
PIDFD_INFO_COREDUMP info is available.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c              | 37 ++++++++++++++++++++++-
 fs/pidfs.c                 | 75 ++++++++++++++++++++++++++++++++++++++++++++++
 include/linux/net.h        |  1 +
 include/linux/pidfs.h      | 10 +++++++
 include/uapi/linux/pidfd.h | 22 ++++++++++++++
 net/unix/af_unix.c         |  7 +++++
 6 files changed, 151 insertions(+), 1 deletion(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index d3599d671c51..ff42688ec9ac 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -46,7 +46,9 @@
 #include <linux/pidfs.h>
 #include <linux/net.h>
 #include <linux/socket.h>
+#include <net/af_unix.h>
 #include <net/net_namespace.h>
+#include <net/sock.h>
 #include <uapi/linux/pidfd.h>
 #include <uapi/linux/un.h>
 
@@ -599,6 +601,8 @@ static int umh_coredump_setup(struct subprocess_info *info, struct cred *new)
 		if (IS_ERR(pidfs_file))
 			return PTR_ERR(pidfs_file);
 
+		pidfs_coredump(cp);
+
 		/*
 		 * Usermode helpers are childen of either
 		 * system_unbound_wq or of kthreadd. So we know that
@@ -899,15 +903,46 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 			goto close_fail;
 		}
 
+		/*
+		 * Set the thread-group leader pid which is used for the
+		 * peer credentials during connect() below. Then
+		 * immediately register it in pidfs...
+		 */
+		cprm.pid = task_tgid(current);
+		retval = pidfs_register_pid(cprm.pid);
+		if (retval) {
+			sock_release(socket);
+			goto close_fail;
+		}
+
+		/*
+		 * ... and set the coredump information so userspace
+		 * has it available after connect()...
+		 */
+		pidfs_coredump(&cprm);
+
+		/*
+		 * ... On connect() the peer credentials are recorded
+		 * and @cprm.pid registered in pidfs...
+		 */
 		retval = kernel_connect(socket,
 					(struct sockaddr *)(&coredump_unix_socket),
-					COREDUMP_UNIX_SOCKET_ADDR_SIZE, O_NONBLOCK);
+					COREDUMP_UNIX_SOCKET_ADDR_SIZE, O_NONBLOCK |
+					SOCK_COREDUMP);
+
+		/* ... So we can safely put our pidfs reference now... */
+		pidfs_put_pid(cprm.pid);
+
 		if (retval) {
 			if (retval == -EAGAIN)
 				coredump_report_failure("Skipping as coredump socket connection %s couldn't complete immediately", cn.corename);
 			goto close_fail;
 		}
 
+		/* ... and validate that @sk_peer_pid matches @cprm.pid. */
+		if (WARN_ON_ONCE(unix_peer(socket->sk)->sk_peer_pid != cprm.pid))
+			goto close_fail;
+
 		cprm.limit = RLIM_INFINITY;
 		cprm.file = no_free_ptr(file);
 #else
diff --git a/fs/pidfs.c b/fs/pidfs.c
index 3b39e471840b..848a12c8f9cf 100644
--- a/fs/pidfs.c
+++ b/fs/pidfs.c
@@ -20,6 +20,7 @@
 #include <linux/time_namespace.h>
 #include <linux/utsname.h>
 #include <net/net_namespace.h>
+#include <linux/coredump.h>
 
 #include "internal.h"
 #include "mount.h"
@@ -33,6 +34,8 @@ static struct kmem_cache *pidfs_cachep __ro_after_init;
 struct pidfs_exit_info {
 	__u64 cgroupid;
 	__s32 exit_code;
+	__u32 coredump_mask;
+	__u64 coredump_cookie;
 };
 
 struct pidfs_inode {
@@ -240,6 +243,22 @@ static inline bool pid_in_current_pidns(const struct pid *pid)
 	return false;
 }
 
+static __u32 pidfs_coredump_mask(unsigned long mm_flags)
+{
+	switch (__get_dumpable(mm_flags)) {
+	case SUID_DUMP_USER:
+		return PIDFD_COREDUMP_USER;
+	case SUID_DUMP_ROOT:
+		return PIDFD_COREDUMP_ROOT;
+	case SUID_DUMP_DISABLE:
+		return PIDFD_COREDUMP_SKIP;
+	default:
+		WARN_ON_ONCE(true);
+	}
+
+	return 0;
+}
+
 static long pidfd_info(struct file *file, unsigned int cmd, unsigned long arg)
 {
 	struct pidfd_info __user *uinfo = (struct pidfd_info __user *)arg;
@@ -280,6 +299,12 @@ static long pidfd_info(struct file *file, unsigned int cmd, unsigned long arg)
 		}
 	}
 
+	if (mask & PIDFD_INFO_COREDUMP) {
+		kinfo.mask |= PIDFD_INFO_COREDUMP;
+		kinfo.coredump_cookie = READ_ONCE(pidfs_i(inode)->__pei.coredump_cookie);
+		kinfo.coredump_mask = READ_ONCE(pidfs_i(inode)->__pei.coredump_mask);
+	}
+
 	task = get_pid_task(pid, PIDTYPE_PID);
 	if (!task) {
 		/*
@@ -296,6 +321,15 @@ static long pidfd_info(struct file *file, unsigned int cmd, unsigned long arg)
 	if (!c)
 		return -ESRCH;
 
+	if (!(kinfo.mask & PIDFD_INFO_COREDUMP)) {
+		task_lock(task);
+		if (task->mm) {
+			kinfo.coredump_cookie = READ_ONCE(pidfs_i(inode)->__pei.coredump_cookie);
+			kinfo.coredump_mask = pidfs_coredump_mask(task->mm->flags);
+		}
+		task_unlock(task);
+	}
+
 	/* Unconditionally return identifiers and credentials, the rest only on request */
 
 	user_ns = current_user_ns();
@@ -559,6 +593,47 @@ void pidfs_exit(struct task_struct *tsk)
 	}
 }
 
+#if defined(CONFIG_COREDUMP) && defined(CONFIG_UNIX)
+void pidfs_coredump_cookie(struct pid *pid, u64 coredump_cookie)
+{
+	struct pidfs_exit_info *exit_info;
+	struct dentry *dentry = pid->stashed;
+	struct inode *inode;
+
+	if (WARN_ON_ONCE(!dentry))
+		return;
+
+	inode = d_inode(dentry);
+	exit_info = &pidfs_i(inode)->__pei;
+	smp_store_release(&exit_info->coredump_cookie, coredump_cookie);
+}
+#endif
+
+#ifdef CONFIG_COREDUMP
+void pidfs_coredump(const struct coredump_params *cprm)
+{
+	struct pid *pid = cprm->pid;
+	struct pidfs_exit_info *exit_info;
+	struct dentry *dentry;
+	struct inode *inode;
+	__u32 coredump_mask = 0;
+
+	dentry = pid->stashed;
+	if (WARN_ON_ONCE(!dentry))
+		return;
+
+	inode = d_inode(dentry);
+	exit_info = &pidfs_i(inode)->__pei;
+	/* Note how we were coredumped. */
+	coredump_mask = pidfs_coredump_mask(cprm->mm_flags);
+	/* Note that we actually did coredump. */
+	coredump_mask |= PIDFD_COREDUMPED;
+	/* If coredumping is set to skip we should never end up here. */
+	VFS_WARN_ON_ONCE(coredump_mask & PIDFD_COREDUMP_SKIP);
+	smp_store_release(&exit_info->coredump_mask, coredump_mask);
+}
+#endif
+
 static struct vfsmount *pidfs_mnt __ro_after_init;
 
 /*
diff --git a/include/linux/net.h b/include/linux/net.h
index 0ff950eecc6b..139c85d0f2ea 100644
--- a/include/linux/net.h
+++ b/include/linux/net.h
@@ -81,6 +81,7 @@ enum sock_type {
 #ifndef SOCK_NONBLOCK
 #define SOCK_NONBLOCK	O_NONBLOCK
 #endif
+#define SOCK_COREDUMP	O_NOCTTY
 
 #endif /* ARCH_HAS_SOCKET_TYPES */
 
diff --git a/include/linux/pidfs.h b/include/linux/pidfs.h
index 2676890c4d0d..497997bc5e34 100644
--- a/include/linux/pidfs.h
+++ b/include/linux/pidfs.h
@@ -2,11 +2,21 @@
 #ifndef _LINUX_PID_FS_H
 #define _LINUX_PID_FS_H
 
+struct coredump_params;
+
 struct file *pidfs_alloc_file(struct pid *pid, unsigned int flags);
 void __init pidfs_init(void);
 void pidfs_add_pid(struct pid *pid);
 void pidfs_remove_pid(struct pid *pid);
 void pidfs_exit(struct task_struct *tsk);
+#ifdef CONFIG_COREDUMP
+void pidfs_coredump(const struct coredump_params *cprm);
+#endif
+#if defined(CONFIG_COREDUMP) && defined(CONFIG_UNIX)
+void pidfs_coredump_cookie(struct pid *pid, u64 coredump_cookie);
+#elif defined(CONFIG_UNIX)
+static inline void pidfs_coredump_cookie(struct pid *pid, u64 coredump_cookie) { }
+#endif
 extern const struct dentry_operations pidfs_dentry_operations;
 int pidfs_register_pid(struct pid *pid);
 void pidfs_get_pid(struct pid *pid);
diff --git a/include/uapi/linux/pidfd.h b/include/uapi/linux/pidfd.h
index 8c1511edd0e9..69267c5ae6d0 100644
--- a/include/uapi/linux/pidfd.h
+++ b/include/uapi/linux/pidfd.h
@@ -25,9 +25,28 @@
 #define PIDFD_INFO_CREDS		(1UL << 1) /* Always returned, even if not requested */
 #define PIDFD_INFO_CGROUPID		(1UL << 2) /* Always returned if available, even if not requested */
 #define PIDFD_INFO_EXIT			(1UL << 3) /* Only returned if requested. */
+#define PIDFD_INFO_COREDUMP		(1UL << 4) /* Only returned if requested. */
 
 #define PIDFD_INFO_SIZE_VER0		64 /* sizeof first published struct */
 
+/*
+ * Values for @coredump_mask in pidfd_info.
+ * Only valid if PIDFD_INFO_COREDUMP is set in @mask.
+ *
+ * Note, the @PIDFD_COREDUMP_ROOT flag indicates that the generated
+ * coredump should be treated as sensitive and access should only be
+ * granted to privileged users.
+ *
+ * If the coredump AF_UNIX socket is used for processing coredumps
+ * @coredump_cookie will be set to the socket SO_COOKIE of the receivers
+ * client socket. This allows the coredump handler to detect whether an
+ * incoming coredump connection was initiated from the crashing task.
+ */
+#define PIDFD_COREDUMPED	(1U << 0) /* Did crash and... */
+#define PIDFD_COREDUMP_SKIP	(1U << 1) /* coredumping generation was skipped. */
+#define PIDFD_COREDUMP_USER	(1U << 2) /* coredump was done as the user. */
+#define PIDFD_COREDUMP_ROOT	(1U << 3) /* coredump was done as root. */
+
 /*
  * The concept of process and threads in userland and the kernel is a confusing
  * one - within the kernel every thread is a 'task' with its own individual PID,
@@ -92,6 +111,9 @@ struct pidfd_info {
 	__u32 fsuid;
 	__u32 fsgid;
 	__s32 exit_code;
+	__u32 coredump_mask;
+	__u32 __spare1;
+	__u64 coredump_cookie;
 };
 
 #define PIDFS_IOCTL_MAGIC 0xFF
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 05e5a4737333..294667b572ee 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -101,6 +101,7 @@
 #include <linux/string.h>
 #include <linux/uaccess.h>
 #include <linux/pidfs.h>
+#include <linux/sock_diag.h>
 #include <linux/coredump.h>
 #include <net/af_unix.h>
 #include <net/net_namespace.h>
@@ -753,6 +754,7 @@ static void unix_release_sock(struct sock *sk, int embrion)
 
 struct unix_peercred {
 	struct pid *peer_pid;
+	u64 cookie;
 	const struct cred *peer_cred;
 };
 
@@ -788,6 +790,8 @@ static void drop_peercred(struct unix_peercred *peercred)
 static inline void init_peercred(struct sock *sk,
 				 const struct unix_peercred *peercred)
 {
+	if (peercred->cookie)
+		pidfs_coredump_cookie(peercred->peer_pid, peercred->cookie);
 	sk->sk_peer_pid = peercred->peer_pid;
 	sk->sk_peer_cred = peercred->peer_cred;
 }
@@ -1699,6 +1703,9 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 	unix_peer(newsk)	= sk;
 	newsk->sk_state		= TCP_ESTABLISHED;
 	newsk->sk_type		= sk->sk_type;
+	/* Prepare a new socket cookie for the receiver. */
+	if (flags & SOCK_COREDUMP)
+		peercred.cookie = sock_gen_cookie(newsk);
 	init_peercred(newsk, &peercred);
 	newu = unix_sk(newsk);
 	newu->listener = other;

-- 
2.47.2

[PATCH v5 6/9] coredump: show supported coredump modes

[Christian Brauner]

Allow userspace to discover what coredump modes are supported.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/fs/coredump.c b/fs/coredump.c
index ff42688ec9ac..fa50a681b225 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -1263,6 +1263,12 @@ static int proc_dostring_coredump(const struct ctl_table *table, int write,
 
 static const unsigned int core_file_note_size_min = CORE_FILE_NOTE_SIZE_DEFAULT;
 static const unsigned int core_file_note_size_max = CORE_FILE_NOTE_SIZE_MAX;
+static char core_modes[] = {
+	"file\npipe"
+#ifdef CONFIG_UNIX
+	"\nlinuxafsk/coredump.socket"
+#endif
+};
 
 static const struct ctl_table coredump_sysctls[] = {
 	{
@@ -1306,6 +1312,13 @@ static const struct ctl_table coredump_sysctls[] = {
 		.extra1		= SYSCTL_ZERO,
 		.extra2		= SYSCTL_ONE,
 	},
+	{
+		.procname	= "core_modes",
+		.data		= core_modes,
+		.maxlen		= sizeof(core_modes) - 1,
+		.mode		= 0444,
+		.proc_handler	= proc_dostring,
+	},
 };
 
 static int __init init_fs_coredump_sysctls(void)

-- 
2.47.2

[PATCH v5 7/9] coredump: validate socket name as it is written

[Christian Brauner]

In contrast to other parameters written into
/proc/sys/kernel/core_pattern that never fail we can validate enabling
the new AF_UNIX support. This is obviously racy as hell but it's always
been that way.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index fa50a681b225..19e776769b21 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -1254,10 +1254,21 @@ void validate_coredump_safety(void)
 static int proc_dostring_coredump(const struct ctl_table *table, int write,
 		  void *buffer, size_t *lenp, loff_t *ppos)
 {
-	int error = proc_dostring(table, write, buffer, lenp, ppos);
+	int error;
+	ssize_t retval;
+	char old_core_pattern[CORENAME_MAX_SIZE];
 
-	if (!error)
-		validate_coredump_safety();
+	retval = strscpy(old_core_pattern, core_pattern, CORENAME_MAX_SIZE);
+
+	error = proc_dostring(table, write, buffer, lenp, ppos);
+	if (error)
+		return error;
+	if (core_pattern[0] == '@' && strcmp(core_pattern, "@linuxafsk/coredump.socket")) {
+		strscpy(core_pattern, old_core_pattern, retval + 1);
+		return -EINVAL;
+	}
+
+	validate_coredump_safety();
 	return error;
 }
 

-- 
2.47.2

[PATCH v5 8/9] selftests/pidfd: add PIDFD_INFO_COREDUMP infrastructure

[Christian Brauner]

Add PIDFD_INFO_COREDUMP infrastructure so we can use it in tests.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 tools/testing/selftests/pidfd/pidfd.h | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/tools/testing/selftests/pidfd/pidfd.h b/tools/testing/selftests/pidfd/pidfd.h
index 55bcf81a2b9a..887c74007086 100644
--- a/tools/testing/selftests/pidfd/pidfd.h
+++ b/tools/testing/selftests/pidfd/pidfd.h
@@ -131,6 +131,26 @@
 #define PIDFD_INFO_EXIT			(1UL << 3) /* Always returned if available, even if not requested */
 #endif
 
+#ifndef PIDFD_INFO_COREDUMP
+#define PIDFD_INFO_COREDUMP	(1UL << 4)
+#endif
+
+#ifndef PIDFD_COREDUMPED
+#define PIDFD_COREDUMPED	(1U << 0) /* Did crash and... */
+#endif
+
+#ifndef PIDFD_COREDUMP_SKIP
+#define PIDFD_COREDUMP_SKIP	(1U << 1) /* coredumping generation was skipped. */
+#endif
+
+#ifndef PIDFD_COREDUMP_USER
+#define PIDFD_COREDUMP_USER	(1U << 2) /* coredump was done as the user. */
+#endif
+
+#ifndef PIDFD_COREDUMP_ROOT
+#define PIDFD_COREDUMP_ROOT	(1U << 3) /* coredump was done as root. */
+#endif
+
 #ifndef PIDFD_THREAD
 #define PIDFD_THREAD O_EXCL
 #endif
@@ -150,6 +170,9 @@ struct pidfd_info {
 	__u32 fsuid;
 	__u32 fsgid;
 	__s32 exit_code;
+	__u32 coredump_mask;
+	__u32 __spare1;
+	__u64 coredump_cookie;
 };
 
 /*

-- 
2.47.2

[PATCH v5 9/9] selftests/coredump: add tests for AF_UNIX coredumps

[Christian Brauner]

Add a simple test for generating coredumps via AF_UNIX sockets.

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 tools/testing/selftests/coredump/stackdump_test.c | 231 +++++++++++++++++++++-
 1 file changed, 230 insertions(+), 1 deletion(-)

diff --git a/tools/testing/selftests/coredump/stackdump_test.c b/tools/testing/selftests/coredump/stackdump_test.c
index fe3c728cd6be..8423499808ff 100644
--- a/tools/testing/selftests/coredump/stackdump_test.c
+++ b/tools/testing/selftests/coredump/stackdump_test.c
@@ -5,10 +5,15 @@
 #include <linux/limits.h>
 #include <pthread.h>
 #include <string.h>
+#include <sys/mount.h>
 #include <sys/resource.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <sys/un.h>
 #include <unistd.h>
 
 #include "../kselftest_harness.h"
+#include "../pidfd/pidfd.h"
 
 #define STACKDUMP_FILE "stack_values"
 #define STACKDUMP_SCRIPT "stackdump"
@@ -35,6 +40,7 @@ static void crashing_child(void)
 FIXTURE(coredump)
 {
 	char original_core_pattern[256];
+	pid_t pid_coredump_server;
 };
 
 FIXTURE_SETUP(coredump)
@@ -44,6 +50,7 @@ FIXTURE_SETUP(coredump)
 	char *dir;
 	int ret;
 
+	self->pid_coredump_server = -ESRCH;
 	file = fopen("/proc/sys/kernel/core_pattern", "r");
 	ASSERT_NE(NULL, file);
 
@@ -61,10 +68,15 @@ FIXTURE_TEARDOWN(coredump)
 {
 	const char *reason;
 	FILE *file;
-	int ret;
+	int ret, status;
 
 	unlink(STACKDUMP_FILE);
 
+	if (self->pid_coredump_server > 0) {
+		kill(self->pid_coredump_server, SIGTERM);
+		waitpid(self->pid_coredump_server, &status, 0);
+	}
+
 	file = fopen("/proc/sys/kernel/core_pattern", "w");
 	if (!file) {
 		reason = "Unable to open core_pattern";
@@ -154,4 +166,221 @@ TEST_F_TIMEOUT(coredump, stackdump, 120)
 	fclose(file);
 }
 
+TEST_F(coredump, socket)
+{
+	int fd, pidfd, ret, status;
+	FILE *file;
+	pid_t pid, pid_coredump_server;
+	struct stat st;
+	char core_file[PATH_MAX];
+	struct pidfd_info info = {};
+	int ipc_sockets[2];
+	char c;
+
+	ASSERT_EQ(unshare(CLONE_NEWNS), 0);
+	ASSERT_EQ(mount(NULL, "/", NULL, MS_PRIVATE | MS_REC, NULL), 0);
+	ASSERT_EQ(mount(NULL, "/tmp", "tmpfs", 0, NULL), 0);
+
+	file = fopen("/proc/sys/kernel/core_pattern", "w");
+	ASSERT_NE(NULL, file);
+
+	ret = fprintf(file, "@linuxafsk/coredump.socket");
+	ASSERT_EQ(ret, strlen("@linuxafsk/coredump.socket"));
+	ASSERT_EQ(fclose(file), 0);
+
+	ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets);
+	ASSERT_EQ(ret, 0);
+
+	pid_coredump_server = fork();
+	ASSERT_GE(pid_coredump_server, 0);
+	if (pid_coredump_server == 0) {
+		int fd_socket, fd_coredump, fd_peer_pidfd, fd_core_file;
+		__u64 peer_cookie;
+		socklen_t fd_peer_pidfd_len, peer_cookie_len;
+		static const struct sockaddr_un coredump_sk = {
+			.sun_family = AF_UNIX,
+			.sun_path = "\0linuxafsk/coredump.socket",
+		};
+		static const size_t coredump_sk_len =
+			offsetof(struct sockaddr_un, sun_path) +
+			sizeof("linuxafsk/coredump.socket"); /* +1 for leading NUL */
+
+		close(ipc_sockets[0]);
+
+		fd_socket = socket(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0);
+		if (fd_socket < 0)
+			_exit(EXIT_FAILURE);
+
+		ret = bind(fd_socket, (const struct sockaddr *)&coredump_sk, coredump_sk_len);
+		if (ret < 0) {
+			fprintf(stderr, "Failed to bind coredump socket\n");
+			close(fd_socket);
+			close(ipc_sockets[1]);
+			_exit(EXIT_FAILURE);
+		}
+
+		ret = listen(fd_socket, 1);
+		if (ret < 0) {
+			fprintf(stderr, "Failed to listen on coredump socket\n");
+			close(fd_socket);
+			close(ipc_sockets[1]);
+			_exit(EXIT_FAILURE);
+		}
+
+		if (write_nointr(ipc_sockets[1], "1", 1) < 0) {
+			close(fd_socket);
+			close(ipc_sockets[1]);
+			_exit(EXIT_FAILURE);
+		}
+
+		close(ipc_sockets[1]);
+
+		fd_coredump = accept4(fd_socket, NULL, NULL, SOCK_CLOEXEC);
+		if (fd_coredump < 0) {
+			fprintf(stderr, "Failed to accept coredump socket connection\n");
+			close(fd_socket);
+			_exit(EXIT_FAILURE);
+		}
+
+		peer_cookie_len = sizeof(peer_cookie);
+		ret = getsockopt(fd_coredump, SOL_SOCKET, SO_COOKIE,
+				 &peer_cookie, &peer_cookie_len);
+		if (ret < 0) {
+			fprintf(stderr, "%m - Failed to retrieve cookie for coredump socket connection\n");
+			close(fd_coredump);
+			close(fd_socket);
+			_exit(EXIT_FAILURE);
+		}
+
+		fd_peer_pidfd_len = sizeof(fd_peer_pidfd);
+		ret = getsockopt(fd_coredump, SOL_SOCKET, SO_PEERPIDFD,
+				 &fd_peer_pidfd, &fd_peer_pidfd_len);
+		if (ret < 0) {
+			fprintf(stderr, "%m - Failed to retrieve peer pidfd for coredump socket connection\n");
+			close(fd_coredump);
+			close(fd_socket);
+			_exit(EXIT_FAILURE);
+		}
+
+		memset(&info, 0, sizeof(info));
+		info.mask = PIDFD_INFO_EXIT | PIDFD_INFO_COREDUMP;
+		ret = ioctl(fd_peer_pidfd, PIDFD_GET_INFO, &info);
+		if (ret < 0) {
+			fprintf(stderr, "Failed to retrieve pidfd info from peer pidfd for coredump socket connection\n");
+			close(fd_coredump);
+			close(fd_socket);
+			close(fd_peer_pidfd);
+			_exit(EXIT_FAILURE);
+		}
+
+		if (!(info.mask & PIDFD_INFO_COREDUMP)) {
+			fprintf(stderr, "Missing coredump information from coredumping task\n");
+			close(fd_coredump);
+			close(fd_socket);
+			close(fd_peer_pidfd);
+			_exit(EXIT_FAILURE);
+		}
+
+		if (!(info.coredump_mask & PIDFD_COREDUMPED)) {
+			fprintf(stderr, "Received connection from non-coredumping task\n");
+			close(fd_coredump);
+			close(fd_socket);
+			close(fd_peer_pidfd);
+			_exit(EXIT_FAILURE);
+		}
+
+		if (!info.coredump_cookie) {
+			fprintf(stderr, "Missing coredump cookie\n");
+			close(fd_coredump);
+			close(fd_socket);
+			close(fd_peer_pidfd);
+			_exit(EXIT_FAILURE);
+		}
+
+		if (info.coredump_cookie != peer_cookie) {
+			fprintf(stderr, "Mismatching coredump cookies\n");
+			close(fd_coredump);
+			close(fd_socket);
+			close(fd_peer_pidfd);
+			_exit(EXIT_FAILURE);
+		}
+
+		fd_core_file = creat("/tmp/coredump.file", 0644);
+		if (fd_core_file < 0) {
+			fprintf(stderr, "Failed to create coredump file\n");
+			close(fd_coredump);
+			close(fd_socket);
+			close(fd_peer_pidfd);
+			_exit(EXIT_FAILURE);
+		}
+
+		for (;;) {
+			char buffer[4096];
+			ssize_t bytes_read, bytes_write;
+
+			bytes_read = read(fd_coredump, buffer, sizeof(buffer));
+			if (bytes_read < 0) {
+				close(fd_coredump);
+				close(fd_socket);
+				close(fd_peer_pidfd);
+				close(fd_core_file);
+				_exit(EXIT_FAILURE);
+			}
+
+			if (bytes_read == 0)
+				break;
+
+			bytes_write = write(fd_core_file, buffer, bytes_read);
+			if (bytes_read != bytes_write) {
+				close(fd_coredump);
+				close(fd_socket);
+				close(fd_peer_pidfd);
+				close(fd_core_file);
+				_exit(EXIT_FAILURE);
+			}
+		}
+
+		close(fd_coredump);
+		close(fd_socket);
+		close(fd_peer_pidfd);
+		close(fd_core_file);
+		_exit(EXIT_SUCCESS);
+	}
+	self->pid_coredump_server = pid_coredump_server;
+
+	EXPECT_EQ(close(ipc_sockets[1]), 0);
+	ASSERT_EQ(read_nointr(ipc_sockets[0], &c, 1), 1);
+	EXPECT_EQ(close(ipc_sockets[0]), 0);
+
+	pid = fork();
+	ASSERT_GE(pid, 0);
+	if (pid == 0)
+		crashing_child();
+
+	pidfd = sys_pidfd_open(pid, 0);
+	ASSERT_GE(pidfd, 0);
+
+	waitpid(pid, &status, 0);
+	ASSERT_TRUE(WIFSIGNALED(status));
+	ASSERT_TRUE(WCOREDUMP(status));
+
+	info.mask = PIDFD_INFO_EXIT | PIDFD_INFO_COREDUMP;
+	ASSERT_EQ(ioctl(pidfd, PIDFD_GET_INFO, &info), 0);
+	ASSERT_GT((info.mask & PIDFD_INFO_COREDUMP), 0);
+	ASSERT_GT((info.coredump_mask & PIDFD_COREDUMPED), 0);
+
+	waitpid(pid_coredump_server, &status, 0);
+	self->pid_coredump_server = -ESRCH;
+	ASSERT_TRUE(WIFEXITED(status));
+	ASSERT_EQ(WEXITSTATUS(status), 0);
+
+	ASSERT_EQ(stat("/tmp/coredump.file", &st), 0);
+	ASSERT_GT(st.st_size, 0);
+	/*
+	 * We should somehow validate the produced core file.
+	 * For now just allow for visual inspection
+	 */
+	system("file /tmp/coredump.file");
+}
+
 TEST_HARNESS_MAIN

-- 
2.47.2

Re: [PATCH v5 4/9] coredump: add coredump socket

[Christian Brauner]

[-- Attachment #1: Type: text/plain, Size: 1448 bytes --]

> Userspace can set /proc/sys/kernel/core_pattern to:
> 
>         @linuxafsk/coredump_socket

I have one other proposal that:

- avoids reserving a specific address
- doesn't require bpf or lsm to be safe
- allows for safe restart and crashes of the coredump sever

To set up a coredump socket the coredump server must allocate a socket
cookie for the listening socket via SO_COOKIE. The socket cookie must be
used as the prefix in the abstract address for the coredump socket. It
can be followed by a \0 byte and then followed by whatever the coredump
server wants. For example:

12345678\0coredump.socket

When a task crashes and generates a coredump it will find the provided
address but also compare the prefixed SO_COOKIE value with the socket
cookie of the socket listening at that address. If they don't match it
will refuse to connect.

So even if the coredump server restarts or crashes and unprivileged
userspace recycles the socket address for an attack the crashing process
will detect this as the new listening socket will have gotten either a
new or no SO_COOKIE and the crashing process will not connect.

The coredump server just sets /proc/sys/kernel/core_pattern to:

        @SO_COOKIE/whatever

The "@" at the beginning indicates to the kernel that the abstract
AF_UNIX coredump socket will be used to process coredumps and the
indicating the end of the SO_COOKIE and the rest of the name.

Appended what that would look like.

[-- Attachment #2: 0001-coredump-add-coredump-socket.patch --]
[-- Type: text/x-diff, Size: 15282 bytes --]

From a99749a48d2aafddeca5fd1fab64b907137c1a68 Mon Sep 17 00:00:00 2001
From: Christian Brauner <brauner@kernel.org>
Date: Mon, 5 May 2025 11:33:44 +0200
Subject: [PATCH] coredump: add coredump socket

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:

        @SO_COOKIE/whatever

The "@" at the beginning indicates to the kernel that the abstract
AF_UNIX coredump socket will be used to process coredumps.

When the coredump server sets up a coredump socket it must allocate a
socket cookie for it and use it as the prefix in the abstract address.
It may be followed by a zero byte and whatever other name the server may
want.

The crashing process uses the socket cookie to verify the socket
connection. Even if the coredump server restarts or crashes and
unprivileged userspace recycles the socket address the kernel will
detect that the address has been recycled as a new socket cookie will
have been allocated for it.

The coredump socket is located in the initial network namespace. To bind
the coredump socket userspace must hold CAP_SYS_ADMIN in the initial
user namespace. Listening and reading can happen from whatever
unprivileged context is necessary to safely process coredumps.

When a task coredumps it opens a client socket in the initial network
namespace and connects to the coredump socket.

- The coredump server uses SO_PEERPIDFD to get a stable handle on the
  connected crashing task. The retrieved pidfd will provide a stable
  reference even if the crashing task gets SIGKILLed while generating
  the coredump.

- By setting core_pipe_limit non-zero userspace can guarantee that the
  crashing task cannot be reaped behind it's back and thus process all
  necessary information in /proc/<pid>. The SO_PEERPIDFD can be used to
  detect whether /proc/<pid> still refers to the same process.

  The core_pipe_limit isn't used to rate-limit connections to the
  socket. This can simply be done via AF_UNIX sockets directly.

- The pidfd for the crashing task will grow new information how the task
  coredumps.

- The coredump server should mark itself as non-dumpable.
  To capture coredumps for the coredump server itself a bpf program
  should be run at connect to redirect it to another socket in
  userspace. This can be useful for debugging crashing coredump servers.

- A container coredump server in a separate network namespace can simply
  bind to another well-know address and systemd-coredump fowards
  coredumps to the container.

- Coredumps could in the future also be handled via per-user/session
  coredump servers that run only with that users privileges.

  The coredump server listens on the coredump socket and accepts a
  new coredump connection. It then retrieves SO_PEERPIDFD for the
  client, inspects uid/gid and hands the accepted client to the users
  own coredump handler which runs with the users privileges only
  (It must of coure pay close attention to not forward crashing suid
  binaries.).

The new coredump socket will allow userspace to not have to rely on
usermode helpers for processing coredumps and provides a safer way to
handle them instead of relying on super privileged coredumping helpers
that have and continue to cause significant CVEs.

This will also be significantly more lightweight since no fork()+exec()
for the usermodehelper is required for each crashing process. The
coredump server in userspace can e.g., just keep a worker pool.

This is easy to test:

(a) coredump processing (we're using socat):

    > cat coredump_socket.sh
    #!/bin/bash

    set -x

    sudo bash -c "echo '@@socket_cookie/whatever-you-want' > /proc/sys/kernel/core_pattern"
    sudo socat --statistics abstract-listen:socket_cookie\0whatever-you-want,fork FILE:core_file,create,append,trunc

(b) trigger a coredump:

    user1@localhost:~/data/scripts$ cat crash.c
    #include <stdio.h>
    #include <unistd.h>

    int main(int argc, char *argv[])
    {
            fprintf(stderr, "%u\n", (1 / 0));
            _exit(0);
    }

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/coredump.c            | 133 ++++++++++++++++++++++++++++++++++++---
 include/linux/coredump.h |   2 +
 include/linux/net.h      |   2 +
 net/unix/af_unix.c       |  22 +++++--
 4 files changed, 145 insertions(+), 14 deletions(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index b2eda7b176e4..6692790bd459 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -44,7 +44,11 @@
 #include <linux/sysctl.h>
 #include <linux/elf.h>
 #include <linux/pidfs.h>
+#include <linux/net.h>
+#include <linux/socket.h>
+#include <net/net_namespace.h>
 #include <uapi/linux/pidfd.h>
+#include <uapi/linux/un.h>
 
 #include <linux/uaccess.h>
 #include <asm/mmu_context.h>
@@ -79,6 +83,7 @@ unsigned int core_file_note_size_limit = CORE_FILE_NOTE_SIZE_DEFAULT;
 enum coredump_type_t {
 	COREDUMP_FILE = 1,
 	COREDUMP_PIPE = 2,
+	COREDUMP_SOCK = 3,
 };
 
 struct core_name {
@@ -232,13 +237,16 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 	cn->corename = NULL;
 	if (*pat_ptr == '|')
 		cn->core_type = COREDUMP_PIPE;
+	else if (*pat_ptr == '@')
+		cn->core_type = COREDUMP_SOCK;
 	else
 		cn->core_type = COREDUMP_FILE;
 	if (expand_corename(cn, core_name_size))
 		return -ENOMEM;
 	cn->corename[0] = '\0';
 
-	if (cn->core_type == COREDUMP_PIPE) {
+	switch (cn->core_type) {
+	case COREDUMP_PIPE: {
 		int argvs = sizeof(core_pattern) / 2;
 		(*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
 		if (!(*argv))
@@ -247,6 +255,29 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 		++pat_ptr;
 		if (!(*pat_ptr))
 			return -ENOMEM;
+		break;
+	}
+	case COREDUMP_SOCK: {
+		/* skip the @ */
+		pat_ptr++;
+		err = cn_printf(cn, "%s", pat_ptr);
+		if (err)
+			return err;
+
+		/*
+		 * Currently no need to parse any other options.
+		 * Relevant information can be retrieved from the peer
+		 * pidfd retrievable via SO_PEERPIDFD by the receiver or
+		 * via /proc/<pid>, using the SO_PEERPIDFD to guard
+		 * against pid recycling when opening /proc/<pid>.
+		 */
+		return 0;
+	}
+	case COREDUMP_FILE:
+		break;
+	default:
+		WARN_ON_ONCE(true);
+		return -EINVAL;
 	}
 
 	/* Repeat as long as we have more pattern to process and more output
@@ -393,11 +424,20 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
 	 * If core_pattern does not include a %p (as is the default)
 	 * and core_uses_pid is set, then .%pid will be appended to
 	 * the filename. Do not do this for piped commands. */
-	if (!(cn->core_type == COREDUMP_PIPE) && !pid_in_pattern && core_uses_pid) {
-		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
-		if (err)
-			return err;
+	if (!pid_in_pattern && core_uses_pid) {
+		switch (cn->core_type) {
+		case COREDUMP_FILE:
+			return cn_printf(cn, ".%d", task_tgid_vnr(current));
+		case COREDUMP_PIPE:
+			break;
+		case COREDUMP_SOCK:
+			break;
+		default:
+			WARN_ON_ONCE(true);
+			return -EINVAL;
+		}
 	}
+
 	return 0;
 }
 
@@ -801,6 +841,59 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 		}
 		break;
 	}
+	case COREDUMP_SOCK: {
+#ifdef CONFIG_UNIX
+		struct file *file __free(fput) = NULL;
+		struct sockaddr_un addr = {
+			.sun_family = AF_UNIX,
+		};
+		unsigned int addr_len;
+		struct socket *socket;
+		char *p;
+
+		/*
+		 * It is possible that the userspace process which is
+		 * supposed to handle the coredump and is listening on
+		 * the AF_UNIX socket coredumps. Userspace should just
+		 * mark itself non dumpable.
+		 */
+
+		retval = sock_create_kern(&init_net, AF_UNIX, SOCK_STREAM, 0, &socket);
+		if (retval < 0)
+			goto close_fail;
+
+		file = sock_alloc_file(socket, 0, NULL);
+		if (IS_ERR(file)) {
+			sock_release(socket);
+			goto close_fail;
+		}
+
+		retval = strscpy(addr.sun_path + 1, cn.corename, strlen(cn.corename));
+		if (retval)
+			goto close_fail;
+
+		/* Format is @socket_cookie\0whatever. */
+		p = strchr(addr.sun_path + 1, '/');
+		if (p)
+			*p = '\0';
+		addr_len = offsetof(struct sockaddr_un, sun_path) + retval + 1;
+
+		retval = kernel_connect(socket, (struct sockaddr *)(&addr),
+					addr_len, O_NONBLOCK | SOCK_COREDUMP);
+		if (retval) {
+			if (retval == -EAGAIN)
+				coredump_report_failure("Skipping as coredump socket connection %s couldn't complete immediately", cn.corename);
+			goto close_fail;
+		}
+
+		cprm.limit = RLIM_INFINITY;
+		cprm.file = no_free_ptr(file);
+#else
+		coredump_report_failure("Core dump socket support %s disabled", cn.corename);
+		goto close_fail;
+#endif
+		break;
+	}
 	default:
 		WARN_ON_ONCE(true);
 		goto close_fail;
@@ -838,8 +931,32 @@ void do_coredump(const kernel_siginfo_t *siginfo)
 		file_end_write(cprm.file);
 		free_vma_snapshot(&cprm);
 	}
-	if ((cn.core_type == COREDUMP_PIPE) && core_pipe_limit)
-		wait_for_dump_helpers(cprm.file);
+
+	/*
+	 * When core_pipe_limit is set we wait for the coredump server
+	 * or usermodehelper to finish before exiting so it can e.g.,
+	 * inspect /proc/<pid>.
+	 */
+	if (core_pipe_limit) {
+		switch (cn.core_type) {
+		case COREDUMP_PIPE:
+			wait_for_dump_helpers(cprm.file);
+			break;
+		case COREDUMP_SOCK: {
+			/*
+			 * We use a simple read to wait for the coredump
+			 * processing to finish. Either the socket is
+			 * closed or we get sent unexpected data. In
+			 * both cases, we're done.
+			 */
+			__kernel_read(cprm.file, &(char){}, 1, NULL);
+			break;
+		}
+		default:
+			break;
+		}
+	}
+
 close_fail:
 	if (cprm.file)
 		filp_close(cprm.file, NULL);
@@ -1069,7 +1186,7 @@ EXPORT_SYMBOL(dump_align);
 void validate_coredump_safety(void)
 {
 	if (suid_dumpable == SUID_DUMP_ROOT &&
-	    core_pattern[0] != '/' && core_pattern[0] != '|') {
+	    core_pattern[0] != '/' && core_pattern[0] != '|' && core_pattern[0] != '@') {
 
 		coredump_report_failure("Unsafe core_pattern used with fs.suid_dumpable=2: "
 			"pipe handler or fully qualified core dump path required. "
diff --git a/include/linux/coredump.h b/include/linux/coredump.h
index 76e41805b92d..3e4b52c769e3 100644
--- a/include/linux/coredump.h
+++ b/include/linux/coredump.h
@@ -7,6 +7,8 @@
 #include <linux/fs.h>
 #include <asm/siginfo.h>
 
+struct sockaddr_un;
+
 #ifdef CONFIG_COREDUMP
 struct core_vma_metadata {
 	unsigned long start, end;
diff --git a/include/linux/net.h b/include/linux/net.h
index 0ff950eecc6b..3f467786bdc9 100644
--- a/include/linux/net.h
+++ b/include/linux/net.h
@@ -82,6 +82,8 @@ enum sock_type {
 #define SOCK_NONBLOCK	O_NONBLOCK
 #endif
 
+#define SOCK_COREDUMP O_NOCTTY
+
 #endif /* ARCH_HAS_SOCKET_TYPES */
 
 /**
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 472f8aa9ea15..944248d7c5be 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -101,6 +101,7 @@
 #include <linux/string.h>
 #include <linux/uaccess.h>
 #include <linux/pidfs.h>
+#include <linux/kstrtox.h>
 #include <net/af_unix.h>
 #include <net/net_namespace.h>
 #include <net/scm.h>
@@ -1191,7 +1192,7 @@ static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
 
 static struct sock *unix_find_abstract(struct net *net,
 				       struct sockaddr_un *sunaddr,
-				       int addr_len, int type)
+				       int addr_len, int type, int flags)
 {
 	unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
 	struct dentry *dentry;
@@ -1201,6 +1202,15 @@ static struct sock *unix_find_abstract(struct net *net,
 	if (!sk)
 		return ERR_PTR(-ECONNREFUSED);
 
+	if (flags & SOCK_COREDUMP) {
+		u64 cookie;
+
+		if (kstrtou64(sunaddr->sun_path, 0, &cookie))
+			return ERR_PTR(-ECONNREFUSED);
+		if (cookie != atomic64_read(&sk->sk_cookie))
+			return ERR_PTR(-ECONNREFUSED);
+	}
+
 	dentry = unix_sk(sk)->path.dentry;
 	if (dentry)
 		touch_atime(&unix_sk(sk)->path);
@@ -1210,14 +1220,14 @@ static struct sock *unix_find_abstract(struct net *net,
 
 static struct sock *unix_find_other(struct net *net,
 				    struct sockaddr_un *sunaddr,
-				    int addr_len, int type)
+				    int addr_len, int type, int flags)
 {
 	struct sock *sk;
 
 	if (sunaddr->sun_path[0])
 		sk = unix_find_bsd(sunaddr, addr_len, type);
 	else
-		sk = unix_find_abstract(net, sunaddr, addr_len, type);
+		sk = unix_find_abstract(net, sunaddr, addr_len, type, flags);
 
 	return sk;
 }
@@ -1473,7 +1483,7 @@ static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 		}
 
 restart:
-		other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
+		other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type, flags);
 		if (IS_ERR(other)) {
 			err = PTR_ERR(other);
 			goto out;
@@ -1620,7 +1630,7 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 
 restart:
 	/*  Find listening sock. */
-	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
+	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, flags);
 	if (IS_ERR(other)) {
 		err = PTR_ERR(other);
 		goto out_free_skb;
@@ -2089,7 +2099,7 @@ static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 	if (msg->msg_namelen) {
 lookup:
 		other = unix_find_other(sock_net(sk), msg->msg_name,
-					msg->msg_namelen, sk->sk_type);
+					msg->msg_namelen, sk->sk_type, 0);
 		if (IS_ERR(other)) {
 			err = PTR_ERR(other);
 			goto out_free;
-- 
2.47.2

Re: [PATCH v5 1/9] coredump: massage format_corname()

[Alexander Mikhalitsyn]

Hi Christian!

Am Fr., 9. Mai 2025 um 12:25 Uhr schrieb Christian Brauner <brauner@kernel.org>:
>
> We're going to extend the coredump code in follow-up patches.
> Clean it up so we can do this more easily.
>
> Signed-off-by: Christian Brauner <brauner@kernel.org>
> ---
>  fs/coredump.c | 41 ++++++++++++++++++++++++-----------------
>  1 file changed, 24 insertions(+), 17 deletions(-)
>
> diff --git a/fs/coredump.c b/fs/coredump.c
> index d740a0411266..281320ea351f 100644
> --- a/fs/coredump.c
> +++ b/fs/coredump.c
> @@ -76,9 +76,15 @@ static char core_pattern[CORENAME_MAX_SIZE] = "core";
>  static int core_name_size = CORENAME_MAX_SIZE;
>  unsigned int core_file_note_size_limit = CORE_FILE_NOTE_SIZE_DEFAULT;
>
> +enum coredump_type_t {
> +       COREDUMP_FILE = 1,
> +       COREDUMP_PIPE = 2,
> +};
> +
>  struct core_name {
>         char *corename;
>         int used, size;
> +       enum coredump_type_t core_type;
>  };
>
>  static int expand_corename(struct core_name *cn, int size)
> @@ -218,18 +224,21 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
>  {
>         const struct cred *cred = current_cred();
>         const char *pat_ptr = core_pattern;
> -       int ispipe = (*pat_ptr == '|');
>         bool was_space = false;
>         int pid_in_pattern = 0;
>         int err = 0;
>
>         cn->used = 0;
>         cn->corename = NULL;
> +       if (*pat_ptr == '|')
> +               cn->core_type = COREDUMP_PIPE;
> +       else
> +               cn->core_type = COREDUMP_FILE;
>         if (expand_corename(cn, core_name_size))
>                 return -ENOMEM;
>         cn->corename[0] = '\0';
>
> -       if (ispipe) {
> +       if (cn->core_type == COREDUMP_PIPE) {
>                 int argvs = sizeof(core_pattern) / 2;
>                 (*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
>                 if (!(*argv))
> @@ -247,7 +256,7 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
>                  * Split on spaces before doing template expansion so that
>                  * %e and %E don't get split if they have spaces in them
>                  */
> -               if (ispipe) {
> +               if (cn->core_type == COREDUMP_PIPE) {
>                         if (isspace(*pat_ptr)) {
>                                 if (cn->used != 0)
>                                         was_space = true;
> @@ -353,7 +362,7 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
>                                  * Installing a pidfd only makes sense if
>                                  * we actually spawn a usermode helper.
>                                  */
> -                               if (!ispipe)
> +                               if (!(cn->core_type != COREDUMP_PIPE))

Shouldn't it be:
if (cn->core_type != COREDUMP_PIPE)

Except this, LGTM

Reviewed-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>

>                                         break;
>
>                                 /*
> @@ -384,12 +393,12 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm,
>          * If core_pattern does not include a %p (as is the default)
>          * and core_uses_pid is set, then .%pid will be appended to
>          * the filename. Do not do this for piped commands. */
> -       if (!ispipe && !pid_in_pattern && core_uses_pid) {
> +       if (!(cn->core_type == COREDUMP_PIPE) && !pid_in_pattern && core_uses_pid) {
>                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
>                 if (err)
>                         return err;
>         }
> -       return ispipe;
> +       return 0;
>  }
>
>  static int zap_process(struct signal_struct *signal, int exit_code)
> @@ -583,7 +592,6 @@ void do_coredump(const kernel_siginfo_t *siginfo)
>         const struct cred *old_cred;
>         struct cred *cred;
>         int retval = 0;
> -       int ispipe;
>         size_t *argv = NULL;
>         int argc = 0;
>         /* require nonrelative corefile path and be extra careful */
> @@ -632,19 +640,18 @@ void do_coredump(const kernel_siginfo_t *siginfo)
>
>         old_cred = override_creds(cred);
>
> -       ispipe = format_corename(&cn, &cprm, &argv, &argc);
> +       retval = format_corename(&cn, &cprm, &argv, &argc);
> +       if (retval < 0) {
> +               coredump_report_failure("format_corename failed, aborting core");
> +               goto fail_unlock;
> +       }
>
> -       if (ispipe) {
> +       if (cn.core_type == COREDUMP_PIPE) {
>                 int argi;
>                 int dump_count;
>                 char **helper_argv;
>                 struct subprocess_info *sub_info;
>
> -               if (ispipe < 0) {
> -                       coredump_report_failure("format_corename failed, aborting core");
> -                       goto fail_unlock;
> -               }
> -
>                 if (cprm.limit == 1) {
>                         /* See umh_coredump_setup() which sets RLIMIT_CORE = 1.
>                          *
> @@ -695,7 +702,7 @@ void do_coredump(const kernel_siginfo_t *siginfo)
>                         coredump_report_failure("|%s pipe failed", cn.corename);
>                         goto close_fail;
>                 }
> -       } else {
> +       } else if (cn.core_type == COREDUMP_FILE) {
>                 struct mnt_idmap *idmap;
>                 struct inode *inode;
>                 int open_flags = O_CREAT | O_WRONLY | O_NOFOLLOW |
> @@ -823,13 +830,13 @@ void do_coredump(const kernel_siginfo_t *siginfo)
>                 file_end_write(cprm.file);
>                 free_vma_snapshot(&cprm);
>         }
> -       if (ispipe && core_pipe_limit)
> +       if ((cn.core_type == COREDUMP_PIPE) && core_pipe_limit)
>                 wait_for_dump_helpers(cprm.file);
>  close_fail:
>         if (cprm.file)
>                 filp_close(cprm.file, NULL);
>  fail_dropcount:
> -       if (ispipe)
> +       if (cn.core_type == COREDUMP_PIPE)
>                 atomic_dec(&core_dump_count);
>  fail_unlock:
>         kfree(argv);
>
> --
> 2.47.2
>

Re: [PATCH v5 2/9] coredump: massage do_coredump()

[Alexander Mikhalitsyn]

Am Fr., 9. Mai 2025 um 12:26 Uhr schrieb Christian Brauner <brauner@kernel.org>:
>
> We're going to extend the coredump code in follow-up patches.
> Clean it up so we can do this more easily.
>
> Signed-off-by: Christian Brauner <brauner@kernel.org>

Reviewed-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>

> ---
>  fs/coredump.c | 122 +++++++++++++++++++++++++++++++---------------------------
>  1 file changed, 65 insertions(+), 57 deletions(-)
>
> diff --git a/fs/coredump.c b/fs/coredump.c
> index 281320ea351f..41491dbfafdf 100644
> --- a/fs/coredump.c
> +++ b/fs/coredump.c
> @@ -646,63 +646,8 @@ void do_coredump(const kernel_siginfo_t *siginfo)
>                 goto fail_unlock;
>         }
>
> -       if (cn.core_type == COREDUMP_PIPE) {
> -               int argi;
> -               int dump_count;
> -               char **helper_argv;
> -               struct subprocess_info *sub_info;
> -
> -               if (cprm.limit == 1) {
> -                       /* See umh_coredump_setup() which sets RLIMIT_CORE = 1.
> -                        *
> -                        * Normally core limits are irrelevant to pipes, since
> -                        * we're not writing to the file system, but we use
> -                        * cprm.limit of 1 here as a special value, this is a
> -                        * consistent way to catch recursive crashes.
> -                        * We can still crash if the core_pattern binary sets
> -                        * RLIM_CORE = !1, but it runs as root, and can do
> -                        * lots of stupid things.
> -                        *
> -                        * Note that we use task_tgid_vnr here to grab the pid
> -                        * of the process group leader.  That way we get the
> -                        * right pid if a thread in a multi-threaded
> -                        * core_pattern process dies.
> -                        */
> -                       coredump_report_failure("RLIMIT_CORE is set to 1, aborting core");
> -                       goto fail_unlock;
> -               }
> -               cprm.limit = RLIM_INFINITY;
> -
> -               dump_count = atomic_inc_return(&core_dump_count);
> -               if (core_pipe_limit && (core_pipe_limit < dump_count)) {
> -                       coredump_report_failure("over core_pipe_limit, skipping core dump");
> -                       goto fail_dropcount;
> -               }
> -
> -               helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
> -                                           GFP_KERNEL);
> -               if (!helper_argv) {
> -                       coredump_report_failure("%s failed to allocate memory", __func__);
> -                       goto fail_dropcount;
> -               }
> -               for (argi = 0; argi < argc; argi++)
> -                       helper_argv[argi] = cn.corename + argv[argi];
> -               helper_argv[argi] = NULL;
> -
> -               retval = -ENOMEM;
> -               sub_info = call_usermodehelper_setup(helper_argv[0],
> -                                               helper_argv, NULL, GFP_KERNEL,
> -                                               umh_coredump_setup, NULL, &cprm);
> -               if (sub_info)
> -                       retval = call_usermodehelper_exec(sub_info,
> -                                                         UMH_WAIT_EXEC);
> -
> -               kfree(helper_argv);
> -               if (retval) {
> -                       coredump_report_failure("|%s pipe failed", cn.corename);
> -                       goto close_fail;
> -               }
> -       } else if (cn.core_type == COREDUMP_FILE) {
> +       switch (cn.core_type) {
> +       case COREDUMP_FILE: {
>                 struct mnt_idmap *idmap;
>                 struct inode *inode;
>                 int open_flags = O_CREAT | O_WRONLY | O_NOFOLLOW |
> @@ -796,6 +741,69 @@ void do_coredump(const kernel_siginfo_t *siginfo)
>                 if (do_truncate(idmap, cprm.file->f_path.dentry,
>                                 0, 0, cprm.file))
>                         goto close_fail;
> +               break;
> +       }
> +       case COREDUMP_PIPE: {
> +               int argi;
> +               int dump_count;
> +               char **helper_argv;
> +               struct subprocess_info *sub_info;
> +
> +               if (cprm.limit == 1) {
> +                       /* See umh_coredump_setup() which sets RLIMIT_CORE = 1.
> +                        *
> +                        * Normally core limits are irrelevant to pipes, since
> +                        * we're not writing to the file system, but we use
> +                        * cprm.limit of 1 here as a special value, this is a
> +                        * consistent way to catch recursive crashes.
> +                        * We can still crash if the core_pattern binary sets
> +                        * RLIM_CORE = !1, but it runs as root, and can do
> +                        * lots of stupid things.
> +                        *
> +                        * Note that we use task_tgid_vnr here to grab the pid
> +                        * of the process group leader.  That way we get the
> +                        * right pid if a thread in a multi-threaded
> +                        * core_pattern process dies.
> +                        */
> +                       coredump_report_failure("RLIMIT_CORE is set to 1, aborting core");
> +                       goto fail_unlock;
> +               }
> +               cprm.limit = RLIM_INFINITY;
> +
> +               dump_count = atomic_inc_return(&core_dump_count);
> +               if (core_pipe_limit && (core_pipe_limit < dump_count)) {
> +                       coredump_report_failure("over core_pipe_limit, skipping core dump");
> +                       goto fail_dropcount;
> +               }
> +
> +               helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
> +                                           GFP_KERNEL);
> +               if (!helper_argv) {
> +                       coredump_report_failure("%s failed to allocate memory", __func__);
> +                       goto fail_dropcount;
> +               }
> +               for (argi = 0; argi < argc; argi++)
> +                       helper_argv[argi] = cn.corename + argv[argi];
> +               helper_argv[argi] = NULL;
> +
> +               retval = -ENOMEM;
> +               sub_info = call_usermodehelper_setup(helper_argv[0],
> +                                               helper_argv, NULL, GFP_KERNEL,
> +                                               umh_coredump_setup, NULL, &cprm);
> +               if (sub_info)
> +                       retval = call_usermodehelper_exec(sub_info,
> +                                                         UMH_WAIT_EXEC);
> +
> +               kfree(helper_argv);
> +               if (retval) {
> +                       coredump_report_failure("|%s pipe failed", cn.corename);
> +                       goto close_fail;
> +               }
> +               break;
> +       }
> +       default:
> +               WARN_ON_ONCE(true);
> +               goto close_fail;
>         }
>
>         /* get us an unshared descriptor table; almost always a no-op */
>
> --
> 2.47.2
>

Re: [PATCH v5 3/9] coredump: reflow dump helpers a little

[Alexander Mikhalitsyn]

Am Fr., 9. Mai 2025 um 12:26 Uhr schrieb Christian Brauner <brauner@kernel.org>:
>
> They look rather messy right now.
>
> Signed-off-by: Christian Brauner <brauner@kernel.org>

Reviewed-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>

> ---
>  fs/coredump.c | 22 +++++++++++-----------
>  1 file changed, 11 insertions(+), 11 deletions(-)
>
> diff --git a/fs/coredump.c b/fs/coredump.c
> index 41491dbfafdf..b2eda7b176e4 100644
> --- a/fs/coredump.c
> +++ b/fs/coredump.c
> @@ -867,10 +867,9 @@ static int __dump_emit(struct coredump_params *cprm, const void *addr, int nr)
>         struct file *file = cprm->file;
>         loff_t pos = file->f_pos;
>         ssize_t n;
> +
>         if (cprm->written + nr > cprm->limit)
>                 return 0;
> -
> -
>         if (dump_interrupted())
>                 return 0;
>         n = __kernel_write(file, addr, nr, &pos);
> @@ -887,20 +886,21 @@ static int __dump_skip(struct coredump_params *cprm, size_t nr)
>  {
>         static char zeroes[PAGE_SIZE];
>         struct file *file = cprm->file;
> +
>         if (file->f_mode & FMODE_LSEEK) {
> -               if (dump_interrupted() ||
> -                   vfs_llseek(file, nr, SEEK_CUR) < 0)
> +               if (dump_interrupted() || vfs_llseek(file, nr, SEEK_CUR) < 0)
>                         return 0;
>                 cprm->pos += nr;
>                 return 1;
> -       } else {
> -               while (nr > PAGE_SIZE) {
> -                       if (!__dump_emit(cprm, zeroes, PAGE_SIZE))
> -                               return 0;
> -                       nr -= PAGE_SIZE;
> -               }
> -               return __dump_emit(cprm, zeroes, nr);
>         }
> +
> +       while (nr > PAGE_SIZE) {
> +               if (!__dump_emit(cprm, zeroes, PAGE_SIZE))
> +                       return 0;
> +               nr -= PAGE_SIZE;
> +       }
> +
> +       return __dump_emit(cprm, zeroes, nr);
>  }
>
>  int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
>
> --
> 2.47.2
>

Re: [PATCH v5 4/9] coredump: add coredump socket

[Luca Boccassi]

On Fri, 9 May 2025 at 16:40, Christian Brauner <brauner@kernel.org> wrote:
>
> > Userspace can set /proc/sys/kernel/core_pattern to:
> >
> >         @linuxafsk/coredump_socket
>
> I have one other proposal that:
>
> - avoids reserving a specific address
> - doesn't require bpf or lsm to be safe
> - allows for safe restart and crashes of the coredump sever
>
> To set up a coredump socket the coredump server must allocate a socket
> cookie for the listening socket via SO_COOKIE. The socket cookie must be
> used as the prefix in the abstract address for the coredump socket. It
> can be followed by a \0 byte and then followed by whatever the coredump
> server wants. For example:
>
> 12345678\0coredump.socket
>
> When a task crashes and generates a coredump it will find the provided
> address but also compare the prefixed SO_COOKIE value with the socket
> cookie of the socket listening at that address. If they don't match it
> will refuse to connect.
>
> So even if the coredump server restarts or crashes and unprivileged
> userspace recycles the socket address for an attack the crashing process
> will detect this as the new listening socket will have gotten either a
> new or no SO_COOKIE and the crashing process will not connect.
>
> The coredump server just sets /proc/sys/kernel/core_pattern to:
>
>         @SO_COOKIE/whatever
>
> The "@" at the beginning indicates to the kernel that the abstract
> AF_UNIX coredump socket will be used to process coredumps and the
> indicating the end of the SO_COOKIE and the rest of the name.
>
> Appended what that would look like.

We set the core pattern via sysctl, so personally I'd prefer if it
remained fixed rather than being dynamic and have to be set at
runtime, so that it doesn't require anything to run and it continues
to be activated on triggers only

Re: [PATCH v5 4/9] coredump: add coredump socket

[Kuniyuki Iwashima]

From: Christian Brauner <brauner@kernel.org>
Date: Fri, 9 May 2025 17:40:14 +0200
> > Userspace can set /proc/sys/kernel/core_pattern to:
> > 
> >         @linuxafsk/coredump_socket
> 
> I have one other proposal that:
> 
> - avoids reserving a specific address
> - doesn't require bpf or lsm to be safe
> - allows for safe restart and crashes of the coredump sever
> 
> To set up a coredump socket the coredump server must allocate a socket
> cookie for the listening socket via SO_COOKIE. The socket cookie must be
> used as the prefix in the abstract address for the coredump socket. It
> can be followed by a \0 byte and then followed by whatever the coredump
> server wants. For example:
> 
> 12345678\0coredump.socket
> 
> When a task crashes and generates a coredump it will find the provided
> address but also compare the prefixed SO_COOKIE value with the socket
> cookie of the socket listening at that address. If they don't match it
> will refuse to connect.
> 
> So even if the coredump server restarts or crashes and unprivileged
> userspace recycles the socket address for an attack the crashing process
> will detect this as the new listening socket will have gotten either a
> new or no SO_COOKIE and the crashing process will not connect.
> 
> The coredump server just sets /proc/sys/kernel/core_pattern to:
> 
>         @SO_COOKIE/whatever
> 
> The "@" at the beginning indicates to the kernel that the abstract
> AF_UNIX coredump socket will be used to process coredumps and the
> indicating the end of the SO_COOKIE and the rest of the name.
> 
> Appended what that would look like.

Thank you, this looks much nicer to me.


[...]
> Userspace can set /proc/sys/kernel/core_pattern to:
> 
>         @SO_COOKIE/whatever
> 
> The "@" at the beginning indicates to the kernel that the abstract
> AF_UNIX coredump socket will be used to process coredumps.
> 
> When the coredump server sets up a coredump socket it must allocate a
> socket cookie for it and use it as the prefix in the abstract address.
> It may be followed by a zero byte and whatever other name the server may
> want.
[...]
> +
> +		/* Format is @socket_cookie\0whatever. */
> +		p = strchr(addr.sun_path + 1, '/');
> +		if (p)
> +			*p = '\0';

nit: the '\0' seems optional, @SO_COOKIEwhatever\0



> diff --git a/include/linux/net.h b/include/linux/net.h
> index 0ff950eecc6b..3f467786bdc9 100644
> --- a/include/linux/net.h
> +++ b/include/linux/net.h
> @@ -82,6 +82,8 @@ enum sock_type {
>  #define SOCK_NONBLOCK	O_NONBLOCK
>  #endif
>  
> +#define SOCK_COREDUMP O_NOCTTY
> +
>  #endif /* ARCH_HAS_SOCKET_TYPES */
>  
>  /**
> diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
> index 472f8aa9ea15..944248d7c5be 100644
> --- a/net/unix/af_unix.c
> +++ b/net/unix/af_unix.c
> @@ -101,6 +101,7 @@
>  #include <linux/string.h>
>  #include <linux/uaccess.h>
>  #include <linux/pidfs.h>
> +#include <linux/kstrtox.h>

nit: please sort in alphabetical order.  It was cleaned up recently.


>  #include <net/af_unix.h>
>  #include <net/net_namespace.h>
>  #include <net/scm.h>
> @@ -1191,7 +1192,7 @@ static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
>  
>  static struct sock *unix_find_abstract(struct net *net,
>  				       struct sockaddr_un *sunaddr,
> -				       int addr_len, int type)
> +				       int addr_len, int type, int flags)
>  {
>  	unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
>  	struct dentry *dentry;
> @@ -1201,6 +1202,15 @@ static struct sock *unix_find_abstract(struct net *net,
>  	if (!sk)
>  		return ERR_PTR(-ECONNREFUSED);
>  
> +	if (flags & SOCK_COREDUMP) {
> +		u64 cookie;
> +
> +		if (kstrtou64(sunaddr->sun_path, 0, &cookie))
> +			return ERR_PTR(-ECONNREFUSED);
> +		if (cookie != atomic64_read(&sk->sk_cookie))
> +			return ERR_PTR(-ECONNREFUSED);
> +	}
> +
>  	dentry = unix_sk(sk)->path.dentry;
>  	if (dentry)
>  		touch_atime(&unix_sk(sk)->path);
> @@ -1210,14 +1220,14 @@ static struct sock *unix_find_abstract(struct net *net,
>  
>  static struct sock *unix_find_other(struct net *net,
>  				    struct sockaddr_un *sunaddr,
> -				    int addr_len, int type)
> +				    int addr_len, int type, int flags)
>  {
>  	struct sock *sk;
>  
>  	if (sunaddr->sun_path[0])
>  		sk = unix_find_bsd(sunaddr, addr_len, type);
>  	else
> -		sk = unix_find_abstract(net, sunaddr, addr_len, type);
> +		sk = unix_find_abstract(net, sunaddr, addr_len, type, flags);
>  
>  	return sk;
>  }
> @@ -1473,7 +1483,7 @@ static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
>  		}
>  
>  restart:
> -		other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
> +		other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type, flags);

The flag should be 0 as we don't use SOCK_DGRAM for coredump.


>  		if (IS_ERR(other)) {
>  			err = PTR_ERR(other);
>  			goto out;
> @@ -1620,7 +1630,7 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
>
>  restart:
>  	/*  Find listening sock. */
> -	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
> +	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, flags);
>  	if (IS_ERR(other)) {
>  		err = PTR_ERR(other);
>  		goto out_free_skb;
> @@ -2089,7 +2099,7 @@ static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
>  	if (msg->msg_namelen) {
>  lookup:
>  		other = unix_find_other(sock_net(sk), msg->msg_name,
> -					msg->msg_namelen, sk->sk_type);
> +					msg->msg_namelen, sk->sk_type, 0);
>  		if (IS_ERR(other)) {
>  			err = PTR_ERR(other);
>  			goto out_free;
> -- 
> 2.47.2