Merge patch series "binfmt_misc: bpf-backed binary type handlers"

Christian Brauner <brauner@kernel.org> says:

binfmt_misc: bpf-backed binary type handlers

This is a POC for the nix people and Farid and Eric in particular. I
would take my hands off the wheel now that I POCed this and hand it to
Farid if he likes to take it forward.

VL;MR (very long, must read):

For a while now Farid has been trying to make relocatable, hermetic
binaries (think Nix-style store layouts) work without patchelf tricks
or wrapper scripts. For such binaries the right dynamic loader can only
be determined relative to the location of the binary itself, which
neither PT_INTERP nor a fixed binfmt_misc interpreter string can
express.

The first attempt was $ORIGIN expansion in PT_INTERP [1]. I pushed back
on that. Userspace guards $ORIGIN behind AT_SECURE so the kernel would
have to make the used loader depend on the type of binary, LSMs would
need a say, it changes long-standing behavior in ways that are ripe for
loader injection attacks, and bprm->file may not have a usable path at
all (memfds, deleted files, unresolvable paths). Making the kernel
splice bprm->file back together with PT_INTERP is terrible. The second
attempt was a pluggable ELF interpreter loader registry [2] which would
mean actual kernel modules for custom binary formats. Also no.
binfmt_misc was invented to kill exactly this horrendous past.

What I suggested instead [3] was to put this where delegating binary
formats to userspace already lives: binfmt_misc. The only things
binfmt_misc cannot do today are matching programmatically and computing
the interpreter per binary instead of using a fixed string recorded at
registration time. Farid prototyped that with an eBPF program [4] and
it turned out quite workable, but the prototype ran a SOCKET_FILTER
program over bprm->buf, added a new helper to the frozen uapi helper
list, and returned the computed path through per-CPU memory.

This series is the proposal turned into what I think the bpf side
{c,sh}ould actually look like. It is a POC: it builds, the selftests
pass, and the design is what I want to discuss. The selftests are
Farid's from his v2 posting, adapted to the contract below.

A handler is an instance of the new binfmt_misc_ops struct_ops with a
name and two ops:

	struct binfmt_misc_ops {
		bool (*match)(struct linux_binprm *bprm);
		int (*load)(struct linux_binprm *bprm);
		char name[BINFMT_MISC_OPS_NAME_MAX];
	};

Both programs receive the bprm as a trusted BTF pointer and both are
sleepable. The match program decides from the entry lookup walk whether
the handler applies, under the same rules as magic matching:
registration order, first match wins. It is not limited to the
prefetched 256 bytes in bprm->buf: it can read arbitrary file content
through bpf_dynptr_from_file(), e.g. to find an ELF interpreter
segment at whatever offset it sits. That is what makes multiple
independent handlers workable at all - a handler that cannot read the
file would have to match broadly and reject from its load program,
stealing the binaries of every handler registered after it. To make
this safe the entry walk becomes an SRCU read-side section. The load
program of the matched handler then selects the interpreter, reading
the file the same way and resolving the binary's location via
bpf_path_d_path() on &bprm->file->f_path. That also solves the
prototype's limitation of only seeing the first 256 bytes of the file.
Selecting is the load program's privilege: the verifier rejects the
selection kfuncs in match, keyed off the struct_ops member a program
attaches to. A match commits the exec to the handler: a failing load
fails the exec instead of falling through to later entries, with
-ENOEXEC handing over to the remaining binary formats, so the walk is
never left and re-entered.

The genuinely new piece of bpf surface is a small family of kfuncs:

	int bpf_binprm_set_interp(struct linux_binprm *bprm,
				  const char *path, size_t path__sz);
	int bpf_binprm_set_interp_arg(struct linux_binprm *bprm,
				      const char *arg, size_t arg__sz);
	int bpf_binprm_set_flags(struct linux_binprm *bprm,
				 enum bpf_binprm_flags flags);

staging the selected interpreter, an optional single argument for it
(the slot the optional argument of a #! interpreter line has), and the
per-exec invocation flags - 'P', 'C' and 'O' equivalents. Selection
cannot go through bprm_change_interp() directly because
load_misc_binary() copies bprm->interp into argv[1] after the program
ran, hence the staging fields added in patch 1.

Registering (attaching) the struct_ops map publishes the handler under
its name in a registry keyed by the registering task's user namespace.
Activation reuses the existing text interface with a new 'B' type where
the interpreter field carries the handler name - it consistently names
whoever supplies the interpreter - and offset, magic, and mask must be
empty:

	echo ':origin:B::::nix:' > /proc/sys/fs/binfmt_misc/register

This keeps the existing permission and namespacing model completely
intact. Activating a handler requires the same write access to a
binfmt_misc instance as any other registration, a container mounting
its own instance escapes the host's entries exactly as before, and
shadowing e.g. all ELF binaries takes the same privilege as a static
'M' entry matching \x7fELF does today.

The only novelty is that matching becomes programmable. Handler lookup
walks the user namespace hierarchy upwards, mirroring how binfmt_misc
instances themselves are resolved, so a handler registered on the host
can be activated from a container's own instance without being forced
upon it.

The computed interpreter is opened with open_exec() under the caller's
credentials and goes through the full LSM vetting as the next binprm
level, exactly like a statically registered interpreter, so the program
cannot widen access. It only ever redirects the caller to something the
caller could exec anyway.

A 'B' entry carries no flags in the register string: the load program
chooses the invocation flags per exec through bpf_binprm_set_flags()
instead. BPF_BINPRM_PRESERVE_ARGV0, BPF_BINPRM_CREDENTIALS and
BPF_BINPRM_EXECFD keep the static 'P', 'C' and 'O' semantics -
BPF_BINPRM_CREDENTIALS honors the matched binary's suid bits exactly
as a static 'C' entry does, with the setuid transition gated by
vfsuid_has_mapping() in the caller's user namespace either way, which
makes 'B' handlers usable for a per-binary loader over setuid
binaries. 'F' (pre-open a fixed interpreter) is rejected: a 'B' entry
has no fixed interpreter. AT_EXECVE_CHECK never invokes programs and
interpreter chains stay capped by the usual ELOOP depth.

A handler for the Nix case then looks roughly like:

	SEC("struct_ops.s/match")
	bool BPF_PROG(nix_match, struct linux_binprm *bprm)
	{
		return !bpf_strncmp(bprm->buf, 4, "\x7f" "ELF");
	}

	SEC("struct_ops.s/load")
	int BPF_PROG(nix_load, struct linux_binprm *bprm)
	{
		char path[256];
		long n;

		n = bpf_path_d_path(&bprm->file->f_path, path, sizeof(path));
		if (n < 0)
			return n;

		/* derive the loader location from the binary's path */

		return bpf_binprm_set_interp(bprm, path, sizeof(path));
	}

	SEC(".struct_ops.link")
	struct binfmt_misc_ops nix = {
		.match = (void *)nix_match,
		.load = (void *)nix_load,
		.name = "nix",
	};

Farid, this should slot underneath your qemu demo from [4] with the
program ported to struct_ops. Feel free to take it from here.

[1]: https://lore.kernel.org/20260622043934.179879-1-farid.m.zakaria@gmail.com
[2]: https://lore.kernel.org/20260702214247.1253741-1-farid.m.zakaria@gmail.com
[3]: https://lore.kernel.org/20260703-meditation-ratsuchende-moratorium-9ecdf1f3f8bb@brauner
[4]: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com

* patches from https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-0-57b7529c002c@kernel.org:
  selftests/exec: add binfmt_misc bpf-backed handler test
  binfmt_misc: let a bpf handler choose the invocation flags per exec
  binfmt_misc: let bpf handlers pass an argument to the interpreter
  bpf: allow fs kfuncs for binfmt_misc_ops programs
  binfmt_misc: wire up bpf-backed 'B' entries
  binfmt_misc: let the entry lookup walk sleep
  binfmt_misc: add binfmt_misc_ops bpf struct_ops
  exec: stash bpf-selected interpreter state in struct linux_binprm

Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-0-57b7529c002c@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>

7 files changed, 642 insertions, 0 deletions

diff --git a/tools/testing/selftests/exec/.gitignore b/tools/testing/selftests/exec/.gitignore
index 7f3d1ae762ec..8b93b405c424 100644
--- a/tools/testing/selftests/exec/.gitignore
+++ b/tools/testing/selftests/exec/.gitignore
@@ -19,3 +19,8 @@ null-argv
 xxxxxxxx*
 pipe
 S_I*.test
+binfmt_misc_bpf
+binfmt_bpf_interp
+binfmt_bpf_app
+*.bpf.o
+vmlinux.h
diff --git a/tools/testing/selftests/exec/Makefile b/tools/testing/selftests/exec/Makefile
index 45a3cfc435cf..ec66c1fecfc0 100644
--- a/tools/testing/selftests/exec/Makefile
+++ b/tools/testing/selftests/exec/Makefile
@@ -21,6 +21,26 @@ TEST_GEN_PROGS += recursion-depth
 TEST_GEN_PROGS += null-argv
 TEST_GEN_PROGS += check-exec
 
+# binfmt_misc bpf-backed ('B') handler test: a libbpf harness plus its
+# struct_ops objects and the test interpreter/app it routes between. Only
+# built when clang, bpftool, the vmlinux BTF and libbpf are all present
+# (HAVE_BPF_TOOLCHAIN=y forces it) so the other exec selftests don't grow
+# a bpf toolchain dependency.
+CLANG ?= clang
+BPFTOOL ?= bpftool
+VMLINUX_BTF ?= /sys/kernel/btf/vmlinux
+HAVE_BPF_TOOLCHAIN ?= $(shell command -v $(CLANG) >/dev/null 2>&1 && \
+			command -v $(BPFTOOL) >/dev/null 2>&1 && \
+			test -r $(VMLINUX_BTF) && \
+			pkg-config --exists libbpf 2>/dev/null && echo y)
+ifeq ($(HAVE_BPF_TOOLCHAIN),y)
+TEST_GEN_PROGS += binfmt_misc_bpf
+TEST_GEN_FILES += bpf_interp.bpf.o nix_origin.bpf.o
+TEST_GEN_FILES += binfmt_bpf_interp binfmt_bpf_app
+else
+$(info exec selftests: skipping binfmt_misc_bpf, needs clang, bpftool, vmlinux BTF and libbpf)
+endif
+
 EXTRA_CLEAN := $(OUTPUT)/subdir.moved $(OUTPUT)/execveat.moved $(OUTPUT)/xxxxx*	\
 	       $(OUTPUT)/S_I*.test
 
@@ -55,3 +75,31 @@ $(OUTPUT)/script-exec.inc: $(CHECK_EXEC_SAMPLES)/script-exec.inc
 	cp $< $@
 $(OUTPUT)/script-noexec.inc: $(CHECK_EXEC_SAMPLES)/script-noexec.inc
 	cp $< $@
+
+# --- binfmt_misc bpf ('B') handler test ---------------------------------
+# The struct_ops bpf objects are compiled against the running kernel's BTF.
+# CLANG/BPFTOOL/VMLINUX_BTF are set above next to the toolchain check;
+# override LIBBPF_CFLAGS/LDLIBS to point at a libbpf install.
+BPF_CFLAGS ?= -I$(OUTPUT)
+LIBBPF_CFLAGS ?=
+LIBBPF_LDLIBS ?= -lbpf -lelf -lz
+
+$(OUTPUT)/vmlinux.h:
+	$(BPFTOOL) btf dump file $(VMLINUX_BTF) format c > $@
+	sed -i '/__ksym;$$/d' $@
+
+$(OUTPUT)/%.bpf.o: %.bpf.c $(OUTPUT)/vmlinux.h
+	$(CLANG) -g -O2 -target bpf -mcpu=v3 $(BPF_CFLAGS) $(LIBBPF_CFLAGS) -c $< -o $@
+
+$(OUTPUT)/binfmt_misc_bpf: binfmt_misc_bpf.c
+	$(CC) $(CFLAGS) $(LIBBPF_CFLAGS) $(LDFLAGS) $< $(LIBBPF_LDLIBS) -o $@
+
+$(OUTPUT)/binfmt_bpf_interp: binfmt_bpf_interp.c
+	$(CC) $(CFLAGS) $(LDFLAGS) $< -o $@
+
+# PT_INTERP is set to the literal "$ORIGIN/binfmt_bpf_interp"; the nix_origin
+# handler resolves it relative to the binary at run time.
+$(OUTPUT)/binfmt_bpf_app: binfmt_bpf_app.c
+	$(CC) $(CFLAGS) $(LDFLAGS) -Wl,--dynamic-linker,'$$ORIGIN/binfmt_bpf_interp' $< -o $@
+
+EXTRA_CLEAN += $(OUTPUT)/vmlinux.h $(OUTPUT)/bpf_interp.bpf.o $(OUTPUT)/nix_origin.bpf.o
diff --git a/tools/testing/selftests/exec/binfmt_bpf_app.c b/tools/testing/selftests/exec/binfmt_bpf_app.c
new file mode 100644
index 000000000000..472270f148bc
--- /dev/null
+++ b/tools/testing/selftests/exec/binfmt_bpf_app.c
@@ -0,0 +1,12 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * A relocatable binary for the binfmt_misc_bpf $ORIGIN case. The Makefile
+ * links it with PT_INTERP set to the literal "$ORIGIN/binfmt_bpf_interp"
+ * (-Wl,--dynamic-linker), which the kernel ELF loader cannot resolve. The
+ * nix_origin bpf handler resolves it relative to this binary's directory and
+ * routes execution to the co-located interpreter.
+ */
+int main(void)
+{
+	return 0;
+}
diff --git a/tools/testing/selftests/exec/binfmt_bpf_interp.c b/tools/testing/selftests/exec/binfmt_bpf_interp.c
new file mode 100644
index 000000000000..2db205f095b2
--- /dev/null
+++ b/tools/testing/selftests/exec/binfmt_bpf_interp.c
@@ -0,0 +1,15 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Test interpreter for the binfmt_misc_bpf selftest. A bpf-backed 'B' handler
+ * routes a matched binary here; printing this marker proves the program's
+ * chosen interpreter actually ran.
+ */
+#include <unistd.h>
+
+int main(int argc, char **argv)
+{
+	(void)argc;
+	(void)argv;
+	write(1, "BPF_INTERP_RAN\n", 15);
+	return 0;
+}
diff --git a/tools/testing/selftests/exec/binfmt_misc_bpf.c b/tools/testing/selftests/exec/binfmt_misc_bpf.c
new file mode 100644
index 000000000000..cb89d2766fe2
--- /dev/null
+++ b/tools/testing/selftests/exec/binfmt_misc_bpf.c
@@ -0,0 +1,277 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Selftest for binfmt_misc bpf-backed ('B') handlers.
+ *
+ * A handler is a struct binfmt_misc_ops struct_ops map with a sleepable match
+ * and a sleepable load program. Attaching it publishes it by name in the
+ * caller's user namespace; a 'B' entry referencing it by name in the
+ * interpreter field activates it:
+ *
+ *     echo ':name:B::::<handler>:' > /proc/sys/fs/binfmt_misc/register
+ *
+ * Two self-contained cases are exercised:
+ *
+ *   1. bpf_interp: the match program matches a synthetic aarch64 ELF header
+ *      from the prefetched bprm->buf and the load program routes it to a
+ *      fixed interpreter of its choosing.
+ *   2. nix_origin: the match program reads the binary's program headers to
+ *      commit only to a "$ORIGIN/..."-relative PT_INTERP and the load program
+ *      resolves it to an interpreter co-located with the binary (the
+ *      relocatable-loader case the kernel ELF loader cannot express).
+ *
+ * Both route to a test interpreter that prints BPF_INTERP_RAN, proving the
+ * program's chosen interpreter actually ran.
+ */
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <libgen.h>
+#include <sys/mount.h>
+#include <sys/stat.h>
+
+#include <bpf/btf.h>
+#include <bpf/libbpf.h>
+
+#define INTERP_PATH	"/tmp/binfmt_bpf_interp"
+#define AARCH64_PATH	"/tmp/binfmt_bpf_aarch64"
+#define RELOC_DIR	"/tmp/binfmt_reloc"
+#define BINFMT_REG	"/proc/sys/fs/binfmt_misc/register"
+#define EXPECT		"BPF_INTERP_RAN"
+
+static char testdir[512]; /* directory holding this test's built artifacts */
+
+static int copy_file(const char *src, const char *dst)
+{
+	char buf[4096];
+	int in, out;
+	ssize_t n;
+
+	in = open(src, O_RDONLY);
+	if (in < 0)
+		return -1;
+	out = open(dst, O_WRONLY | O_CREAT | O_TRUNC, 0755);
+	if (out < 0) {
+		close(in);
+		return -1;
+	}
+	while ((n = read(in, buf, sizeof(buf))) > 0) {
+		if (write(out, buf, n) != n) {
+			close(in);
+			close(out);
+			return -1;
+		}
+	}
+	close(in);
+	close(out);
+	return n < 0 ? -1 : 0;
+}
+
+/* A minimal 64-bit little-endian aarch64 ELF header, padded to the read size. */
+static int create_fake_aarch64(const char *path)
+{
+	unsigned char hdr[256] = {0};
+	int fd;
+
+	hdr[0] = 0x7f; hdr[1] = 'E'; hdr[2] = 'L'; hdr[3] = 'F';
+	hdr[4] = 2;			/* ELFCLASS64 */
+	hdr[5] = 1;			/* ELFDATA2LSB */
+	hdr[6] = 1;			/* EV_CURRENT */
+	hdr[16] = 2;			/* e_type = ET_EXEC */
+	hdr[18] = 183 & 0xff;		/* e_machine = EM_AARCH64 */
+	hdr[19] = (183 >> 8) & 0xff;
+	hdr[20] = 1;			/* e_version */
+
+	fd = open(path, O_WRONLY | O_CREAT | O_TRUNC, 0755);
+	if (fd < 0)
+		return -1;
+	if (write(fd, hdr, sizeof(hdr)) != (ssize_t)sizeof(hdr)) {
+		close(fd);
+		return -1;
+	}
+	close(fd);
+	return 0;
+}
+
+static int register_entry(const char *name, const char *handler)
+{
+	char rule[128];
+	int fd;
+	ssize_t n;
+
+	snprintf(rule, sizeof(rule), ":%s:B::::%s:", name, handler);
+	fd = open(BINFMT_REG, O_WRONLY);
+	if (fd < 0)
+		return -1;
+	n = write(fd, rule, strlen(rule));
+	close(fd);
+	return n < 0 ? -1 : 0;
+}
+
+static void unregister_entry(const char *name)
+{
+	char path[128];
+	int fd;
+
+	snprintf(path, sizeof(path), "/proc/sys/fs/binfmt_misc/%s", name);
+	fd = open(path, O_WRONLY);
+	if (fd >= 0) {
+		if (write(fd, "-1", 2) < 0)
+			; /* best effort */
+		close(fd);
+	}
+}
+
+static int check_output(const char *cmd, const char *expected)
+{
+	char buf[128];
+	FILE *fp;
+
+	fp = popen(cmd, "r");
+	if (!fp)
+		return -1;
+	if (!fgets(buf, sizeof(buf), fp)) {
+		pclose(fp);
+		return -1;
+	}
+	pclose(fp);
+	return strncmp(buf, expected, strlen(expected)) ? -1 : 0;
+}
+
+/*
+ * Load @objfile, attach its struct_ops map @handler (which publishes the
+ * handler), activate a 'B' entry named @entry that references it, run @target
+ * and check it produced @expect.
+ */
+static int run_case(const char *objfile, const char *handler,
+		    const char *entry, const char *target, const char *expect)
+{
+	struct bpf_object *obj;
+	struct bpf_map *map;
+	struct bpf_link *link;
+	int ret = -1;
+
+	obj = bpf_object__open_file(objfile, NULL);
+	if (!obj || libbpf_get_error(obj)) {
+		fprintf(stderr, "open %s failed\n", objfile);
+		return -1;
+	}
+	if (bpf_object__load(obj)) {
+		fprintf(stderr, "load %s failed (check dmesg for the verifier log)\n",
+			objfile);
+		goto close;
+	}
+	map = bpf_object__find_map_by_name(obj, handler);
+	if (!map) {
+		fprintf(stderr, "no struct_ops map '%s' in %s\n", handler, objfile);
+		goto close;
+	}
+	link = bpf_map__attach_struct_ops(map);
+	if (!link || libbpf_get_error(link)) {
+		fprintf(stderr, "attach struct_ops '%s' failed\n", handler);
+		goto close;
+	}
+	if (register_entry(entry, handler)) {
+		fprintf(stderr, "register 'B' entry '%s' failed\n", entry);
+		goto detach;
+	}
+	ret = check_output(target, expect);
+	unregister_entry(entry);
+detach:
+	bpf_link__destroy(link);
+close:
+	bpf_object__close(obj);
+	return ret;
+}
+
+int main(void)
+{
+	char src[600], obj[600], appdst[600], interpdst[600];
+	char exe[512];
+	ssize_t n;
+	int fail = 0;
+	struct stat st;
+	struct btf *btf;
+
+	if (getuid() != 0) {
+		fprintf(stderr, "Skipping: test must be run as root\n");
+		return 4; /* KSFT_SKIP */
+	}
+
+	/* The kernel must know struct binfmt_misc_ops (CONFIG_BINFMT_MISC_BPF). */
+	btf = btf__load_vmlinux_btf();
+	if (!btf || btf__find_by_name_kind(btf, "binfmt_misc_ops",
+					   BTF_KIND_STRUCT) < 0) {
+		fprintf(stderr,
+			"Skipping: no struct binfmt_misc_ops in the kernel BTF (CONFIG_BINFMT_MISC_BPF)\n");
+		btf__free(btf);
+		return 4; /* KSFT_SKIP */
+	}
+	btf__free(btf);
+
+	n = readlink("/proc/self/exe", exe, sizeof(exe) - 1);
+	if (n < 0) {
+		perror("readlink");
+		return 1;
+	}
+	exe[n] = '\0';
+	snprintf(testdir, sizeof(testdir), "%s", dirname(exe));
+
+	if (stat("/sys/fs/bpf", &st) < 0)
+		mkdir("/sys/fs/bpf", 0755);
+	mount("bpf", "/sys/fs/bpf", "bpf", 0, NULL);
+	if (access(BINFMT_REG, F_OK) < 0)
+		mount("binfmt_misc", "/proc/sys/fs/binfmt_misc", "binfmt_misc", 0, NULL);
+
+	/* Shared test interpreter. */
+	snprintf(src, sizeof(src), "%s/binfmt_bpf_interp", testdir);
+	if (copy_file(src, INTERP_PATH)) {
+		fprintf(stderr, "cannot install %s\n", INTERP_PATH);
+		return 1;
+	}
+
+	/* Case 1: match a synthetic aarch64 header -> fixed interpreter. */
+	printf("[*] case 1: match aarch64 header -> program-chosen interpreter\n");
+	if (create_fake_aarch64(AARCH64_PATH)) {
+		fprintf(stderr, "cannot create %s\n", AARCH64_PATH);
+		return 1;
+	}
+	snprintf(obj, sizeof(obj), "%s/bpf_interp.bpf.o", testdir);
+	if (run_case(obj, "bpf_interp", "test_bpf_interp", AARCH64_PATH, EXPECT) == 0)
+		printf("[+] case 1 passed\n");
+	else {
+		printf("[-] case 1 FAILED\n");
+		fail = 1;
+	}
+	unlink(AARCH64_PATH);
+
+	/* Case 2: $ORIGIN-relative PT_INTERP -> co-located interpreter. */
+	printf("[*] case 2: $ORIGIN interpreter resolved relative to the binary\n");
+	mkdir(RELOC_DIR, 0755);
+	snprintf(appdst, sizeof(appdst), "%s/app", RELOC_DIR);
+	snprintf(interpdst, sizeof(interpdst), "%s/binfmt_bpf_interp", RELOC_DIR);
+	snprintf(src, sizeof(src), "%s/binfmt_bpf_app", testdir);
+	if (copy_file(src, appdst) ||
+	    copy_file(INTERP_PATH, interpdst)) {
+		fprintf(stderr, "cannot set up %s\n", RELOC_DIR);
+		fail = 1;
+	} else {
+		snprintf(obj, sizeof(obj), "%s/nix_origin.bpf.o", testdir);
+		if (run_case(obj, "nix_origin", "test_bpf_origin", appdst, EXPECT) == 0)
+			printf("[+] case 2 passed\n");
+		else {
+			printf("[-] case 2 FAILED\n");
+			fail = 1;
+		}
+	}
+	unlink(appdst);
+	unlink(interpdst);
+	rmdir(RELOC_DIR);
+	unlink(INTERP_PATH);
+
+	if (!fail)
+		printf("[*] all binfmt_misc bpf cases passed\n");
+	return fail;
+}
diff --git a/tools/testing/selftests/exec/bpf_interp.bpf.c b/tools/testing/selftests/exec/bpf_interp.bpf.c
new file mode 100644
index 000000000000..8df2d2d01e25
--- /dev/null
+++ b/tools/testing/selftests/exec/bpf_interp.bpf.c
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * binfmt_misc_ops handler for the selftest's fixed-interpreter case: match a
+ * 64-bit aarch64 ELF header from the prefetched buffer and route it to a fixed
+ * interpreter chosen by the program. This is the portable, self-contained
+ * equivalent of routing a foreign binary to an emulator: it matches
+ * programmatically and computes the interpreter, but points at a test binary
+ * the harness installs rather than a system emulator.
+ */
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char _license[] SEC("license") = "GPL";
+
+#define EI_CLASS	4
+#define ELFCLASS64	2
+#define EM_AARCH64	183
+
+extern int bpf_binprm_set_interp(struct linux_binprm *bprm, const char *path,
+				 size_t path__sz) __ksym;
+
+/*
+ * A magic-style decision needs nothing beyond the prefetched bprm->buf,
+ * even though the match program could read the file.
+ */
+SEC("struct_ops.s/match")
+bool BPF_PROG(bpf_interp_match, struct linux_binprm *bprm)
+{
+	__u16 machine;
+
+	if (bprm->buf[0] != 0x7f || bprm->buf[1] != 'E' ||
+	    bprm->buf[2] != 'L' || bprm->buf[3] != 'F' ||
+	    bprm->buf[EI_CLASS] != ELFCLASS64)
+		return false;
+
+	/* e_machine is a 16-bit little-endian field at offset 18. */
+	machine = (__u8)bprm->buf[18] | ((__u16)(__u8)bprm->buf[19] << 8);
+	return machine == EM_AARCH64;
+}
+
+SEC("struct_ops.s/load")
+int BPF_PROG(bpf_interp_load, struct linux_binprm *bprm)
+{
+	/*
+	 * Keep the path on the (writable) stack: bpf_binprm_set_interp() takes
+	 * a sized memory arg and the verifier rejects a read-only .rodata
+	 * buffer for it. The harness installs the interpreter at this path.
+	 */
+	char interp[] = "/tmp/binfmt_bpf_interp";
+
+	/* @path__sz includes the terminating NUL; 0 commits the selection. */
+	return bpf_binprm_set_interp(bprm, interp, sizeof(interp));
+}
+
+SEC(".struct_ops.link")
+struct binfmt_misc_ops bpf_interp = {
+	.match = (void *)bpf_interp_match,
+	.load = (void *)bpf_interp_load,
+	.name = "bpf_interp",
+};
diff --git a/tools/testing/selftests/exec/nix_origin.bpf.c b/tools/testing/selftests/exec/nix_origin.bpf.c
new file mode 100644
index 000000000000..378e22a4c43b
--- /dev/null
+++ b/tools/testing/selftests/exec/nix_origin.bpf.c
@@ -0,0 +1,224 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * nix_origin.bpf.c - $ORIGIN-relative PT_INTERP resolution
+ *
+ * A binfmt_misc_ops handler that makes relocatable (Nix-style) ELF
+ * binaries work: if PT_INTERP starts with "$ORIGIN/", the loader is
+ * resolved relative to the directory of the binary being executed and
+ * selected via bpf_binprm_set_interp(). The match program reads the
+ * program headers itself, so anything else never commits to this
+ * handler and passes through untouched.
+ *
+ * Activate with:
+ *   bpftool struct_ops register nix_origin.bpf.o /sys/fs/bpf
+ *   echo ':nix-origin:B::::nix_origin:' > /proc/sys/fs/binfmt_misc/register
+ */
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char _license[] SEC("license") = "GPL";
+
+#define PATH_MAX	4096
+#define EI_CLASS	4
+#define ELFCLASSXX	2	/* ELFCLASS64; flip to 1 for 32-bit */
+#define PT_INTERP	3
+#define MAX_PHDRS	64
+
+#define ORIGIN		"$ORIGIN"
+#define ORIGIN_LEN	(sizeof(ORIGIN) - 1)
+
+#define ENOENT		2
+#define ENOEXEC		8
+#define ENAMETOOLONG	36
+
+extern int bpf_dynptr_from_file(struct file *file, __u32 flags,
+				struct bpf_dynptr *ptr__uninit) __ksym;
+extern int bpf_dynptr_file_discard(struct bpf_dynptr *dynptr) __ksym;
+extern int bpf_path_d_path(const struct path *path, char *buf,
+			   size_t buf__sz) __ksym;
+extern int bpf_binprm_set_interp(struct linux_binprm *bprm, const char *path,
+				 size_t path__sz) __ksym;
+
+struct scratch {
+	char interp[PATH_MAX];	/* PT_INTERP as embedded in the binary */
+	char path[PATH_MAX];	/* d_path of the binary, becomes the result */
+};
+
+/* Keyed by pid: execs run concurrently and the programs can sleep. */
+struct {
+	__uint(type, BPF_MAP_TYPE_HASH);
+	__uint(max_entries, 512);
+	__type(key, __u64);
+	__type(value, struct scratch);
+} scratch_map SEC(".maps");
+
+static const struct scratch zero_scratch;
+
+/* An ELF64 binary per the prefetched header? */
+static bool is_elf64(struct linux_binprm *bprm)
+{
+	return bprm->buf[0] == 0x7f && bprm->buf[1] == 'E' &&
+	       bprm->buf[2] == 'L' && bprm->buf[3] == 'F' &&
+	       bprm->buf[EI_CLASS] == ELFCLASSXX;
+}
+
+/* Locate PT_INTERP; false if the file has none or looks malformed. */
+static bool find_pt_interp(struct bpf_dynptr *dp, struct elf64_phdr *phdr)
+{
+	struct elf64_hdr ehdr;
+	bool found = false;
+	int i;
+
+	if (bpf_dynptr_read(&ehdr, sizeof(ehdr), dp, 0, 0))
+		return false;
+	if (ehdr.e_phentsize != sizeof(struct elf64_phdr))
+		return false;
+
+	bpf_for(i, 0, ehdr.e_phnum) {
+		if (i >= MAX_PHDRS)
+			break;
+		if (bpf_dynptr_read(phdr, sizeof(*phdr), dp,
+				    ehdr.e_phoff + i * sizeof(*phdr), 0))
+			return false;
+		if (phdr->p_type == PT_INTERP) {
+			found = true;
+			break;
+		}
+	}
+	return found;
+}
+
+/*
+ * An ELF64 binary whose PT_INTERP starts with "$ORIGIN/" is ours. The
+ * match can sleep and read the file, so the decision is made here and
+ * regular binaries never commit to this handler: later binfmt_misc
+ * entries and binfmt_elf see them as if we did not exist.
+ */
+SEC("struct_ops.s/match")
+bool BPF_PROG(nix_origin_match, struct linux_binprm *bprm)
+{
+	char prefix[ORIGIN_LEN + 1] = {};
+	struct elf64_phdr phdr;
+	struct bpf_dynptr dp;
+	bool ours = false;
+
+	if (!is_elf64(bprm))
+		return false;
+
+	/* The dynptr must be discarded on every path once requested. */
+	if (bpf_dynptr_from_file(bprm->file, 0, &dp))
+		goto out;
+	if (find_pt_interp(&dp, &phdr) &&
+	    phdr.p_filesz > ORIGIN_LEN + 1 &&
+	    !bpf_dynptr_read(prefix, sizeof(prefix), &dp, phdr.p_offset, 0))
+		ours = !bpf_strncmp(prefix, sizeof(prefix), ORIGIN "/");
+out:
+	bpf_dynptr_file_discard(&dp);
+	return ours;
+}
+
+/*
+ * The match is committed and already vetted the "$ORIGIN/" prefix, so
+ * everything here reads the file again from scratch: -ENOEXEC only
+ * covers a binary that changed under us and stopped being ours.
+ */
+SEC("struct_ops.s/load")
+int BPF_PROG(nix_origin_load, struct linux_binprm *bprm)
+{
+	__u32 isz, sfx, rsz, slash;
+	struct elf64_phdr phdr;
+	struct bpf_dynptr dp;
+	struct scratch *sc;
+	__u64 id;
+	int ret = -ENOEXEC, len, i;
+
+	if (bpf_dynptr_from_file(bprm->file, 0, &dp))
+		goto out;
+
+	if (!find_pt_interp(&dp, &phdr))
+		goto out;
+
+	isz = phdr.p_filesz;
+	if (isz <= ORIGIN_LEN + 1 || isz >= sizeof(sc->interp))
+		goto out;
+	/*
+	 * The range check above compiles to a test on a zero-extended copy of
+	 * the u64 p_filesz, so the verifier does not carry the bound to the
+	 * dynptr_read() length below ("unbounded memory access"). Mask isz to
+	 * the buffer size (a power of two) and force the masked value to be
+	 * materialized with a barrier so the read uses the bounded register.
+	 */
+	isz &= sizeof(sc->interp) - 1;
+	barrier_var(isz);
+
+	id = bpf_get_current_pid_tgid();
+	if (bpf_map_update_elem(&scratch_map, &id, &zero_scratch, BPF_ANY))
+		goto out;
+	sc = bpf_map_lookup_elem(&scratch_map, &id);
+	if (!sc)
+		goto out_del;
+
+	if (bpf_dynptr_read(sc->interp, isz, &dp, phdr.p_offset, 0))
+		goto out_del;
+	if (sc->interp[isz - 1] != '\0')
+		goto out_del;
+
+	/* Not "$ORIGIN/..." anymore? Then it is not ours anymore either. */
+	if (sc->interp[0] != '$' || sc->interp[1] != 'O' ||
+	    sc->interp[2] != 'R' || sc->interp[3] != 'I' ||
+	    sc->interp[4] != 'G' || sc->interp[5] != 'I' ||
+	    sc->interp[6] != 'N' || sc->interp[7] != '/')
+		goto out_del;
+
+	/*
+	 * From here on resolution failures fail the exec instead of falling
+	 * back to binfmt_elf, which would resolve the literal "$ORIGIN/..."
+	 * relative to the caller's cwd.
+	 */
+	ret = -ENOENT;
+	len = bpf_path_d_path(&bprm->file->f_path, sc->path, sizeof(sc->path));
+	if (len <= 0 || len > sizeof(sc->path))
+		goto out_del;
+	/* Unreachable or unlinked ("... (deleted)") binaries can't resolve. */
+	if (sc->path[0] != '/')
+		goto out_del;
+
+	/* $ORIGIN = dirname of the binary. */
+	slash = 0;
+	bpf_for(i, 1, len - 1) {
+		if (i >= sizeof(sc->path))
+			break;
+		if (sc->path[i] == '/')
+			slash = i;
+	}
+
+	/* Splice the suffix (leading '/' and NUL included) onto the dir. */
+	sfx = isz - ORIGIN_LEN;
+	rsz = slash + sfx;
+	if (rsz > sizeof(sc->path)) {
+		ret = -ENAMETOOLONG;
+		goto out_del;
+	}
+	bpf_for(i, 0, sfx) {
+		__u32 s = ORIGIN_LEN + i, d = slash + i;
+
+		if (s >= sizeof(sc->interp) || d >= sizeof(sc->path))
+			break;
+		sc->path[d] = sc->interp[s];
+	}
+
+	ret = bpf_binprm_set_interp(bprm, sc->path, rsz);
+out_del:
+	bpf_map_delete_elem(&scratch_map, &id);
+out:
+	bpf_dynptr_file_discard(&dp);
+	return ret;
+}
+
+SEC(".struct_ops.link")
+struct binfmt_misc_ops nix_origin = {
+	.match = (void *)nix_origin_match,
+	.load = (void *)nix_origin_load,
+	.name = "nix_origin",
+};