bpf: Add kernel module with user mode driver that populates bpffs.

Add kernel module with user mode driver that populates bpffs with
BPF iterators.

$ mount bpffs /my/bpffs/ -t bpf
$ ls -la /my/bpffs/
total 4
drwxrwxrwt  2 root root    0 Jul  2 00:27 .
drwxr-xr-x 19 root root 4096 Jul  2 00:09 ..
-rw-------  1 root root    0 Jul  2 00:27 maps.debug
-rw-------  1 root root    0 Jul  2 00:27 progs.debug

The user mode driver will load BPF Type Formats, create BPF maps, populate BPF
maps, load two BPF programs, attach them to BPF iterators, and finally send two
bpf_link IDs back to the kernel.
The kernel will pin two bpf_links into newly mounted bpffs instance under
names "progs.debug" and "maps.debug". These two files become human readable.

$ cat /my/bpffs/progs.debug
  id name            attached
  11 dump_bpf_map    bpf_iter_bpf_map
  12 dump_bpf_prog   bpf_iter_bpf_prog
  27 test_pkt_access
  32 test_main       test_pkt_access test_pkt_access
  33 test_subprog1   test_pkt_access_subprog1 test_pkt_access
  34 test_subprog2   test_pkt_access_subprog2 test_pkt_access
  35 test_subprog3   test_pkt_access_subprog3 test_pkt_access
  36 new_get_skb_len get_skb_len test_pkt_access
  37 new_get_skb_ifindex get_skb_ifindex test_pkt_access
  38 new_get_constant get_constant test_pkt_access

The BPF program dump_bpf_prog() in iterators.bpf.c is printing this data about
all BPF programs currently loaded in the system. This information is unstable
and will change from kernel to kernel as ".debug" suffix conveys.

Signed-off-by: Alexei Starovoitov <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]

init/Kconfig

@@ -1710,6 +1710,8 @@ config BPF_JIT_DEFAULT_ON
 	def_bool ARCH_WANT_DEFAULT_BPF_JIT || BPF_JIT_ALWAYS_ON
 	depends on HAVE_EBPF_JIT && BPF_JIT
 
+source "kernel/bpf/preload/Kconfig"
+
 config USERFAULTFD
 	bool "Enable userfaultfd() system call"
 	depends on MMU

kernel/Makefile

@@ -12,7 +12,7 @@ obj-y     = fork.o exec_domain.o panic.o \
 	    notifier.o ksysfs.o cred.o reboot.o \
 	    async.o range.o smpboot.o ucount.o regset.o
 
-obj-$(CONFIG_BPFILTER) += usermode_driver.o
+obj-$(CONFIG_USERMODE_DRIVER) += usermode_driver.o
 obj-$(CONFIG_MODULES) += kmod.o
 obj-$(CONFIG_MULTIUSER) += groups.o
 

kernel/bpf/Makefile

@@ -29,3 +29,4 @@ ifeq ($(CONFIG_BPF_JIT),y)
 obj-$(CONFIG_BPF_SYSCALL) += bpf_struct_ops.o
 obj-${CONFIG_BPF_LSM} += bpf_lsm.o
 endif
+obj-$(CONFIG_BPF_PRELOAD) += preload/

kernel/bpf/inode.c

@@ -20,6 +20,7 @@
 #include <linux/filter.h>
 #include <linux/bpf.h>
 #include <linux/bpf_trace.h>
+#include "preload/bpf_preload.h"
 
 enum bpf_type {
 	BPF_TYPE_UNSPEC	= 0,
@@ -369,9 +370,10 @@ static struct dentry *
 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
 {
 	/* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
-	 * extensions.
+	 * extensions. That allows popoulate_bpffs() create special files.
 	 */
-	if (strchr(dentry->d_name.name, '.'))
+	if ((dir->i_mode & S_IALLUGO) &&
+	    strchr(dentry->d_name.name, '.'))
 		return ERR_PTR(-EPERM);
 
 	return simple_lookup(dir, dentry, flags);
@@ -409,6 +411,27 @@ static const struct inode_operations bpf_dir_iops = {
 	.unlink		= simple_unlink,
 };
 
+/* pin iterator link into bpffs */
+static int bpf_iter_link_pin_kernel(struct dentry *parent,
+				    const char *name, struct bpf_link *link)
+{
+	umode_t mode = S_IFREG | S_IRUSR;
+	struct dentry *dentry;
+	int ret;
+
+	inode_lock(parent->d_inode);
+	dentry = lookup_one_len(name, parent, strlen(name));
+	if (IS_ERR(dentry)) {
+		inode_unlock(parent->d_inode);
+		return PTR_ERR(dentry);
+	}
+	ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
+			    &bpf_iter_fops);
+	dput(dentry);
+	inode_unlock(parent->d_inode);
+	return ret;
+}
+
 static int bpf_obj_do_pin(const char __user *pathname, void *raw,
 			  enum bpf_type type)
 {
@@ -638,6 +661,91 @@ static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
 	return 0;
 }
 
+struct bpf_preload_ops *bpf_preload_ops;
+EXPORT_SYMBOL_GPL(bpf_preload_ops);
+
+static bool bpf_preload_mod_get(void)
+{
+	/* If bpf_preload.ko wasn't loaded earlier then load it now.
+	 * When bpf_preload is built into vmlinux the module's __init
+	 * function will populate it.
+	 */
+	if (!bpf_preload_ops) {
+		request_module("bpf_preload");
+		if (!bpf_preload_ops)
+			return false;
+	}
+	/* And grab the reference, so the module doesn't disappear while the
+	 * kernel is interacting with the kernel module and its UMD.
+	 */
+	if (!try_module_get(bpf_preload_ops->owner)) {
+		pr_err("bpf_preload module get failed.\n");
+		return false;
+	}
+	return true;
+}
+
+static void bpf_preload_mod_put(void)
+{
+	if (bpf_preload_ops)
+		/* now user can "rmmod bpf_preload" if necessary */
+		module_put(bpf_preload_ops->owner);
+}
+
+static DEFINE_MUTEX(bpf_preload_lock);
+
+static int populate_bpffs(struct dentry *parent)
+{
+	struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
+	struct bpf_link *links[BPF_PRELOAD_LINKS] = {};
+	int err = 0, i;
+
+	/* grab the mutex to make sure the kernel interactions with bpf_preload
+	 * UMD are serialized
+	 */
+	mutex_lock(&bpf_preload_lock);
+
+	/* if bpf_preload.ko wasn't built into vmlinux then load it */
+	if (!bpf_preload_mod_get())
+		goto out;
+
+	if (!bpf_preload_ops->info.tgid) {
+		/* preload() will start UMD that will load BPF iterator programs */
+		err = bpf_preload_ops->preload(objs);
+		if (err)
+			goto out_put;
+		for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
+			links[i] = bpf_link_by_id(objs[i].link_id);
+			if (IS_ERR(links[i])) {
+				err = PTR_ERR(links[i]);
+				goto out_put;
+			}
+		}
+		for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
+			err = bpf_iter_link_pin_kernel(parent,
+						       objs[i].link_name, links[i]);
+			if (err)
+				goto out_put;
+			/* do not unlink successfully pinned links even
+			 * if later link fails to pin
+			 */
+			links[i] = NULL;
+		}
+		/* finish() will tell UMD process to exit */
+		err = bpf_preload_ops->finish();
+		if (err)
+			goto out_put;
+	}
+out_put:
+	bpf_preload_mod_put();
+out:
+	mutex_unlock(&bpf_preload_lock);
+	for (i = 0; i < BPF_PRELOAD_LINKS && err; i++)
+		if (!IS_ERR_OR_NULL(links[i]))
+			bpf_link_put(links[i]);
+	return err;
+}
+
 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
 {
 	static const struct tree_descr bpf_rfiles[] = { { "" } };
@@ -654,8 +762,8 @@ static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
 	inode = sb->s_root->d_inode;
 	inode->i_op = &bpf_dir_iops;
 	inode->i_mode &= ~S_IALLUGO;
+	populate_bpffs(sb->s_root);
 	inode->i_mode |= S_ISVTX | opts->mode;
-
 	return 0;
 }
 
@@ -705,6 +813,8 @@ static int __init bpf_init(void)
 {
 	int ret;
 
+	mutex_init(&bpf_preload_lock);
+
 	ret = sysfs_create_mount_point(fs_kobj, "bpf");
 	if (ret)
 		return ret;

kernel/bpf/preload/Kconfig

@@ -0,0 +1,23 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config USERMODE_DRIVER
+	bool
+	default n
+
+menuconfig BPF_PRELOAD
+	bool "Preload BPF file system with kernel specific program and map iterators"
+	depends on BPF
+	select USERMODE_DRIVER
+	help
+	  This builds kernel module with several embedded BPF programs that are
+	  pinned into BPF FS mount point as human readable files that are
+	  useful in debugging and introspection of BPF programs and maps.
+
+if BPF_PRELOAD
+config BPF_PRELOAD_UMD
+	tristate "bpf_preload kernel module with user mode driver"
+	depends on CC_CAN_LINK
+	depends on m || CC_CAN_LINK_STATIC
+	default m
+	help
+	  This builds bpf_preload kernel module with embedded user mode driver.
+endif

kernel/bpf/preload/Makefile

@@ -0,0 +1,23 @@
+# SPDX-License-Identifier: GPL-2.0
+
+LIBBPF_SRCS = $(srctree)/tools/lib/bpf/
+LIBBPF_A = $(obj)/libbpf.a
+LIBBPF_OUT = $(abspath $(obj))
+
+$(LIBBPF_A):
+	$(Q)$(MAKE) -C $(LIBBPF_SRCS) OUTPUT=$(LIBBPF_OUT)/ $(LIBBPF_OUT)/libbpf.a
+
+userccflags += -I $(srctree)/tools/include/ -I $(srctree)/tools/include/uapi \
+	-I $(srctree)/tools/lib/ -Wno-unused-result
+
+userprogs := bpf_preload_umd
+
+bpf_preload_umd-objs := iterators/iterators.o
+bpf_preload_umd-userldlibs := $(LIBBPF_A) -lelf -lz
+
+$(obj)/bpf_preload_umd: $(LIBBPF_A)
+
+$(obj)/bpf_preload_umd_blob.o: $(obj)/bpf_preload_umd
+
+obj-$(CONFIG_BPF_PRELOAD_UMD) += bpf_preload.o
+bpf_preload-objs += bpf_preload_kern.o bpf_preload_umd_blob.o

kernel/bpf/preload/bpf_preload.h

@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BPF_PRELOAD_H
+#define _BPF_PRELOAD_H
+
+#include <linux/usermode_driver.h>
+#include "iterators/bpf_preload_common.h"
+
+struct bpf_preload_ops {
+        struct umd_info info;
+	int (*preload)(struct bpf_preload_info *);
+	int (*finish)(void);
+	struct module *owner;
+};
+extern struct bpf_preload_ops *bpf_preload_ops;
+#define BPF_PRELOAD_LINKS 2
+#endif

kernel/bpf/preload/bpf_preload_kern.c

@@ -0,0 +1,91 @@
+// SPDX-License-Identifier: GPL-2.0
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/pid.h>
+#include <linux/fs.h>
+#include <linux/sched/signal.h>
+#include "bpf_preload.h"
+
+extern char bpf_preload_umd_start;
+extern char bpf_preload_umd_end;
+
+static int preload(struct bpf_preload_info *obj);
+static int finish(void);
+
+static struct bpf_preload_ops umd_ops = {
+	.info.driver_name = "bpf_preload",
+	.preload = preload,
+	.finish = finish,
+	.owner = THIS_MODULE,
+};
+
+static int preload(struct bpf_preload_info *obj)
+{
+	int magic = BPF_PRELOAD_START;
+	loff_t pos = 0;
+	int i, err;
+	ssize_t n;
+
+	err = fork_usermode_driver(&umd_ops.info);
+	if (err)
+		return err;
+
+	/* send the start magic to let UMD proceed with loading BPF progs */
+	n = kernel_write(umd_ops.info.pipe_to_umh,
+			 &magic, sizeof(magic), &pos);
+	if (n != sizeof(magic))
+		return -EPIPE;
+
+	/* receive bpf_link IDs and names from UMD */
+	pos = 0;
+	for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
+		n = kernel_read(umd_ops.info.pipe_from_umh,
+				&obj[i], sizeof(*obj), &pos);
+		if (n != sizeof(*obj))
+			return -EPIPE;
+	}
+	return 0;
+}
+
+static int finish(void)
+{
+	int magic = BPF_PRELOAD_END;
+	struct pid *tgid;
+	loff_t pos = 0;
+	ssize_t n;
+
+	/* send the last magic to UMD. It will do a normal exit. */
+	n = kernel_write(umd_ops.info.pipe_to_umh,
+			 &magic, sizeof(magic), &pos);
+	if (n != sizeof(magic))
+		return -EPIPE;
+	tgid = umd_ops.info.tgid;
+	wait_event(tgid->wait_pidfd, thread_group_exited(tgid));
+	umd_ops.info.tgid = NULL;
+	return 0;
+}
+
+static int __init load_umd(void)
+{
+	int err;
+
+	err = umd_load_blob(&umd_ops.info, &bpf_preload_umd_start,
+			    &bpf_preload_umd_end - &bpf_preload_umd_start);
+	if (err)
+		return err;
+	bpf_preload_ops = &umd_ops;
+	return err;
+}
+
+static void __exit fini_umd(void)
+{
+	bpf_preload_ops = NULL;
+	/* kill UMD in case it's still there due to earlier error */
+	kill_pid(umd_ops.info.tgid, SIGKILL, 1);
+	umd_ops.info.tgid = NULL;
+	umd_unload_blob(&umd_ops.info);
+}
+late_initcall(load_umd);
+module_exit(fini_umd);
+MODULE_LICENSE("GPL");

kernel/bpf/preload/bpf_preload_umd_blob.S

@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+	.section .init.rodata, "a"
+	.global bpf_preload_umd_start
+bpf_preload_umd_start:
+	.incbin "kernel/bpf/preload/bpf_preload_umd"
+	.global bpf_preload_umd_end
+bpf_preload_umd_end:

kernel/bpf/preload/iterators/bpf_preload_common.h

@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BPF_PRELOAD_COMMON_H
+#define _BPF_PRELOAD_COMMON_H
+
+#define BPF_PRELOAD_START 0x5555
+#define BPF_PRELOAD_END 0xAAAA
+
+struct bpf_preload_info {
+	char link_name[16];
+	int link_id;
+};
+
+#endif