niceSOFT/systemd: The systemd System and Service Manager

src/shared/bpf-program.c
581 строка20 KB
Yu Watanabe
tree-wide: normalize comment style
15 дек 2025, 02:50
15 дек 2025, 02:509337814
/* SPDX-License-Identifier: LGPL-2.1-or-later */

#include <fcntl.h>
#include <linux/bpf.h>
#include <linux/bpf_insn.h>
#include <sys/bpf.h>
#include <unistd.h>

#include "alloc-util.h"
#include "bpf-program.h"
#include "errno-util.h"
#include "escape.h"
#include "extract-word.h"
#include "fd-util.h"
#include "fdset.h"
#include "log.h"
#include "memory-util.h"
#include "parse-util.h"
#include "path-util.h"
#include "serialize.h"
#include "set.h"
#include "string-table.h"
#include "string-util.h"

static const char *const bpf_cgroup_attach_type_table[__MAX_BPF_ATTACH_TYPE] = {
        [BPF_CGROUP_INET_INGRESS] =     "ingress",
        [BPF_CGROUP_INET_EGRESS] =      "egress",
        [BPF_CGROUP_INET_SOCK_CREATE] = "sock_create",
        [BPF_CGROUP_SOCK_OPS] =         "sock_ops",
        [BPF_CGROUP_DEVICE] =           "device",
        [BPF_CGROUP_INET4_BIND] =       "bind4",
        [BPF_CGROUP_INET6_BIND] =       "bind6",
        [BPF_CGROUP_INET4_CONNECT] =    "connect4",
        [BPF_CGROUP_INET6_CONNECT] =    "connect6",
        [BPF_CGROUP_INET4_POST_BIND] =  "post_bind4",
        [BPF_CGROUP_INET6_POST_BIND] =  "post_bind6",
        [BPF_CGROUP_UDP4_SENDMSG] =     "sendmsg4",
        [BPF_CGROUP_UDP6_SENDMSG] =     "sendmsg6",
        [BPF_CGROUP_SYSCTL] =           "sysctl",
        [BPF_CGROUP_UDP4_RECVMSG] =     "recvmsg4",
        [BPF_CGROUP_UDP6_RECVMSG] =     "recvmsg6",
        [BPF_CGROUP_GETSOCKOPT] =       "getsockopt",
        [BPF_CGROUP_SETSOCKOPT] =       "setsockopt",
};

DEFINE_STRING_TABLE_LOOKUP(bpf_cgroup_attach_type, int);

DEFINE_HASH_OPS_WITH_KEY_DESTRUCTOR(bpf_program_hash_ops, void, trivial_hash_func, trivial_compare_func, bpf_program_free);

int bpf_program_supported(void) {
        static int cached = 0;
        int r;

        if (cached != 0)
                return cached;

        /* Currently, we only use the following three types:
         * - BPF_PROG_TYPE_CGROUP_SKB, supported since kernel v4.10 (0e33661de493db325435d565a4a722120ae4cbf3),
         * - BPF_PROG_TYPE_CGROUP_DEVICE, supported since kernel v4.15 (ebc614f687369f9df99828572b1d85a7c2de3d92),
         * - BPF_PROG_TYPE_CGROUP_SOCK_ADDR, supported since kernel v4.17 (4fbac77d2d092b475dda9eea66da674369665427).
         * As our baseline on the kernel is v5.4, it is enough to check if one BPF program can be created and loaded. */

        _cleanup_(bpf_program_freep) BPFProgram *program = NULL;
        r = bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, /* prog_name= */ NULL, &program);
        if (r < 0)
                return cached = log_debug_errno(r, "Can't allocate CGROUP SKB BPF program, assuming BPF is not supported: %m");

        static const struct bpf_insn trivial[] = {
                BPF_MOV64_IMM(BPF_REG_0, 1),
                BPF_EXIT_INSN()
        };
        r = bpf_program_add_instructions(program, trivial, ELEMENTSOF(trivial));
        if (r < 0)
                return cached = log_debug_errno(r, "Can't add trivial instructions to CGROUP SKB BPF program, assuming BPF is not supported: %m");

        r = bpf_program_load_kernel(program, /* log_buf= */ NULL, /* log_size= */ 0);
        if (r < 0)
                return cached = log_debug_errno(r, "Can't load kernel CGROUP SKB BPF program, assuming BPF is not supported: %m");

        /* Unfortunately the kernel allows us to create BPF_PROG_TYPE_CGROUP_SKB (maybe also other types)
         * programs even when CONFIG_CGROUP_BPF is turned off at kernel compilation time. This sucks of course:
         * why does it allow us to create a cgroup BPF program if we can't do a thing with it later?
         *
         * We detect this case by issuing the BPF_PROG_DETACH bpf() call with invalid file descriptors: if
         * CONFIG_CGROUP_BPF is turned off, then the call will fail early with EINVAL. If it is turned on the
         * parameters are validated however, and that'll fail with EBADF then.
         *
         * The check seems also important when we are running with sanitizers. With sanitizers (at least with
         * LLVM v20), the following check and other bpf() calls fails even if the kernel supports BPF. To
         * avoid unexpected fail when running with sanitizers, let's explicitly check if bpf() syscall works. */

        /* Clang and GCC (>=15) do not 0-pad with structured initialization, causing the kernel to reject the
         * bpf_attr as invalid. See: https://github.com/torvalds/linux/blob/v5.9/kernel/bpf/syscall.c#L65
         * Hence, we cannot use structured initialization here, and need to clear the structure with zero
         * explicitly before use. */
        union bpf_attr attr;
        zero(attr);
        attr.attach_type = BPF_CGROUP_INET_EGRESS; /* since kernel v4.10 (0e33661de493db325435d565a4a722120ae4cbf3) */
        attr.target_fd = -EBADF;
        attr.attach_bpf_fd = -EBADF;

        if (bpf(BPF_PROG_DETACH, &attr, sizeof(attr)) < 0) {
                if (errno == EBADF) /* YAY! */
                        return cached = true;

                return cached = log_debug_errno(errno, "Didn't get EBADF from invalid BPF_PROG_DETACH call: %m");
        }

        return cached = log_debug_errno(SYNTHETIC_ERRNO(EBADE),
                                        "Wut? Kernel accepted our invalid BPF_PROG_DETACH call? Something is weird, assuming BPF is broken and hence not supported.");
}

BPFProgram *bpf_program_free(BPFProgram *p) {
        if (!p)
                return NULL;
        /* Unfortunately, the kernel currently doesn't implicitly detach BPF programs from their cgroups when the last
         * fd to the BPF program is closed. This has nasty side-effects since this means that abnormally terminated
         * programs that attached one of their BPF programs to a cgroup will leave this program pinned for good with
         * zero chance of recovery, until the cgroup is removed. This is particularly problematic if the cgroup in
         * question is the root cgroup (or any other cgroup belonging to a service that cannot be restarted during
         * operation, such as dbus), as the memory for the BPF program can only be reclaimed through a reboot. To
         * counter this, we track closely to which cgroup a program was attached to and will detach it on our own
         * whenever we close the BPF fd. */
        (void) bpf_program_cgroup_detach(p);

        safe_close(p->kernel_fd);
        free(p->prog_name);
        free(p->instructions);
        free(p->attached_path);

        return mfree(p);
}

 /* struct bpf_prog_info info must be initialized since its value is both input and output
  * for BPF_OBJ_GET_INFO_BY_FD syscall. */
static int bpf_program_get_info_by_fd(int prog_fd, struct bpf_prog_info *info, uint32_t info_len) {
        union bpf_attr attr;

        /* Explicitly memset to zero since some compilers may produce non-zero-initialized padding when
         * structured initialization is used.
         * Refer to https://github.com/systemd/systemd/issues/18164
         */
        zero(attr);
        attr.info.bpf_fd = prog_fd;
        attr.info.info_len = info_len;
        attr.info.info = PTR_TO_UINT64(info);

        return RET_NERRNO(bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)));
}

int bpf_program_new(uint32_t prog_type, const char *prog_name, BPFProgram **ret) {
        _cleanup_(bpf_program_freep) BPFProgram *p = NULL;
        _cleanup_free_ char *name = NULL;

        if (prog_name) {
                if (strlen(prog_name) >= BPF_OBJ_NAME_LEN)
                        return -ENAMETOOLONG;

                name = strdup(prog_name);
                if (!name)
                        return -ENOMEM;
        }

        p = new(BPFProgram, 1);
        if (!p)
                return -ENOMEM;

        *p = (BPFProgram) {
                .prog_type = prog_type,
                .kernel_fd = -EBADF,
                .prog_name = TAKE_PTR(name),
        };

        *ret = TAKE_PTR(p);

        return 0;
}

int bpf_program_new_from_bpffs_path(const char *path, BPFProgram **ret) {
        _cleanup_(bpf_program_freep) BPFProgram *p = NULL;
        struct bpf_prog_info info = {};
        int r;

        assert(path);
        assert(ret);

        p = new(BPFProgram, 1);
        if (!p)
                return -ENOMEM;

        *p = (BPFProgram) {
                .prog_type = BPF_PROG_TYPE_UNSPEC,
                .kernel_fd = -EBADF,
        };

        r = bpf_program_load_from_bpf_fs(p, path);
        if (r < 0)
                return r;

        r = bpf_program_get_info_by_fd(p->kernel_fd, &info, sizeof(info));
        if (r < 0)
                return r;

        p->prog_type = info.type;
        *ret = TAKE_PTR(p);

        return 0;
}

int bpf_program_add_instructions(BPFProgram *p, const struct bpf_insn *instructions, size_t count) {

        assert(p);

        if (p->kernel_fd >= 0) /* don't allow modification after we uploaded things to the kernel */
                return -EBUSY;

        if (!GREEDY_REALLOC(p->instructions, p->n_instructions + count))
                return -ENOMEM;

        memcpy(p->instructions + p->n_instructions, instructions, sizeof(struct bpf_insn) * count);
        p->n_instructions += count;

        return 0;
}

int bpf_program_load_kernel(BPFProgram *p, char *log_buf, size_t log_size) {
        union bpf_attr attr;

        assert(p);

        if (p->kernel_fd >= 0) { /* make this idempotent */
                memzero(log_buf, log_size);
                return 0;
        }

        // FIXME: Clang doesn't 0-pad with structured initialization, causing
        // the kernel to reject the bpf_attr as invalid. See:
        // https://github.com/torvalds/linux/blob/v5.9/kernel/bpf/syscall.c#L65
        // Ideally it should behave like GCC, so that we can remove these workarounds.
        zero(attr);
        attr.prog_type = p->prog_type;
        attr.insns = PTR_TO_UINT64(p->instructions);
        attr.insn_cnt = p->n_instructions;
        attr.license = PTR_TO_UINT64("GPL");
        attr.log_buf = PTR_TO_UINT64(log_buf);
        attr.log_level = !!log_buf;
        attr.log_size = log_size;
        if (p->prog_name)
                strncpy(attr.prog_name, p->prog_name, BPF_OBJ_NAME_LEN - 1);

        p->kernel_fd = bpf(BPF_PROG_LOAD, &attr, sizeof(attr));
        if (p->kernel_fd < 0)
                return -errno;

        return 0;
}

int bpf_program_load_from_bpf_fs(BPFProgram *p, const char *path) {
        union bpf_attr attr;

        assert(p);

        if (p->kernel_fd >= 0) /* don't overwrite an assembled or loaded program */
                return -EBUSY;

        zero(attr);
        attr.pathname = PTR_TO_UINT64(path);

        p->kernel_fd = bpf(BPF_OBJ_GET, &attr, sizeof(attr));
        if (p->kernel_fd < 0)
                return -errno;

        return 0;
}

int bpf_program_cgroup_attach(BPFProgram *p, int type, const char *path, uint32_t flags) {
        _cleanup_free_ char *copy = NULL;
        _cleanup_close_ int fd = -EBADF;
        union bpf_attr attr;
        int r;

        assert(p);
        assert(type >= 0);
        assert(path);

        if (!IN_SET(flags, 0, BPF_F_ALLOW_OVERRIDE, BPF_F_ALLOW_MULTI))
                return -EINVAL;

        /* We need to track which cgroup the program is attached to, and we can only track one attachment, hence let's
        * refuse this early. */
        if (p->attached_path) {
                if (!path_equal(p->attached_path, path))
                        return -EBUSY;
                if (p->attached_type != type)
                        return -EBUSY;
                if (p->attached_flags != flags)
                        return -EBUSY;

                /* Here's a shortcut: if we previously attached this program already, then we don't have to do so
                 * again. Well, with one exception: if we are in BPF_F_ALLOW_OVERRIDE mode then someone else might have
                 * replaced our program since the last time, hence let's reattach it again, just to be safe. In flags
                 * == 0 mode this is not an issue since nobody else can replace our program in that case, and in flags
                 * == BPF_F_ALLOW_MULTI mode any other's program would be installed in addition to ours hence ours
                 * would remain in effect. */
                if (flags != BPF_F_ALLOW_OVERRIDE)
                        return 0;
        }

        /* Ensure we have a kernel object for this. */
        r = bpf_program_load_kernel(p, NULL, 0);
        if (r < 0)
                return r;

        copy = strdup(path);
        if (!copy)
                return -ENOMEM;

        fd = open(path, O_DIRECTORY|O_RDONLY|O_CLOEXEC);
        if (fd < 0)
                return -errno;

        zero(attr);
        attr.attach_type = type;
        attr.target_fd = fd;
        attr.attach_bpf_fd = p->kernel_fd;
        attr.attach_flags = flags;

        if (bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)) < 0)
                return -errno;

        free_and_replace(p->attached_path, copy);
        p->attached_type = type;
        p->attached_flags = flags;

        return 0;
}

int bpf_program_cgroup_detach(BPFProgram *p) {
        _cleanup_close_ int fd = -EBADF;

        assert(p);

        if (!p->attached_path)
                return -EUNATCH;

        fd = open(p->attached_path, O_DIRECTORY|O_RDONLY|O_CLOEXEC);
        if (fd < 0) {
                if (errno != ENOENT)
                        return -errno;

                /* If the cgroup does not exist anymore, then we don't have to explicitly detach, it got detached
                 * implicitly by the removal, hence don't complain */

        } else {
                union bpf_attr attr;

                zero(attr);
                attr.attach_type = p->attached_type;
                attr.target_fd = fd;
                attr.attach_bpf_fd = p->kernel_fd;

                if (bpf(BPF_PROG_DETACH, &attr, sizeof(attr)) < 0)
                        return -errno;
        }

        p->attached_path = mfree(p->attached_path);

        return 0;
}

int bpf_map_new(
                const char *name,
                enum bpf_map_type type,
                size_t key_size,
                size_t value_size,
                size_t max_entries,
                uint32_t flags) {

        union bpf_attr attr;
        const char *n = name;

        zero(attr);
        attr.map_type = type;
        attr.key_size = key_size;
        attr.value_size = value_size;
        attr.max_entries = max_entries;
        attr.map_flags = flags;

        /* The map name is primarily informational for debugging purposes, and typically too short
         * to carry the full unit name, hence we employ a trivial lossy escaping to make it fit
         * (truncation + only alphanumerical, "." and "_" are allowed as per
         * https://docs.kernel.org/bpf/maps.html#usage-notes) */
        for (size_t i = 0; i < sizeof(attr.map_name) - 1 && *n; i++, n++)
                attr.map_name[i] = strchr(ALPHANUMERICAL ".", *n) ? *n : '_';

        return RET_NERRNO(bpf(BPF_MAP_CREATE, &attr, sizeof(attr)));
}

int bpf_map_update_element(int fd, const void *key, void *value) {
        union bpf_attr attr;

        zero(attr);
        attr.map_fd = fd;
        attr.key = PTR_TO_UINT64(key);
        attr.value = PTR_TO_UINT64(value);

        return RET_NERRNO(bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)));
}

int bpf_map_lookup_element(int fd, const void *key, void *value) {
        union bpf_attr attr;

        zero(attr);
        attr.map_fd = fd;
        attr.key = PTR_TO_UINT64(key);
        attr.value = PTR_TO_UINT64(value);

        return RET_NERRNO(bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr)));
}

int bpf_program_pin(int prog_fd, const char *bpffs_path) {
        union bpf_attr attr;

        zero(attr);
        attr.pathname = PTR_TO_UINT64((void *) bpffs_path);
        attr.bpf_fd = prog_fd;

        return RET_NERRNO(bpf(BPF_OBJ_PIN, &attr, sizeof(attr)));
}

int bpf_program_get_id_by_fd(int prog_fd, uint32_t *ret_id) {
        struct bpf_prog_info info = {};
        int r;

        assert(ret_id);

        r = bpf_program_get_info_by_fd(prog_fd, &info, sizeof(info));
        if (r < 0)
                return r;

        *ret_id = info.id;

        return 0;
};

int bpf_program_serialize_attachment(
                FILE *f,
                FDSet *fds,
                const char *key,
                BPFProgram *p) {

        _cleanup_free_ char *escaped = NULL;
        int copy, r;

        if (!p || !p->attached_path)
                return 0;

        assert(p->kernel_fd >= 0);

        escaped = cescape(p->attached_path);
        if (!escaped)
                return -ENOMEM;

        copy = fdset_put_dup(fds, p->kernel_fd);
        if (copy < 0)
                return log_error_errno(copy, "Failed to add BPF kernel fd to serialize: %m");

        r = serialize_item_format(
                        f,
                        key,
                        "%i %s %s",
                        copy,
                        bpf_cgroup_attach_type_to_string(p->attached_type),
                        escaped);
        if (r < 0)
                return r;

        /* After serialization, let's forget the fact that this program is attached. The attachment — if you
         * so will — is now 'owned' by the serialization, and not us anymore. Why does that matter? Because
         * of BPF's less-than-ideal lifecycle handling: to detach a program from a cgroup we have to
         * explicitly do so, it's not done implicitly on close(). Now, since we are serializing here we don't
         * want the program to be detached while freeing things, so that the attachment can be retained after
         * deserializing again. bpf_program_free() implicitly detaches things, if attached_path is non-NULL,
         * hence we set it to NULL here. */

        p->attached_path = mfree(p->attached_path);
        return 0;
}

int bpf_program_serialize_attachment_set(FILE *f, FDSet *fds, const char *key, Set *set) {
        BPFProgram *p;
        int r;

        SET_FOREACH(p, set) {
                r = bpf_program_serialize_attachment(f, fds, key, p);
                if (r < 0)
                        return r;
        }

        return 0;
}

int bpf_program_deserialize_attachment(const char *v, FDSet *fds, BPFProgram **bpfp) {
        _cleanup_free_ char *sfd = NULL, *sat = NULL, *unescaped = NULL;
        _cleanup_(bpf_program_freep) BPFProgram *p = NULL;
        _cleanup_close_ int fd = -EBADF;
        ssize_t l;
        int ifd, at, r;

        assert(v);
        assert(bpfp);

        /* Extract first word: the fd number */
        r = extract_first_word(&v, &sfd, NULL, 0);
        if (r < 0)
                return r;
        if (r == 0)
                return -EINVAL;

        ifd = parse_fd(sfd);
        if (ifd < 0)
                return r;

        /* Extract second word: the attach type */
        r = extract_first_word(&v, &sat, NULL, 0);
        if (r < 0)
                return r;
        if (r == 0)
                return -EINVAL;

        at = bpf_cgroup_attach_type_from_string(sat);
        if (at < 0)
                return at;

        /* The rest is the path */
        if (isempty(v))
                return -EINVAL;

        l = cunescape(v, 0, &unescaped);
        if (l < 0)
                return l;

        fd = fdset_remove(fds, ifd);
        if (fd < 0)
                return fd;

        p = new(BPFProgram, 1);
        if (!p)
                return -ENOMEM;

        *p = (BPFProgram) {
                .kernel_fd = TAKE_FD(fd),
                .prog_type = BPF_PROG_TYPE_UNSPEC,
                .attached_path = TAKE_PTR(unescaped),
                .attached_type = at,
        };

        if (*bpfp)
                bpf_program_free(*bpfp);

        *bpfp = TAKE_PTR(p);
        return 0;
}

int bpf_program_deserialize_attachment_set(const char *v, FDSet *fds, Set **bpfsetp) {
        BPFProgram *p = NULL;
        int r;

        assert(v);
        assert(bpfsetp);

        r = bpf_program_deserialize_attachment(v, fds, &p);
        if (r < 0)
                return r;

        r = set_ensure_consume(bpfsetp, &bpf_program_hash_ops, p);
        if (r < 0)
                return r;

        return 0;
}