bpf(): improve map and program compatibility (microsoft#3980)

* bpf(): improve map and program compatibility

Allow passing a name for maps and programs and allow retrieving it via
BPF_OBJ_GET_INFO_BY_FD. This requires a bunch of glue code because the
bpf_prog_info, etc. structs are not compatible with Linux.

* Fix test failure due to incorrect link type

* Address feedback from Dave Thaler

* Fix second test failure due to wrong link type

* Use program ID 0 on detached link ids

Internally both zero and EBPF_ID_NONE are used to refer to non-existing ID.
Use zero when retrieving link info, since that aligns with Linux better.

* Temporarily use BPF_LINK_TYPE_UNSPEC

* Only test program info if JIT is enabled

* Apply suggestions from dthaler's review

Co-authored-by: Dave Thaler <dthaler1968@gmail.com>

---------

Co-authored-by: Dave Thaler <dthaler1968@gmail.com>

include/ebpf_core_structs.h

@@ -31,3 +31,11 @@ typedef struct _ebpf_ring_buffer_map_async_query_result
     size_t producer;
     size_t consumer;
 } ebpf_ring_buffer_map_async_query_result_t;
+
+typedef enum _ebpf_object_type
+{
+    EBPF_OBJECT_UNKNOWN,
+    EBPF_OBJECT_MAP,
+    EBPF_OBJECT_LINK,
+    EBPF_OBJECT_PROGRAM,
+} ebpf_object_type_t;

include/linux/bpf.h

@@ -54,6 +54,8 @@ enum bpf_stats_type
 // Names do not have to match, but try to keep them the same as much as possible.
 // In case of conflicts prefix them with "sys_" or "SYS_".
 
+#define SYS_BPF_OBJ_NAME_LEN 16U
+
 enum bpf_cmd_id
 {
     BPF_MAP_CREATE,
@@ -88,12 +90,15 @@ enum bpf_cmd_id
 /// Attributes used by BPF_MAP_CREATE.
 typedef struct
 {
-    enum bpf_map_type map_type; ///< Type of map to create.
-    uint32_t key_size;          ///< Size in bytes of keys.
-    uint32_t value_size;        ///< Size in bytes of values.
-    uint32_t max_entries;       ///< Maximum number of entries in the map.
-    uint32_t map_flags;         ///< Not supported, must be zero.
-    uint32_t inner_map_fd;      ///< File descriptor of inner map.
+    enum bpf_map_type map_type;          ///< Type of map to create.
+    uint32_t key_size;                   ///< Size in bytes of keys.
+    uint32_t value_size;                 ///< Size in bytes of values.
+    uint32_t max_entries;                ///< Maximum number of entries in the map.
+    uint32_t map_flags;                  ///< Not supported, must be zero.
+    uint32_t inner_map_fd;               ///< File descriptor of inner map.
+    uint32_t numa_node;                  ///< Not supported, must be zero.
+    char map_name[SYS_BPF_OBJ_NAME_LEN]; ///< Map name.
+    uint32_t map_ifindex;                ///< Not supported, must be zero.
 } sys_bpf_map_create_attr_t;
 
 typedef struct
@@ -137,6 +142,7 @@ typedef struct
     uint64_t log_buf;      ///< Pointer to a buffer in which log info can be written.
     uint32_t kern_version; ///< Kernel version (currently ignored on Windows).
     uint32_t prog_flags;   ///< Not supported, must be zero.
+    char prog_name[SYS_BPF_OBJ_NAME_LEN]; ///< Program name.
 } sys_bpf_prog_load_attr_t;
 
 typedef struct
@@ -186,6 +192,40 @@ typedef struct
     uint64_t info; ///< Pointer to memory in which to write the info obtained.
 } sys_bpf_obj_info_attr_t;
 
+typedef struct
+{
+    ebpf_map_type_t type;            ///< Type of map.
+    ebpf_id_t id;                    ///< Map ID.
+    uint32_t key_size;               ///< Size in bytes of a map key.
+    uint32_t value_size;             ///< Size in bytes of a map value.
+    uint32_t max_entries;            ///< Maximum number of entries allowed in the map.
+    uint32_t map_flags;              ///< Map flags.
+    char name[SYS_BPF_OBJ_NAME_LEN]; ///< Null-terminated map name.
+} sys_bpf_map_info_t;
+
+typedef struct
+{
+    enum bpf_prog_type type;         ///< Program type.
+    ebpf_id_t id;                    ///< Program ID.
+    char tag[8];                     ///< Program tag.
+    uint32_t jited_prog_len;         ///< Not supported.
+    uint32_t xlated_prog_len;        ///< Not supported.
+    uint64_t jited_prog_insns;       ///< Not supported.
+    uint64_t xlated_prog_insns;      ///< Not supported.
+    uint64_t load_time;              ///< Not supported.
+    uint32_t created_by_uid;         ///< Not supported.
+    uint32_t nr_map_ids;             ///< Number of maps associated with this program.
+    uint64_t map_ids;                ///< Pointer to caller-allocated array to fill map IDs into.
+    char name[SYS_BPF_OBJ_NAME_LEN]; ///< Null-terminated program name.
+} sys_bpf_prog_info_t;
+
+typedef struct
+{
+    enum bpf_link_type type; ///< Link type.
+    ebpf_id_t id;            ///< Link ID.
+    ebpf_id_t prog_id;       ///< Program ID.
+} sys_bpf_link_info_t;
+
 /// Attributes used by BPF_LINK_DETACH.
 typedef struct
 {

libs/api/api_internal.h

@@ -521,7 +521,10 @@ ebpf_get_next_program_id(ebpf_id_t start_id, ebpf_id_t _Out_* next_id) noexcept;
  */
 _Must_inspect_result_ ebpf_result_t
 ebpf_object_get_info_by_fd(
-    fd_t bpf_fd, _Inout_updates_bytes_to_(*info_size, *info_size) void* info, _Inout_ uint32_t* info_size) noexcept;
+    fd_t bpf_fd,
+    _Inout_updates_bytes_to_(*info_size, *info_size) void* info,
+    _Inout_ uint32_t* info_size,
+    _Out_opt_ ebpf_object_type_t* type) noexcept;
 
 /**
  * @brief Pin an object to the specified path.

libs/api/bpf_syscall.cpp

@@ -1,7 +1,10 @@
 // Copyright (c) eBPF for Windows contributors
 // SPDX-License-Identifier: MIT
+#include "api_internal.h"
 #include "bpf.h"
 #include "libbpf.h"
+#include "libbpf_internal.h"
+#include "linux/bpf.h"
 
 #include <windows.h>
 #include <WinError.h>
@@ -69,6 +72,105 @@ template <typename T> class ExtensibleStruct
     }
 };
 
+static bool
+is_valid_name(_In_z_ const char* name, _Out_opt_ size_t* length)
+{
+    size_t name_length = strnlen(name, SYS_BPF_OBJ_NAME_LEN);
+
+    if (length != NULL) {
+        *length = name_length;
+    }
+
+    return name_length < SYS_BPF_OBJ_NAME_LEN;
+}
+
+static void
+convert_to_map_info(_Out_ struct bpf_map_info* bpf, _In_ const sys_bpf_map_info_t* sys);
+static void
+convert_to_sys_map_info(_Out_ sys_bpf_map_info_t* sys, _In_ const struct bpf_map_info* bpf);
+static void
+convert_to_prog_info(_Out_ struct bpf_prog_info* bpf, _In_ const sys_bpf_prog_info_t* sys);
+static void
+convert_to_sys_prog_info(_Out_ sys_bpf_prog_info_t* sys, _In_ const struct bpf_prog_info* bpf);
+static void
+convert_to_link_info(_Out_ struct bpf_link_info* bpf, _In_ const sys_bpf_link_info_t* sys);
+static void
+convert_to_sys_link_info(_Out_ sys_bpf_link_info_t* sys, _In_ const struct bpf_link_info* bpf);
+
+static int
+obj_get_info_by_fd(_In_ sys_bpf_obj_info_attr_t* attr)
+{
+    union
+    {
+        struct bpf_map_info map;
+        struct bpf_prog_info prog;
+        struct bpf_link_info link;
+    } tmp = {};
+    uint32_t info_size = sizeof(tmp);
+    ebpf_object_type_t type;
+
+    ebpf_result_t result = ebpf_object_get_info_by_fd((fd_t)attr->bpf_fd, &tmp, &info_size, &type);
+    if (result != EBPF_SUCCESS) {
+        return libbpf_result_err(result);
+    }
+
+    switch (type) {
+    case EBPF_OBJECT_MAP: {
+        ExtensibleStruct<sys_bpf_map_info_t> info((void*)attr->info, (size_t)attr->info_len);
+
+        convert_to_map_info(&tmp.map, &info);
+
+        info_size = sizeof(tmp.map);
+        result = ebpf_object_get_info_by_fd((fd_t)attr->bpf_fd, &tmp.map, &info_size, nullptr);
+        if (result != EBPF_SUCCESS) {
+            return libbpf_result_err(result);
+        }
+
+        convert_to_sys_map_info(&info, &tmp.map);
+        return 0;
+    }
+
+    case EBPF_OBJECT_PROGRAM: {
+        ExtensibleStruct<sys_bpf_prog_info_t> info((void*)attr->info, (size_t)attr->info_len);
+        sys_bpf_prog_info_t* sys = &info;
+
+        if (sys->jited_prog_len != 0 || sys->xlated_prog_len != 0 || sys->jited_prog_insns != 0 ||
+            sys->xlated_prog_insns != 0) {
+            return -EINVAL;
+        }
+
+        convert_to_prog_info(&tmp.prog, &info);
+
+        info_size = sizeof(tmp.prog);
+        result = ebpf_object_get_info_by_fd((fd_t)attr->bpf_fd, &tmp.prog, &info_size, nullptr);
+        if (result != EBPF_SUCCESS) {
+            return libbpf_result_err(result);
+        }
+
+        convert_to_sys_prog_info(&info, &tmp.prog);
+        return 0;
+    }
+
+    case EBPF_OBJECT_LINK: {
+        ExtensibleStruct<sys_bpf_link_info_t> info((void*)attr->info, (size_t)attr->info_len);
+
+        convert_to_link_info(&tmp.link, &info);
+
+        info_size = sizeof(tmp.link);
+        result = ebpf_object_get_info_by_fd((fd_t)attr->bpf_fd, &tmp.link, &info_size, nullptr);
+        if (result != EBPF_SUCCESS) {
+            return libbpf_result_err(result);
+        }
+
+        convert_to_sys_link_info(&info, &tmp.link);
+        return 0;
+    }
+
+    default:
+        return -EINVAL;
+    }
+}
+
 int
 bpf(int cmd, union bpf_attr* attr, unsigned int size)
 {
@@ -97,11 +199,17 @@ bpf(int cmd, union bpf_attr* attr, unsigned int size)
             struct bpf_map_create_opts opts = {
                 .inner_map_fd = map_create_attr->inner_map_fd,
                 .map_flags = map_create_attr->map_flags,
+                .numa_node = map_create_attr->numa_node,
+                .map_ifindex = map_create_attr->map_ifindex,
             };
 
+            if (!is_valid_name(map_create_attr->map_name, NULL)) {
+                return -EINVAL;
+            }
+
             return bpf_map_create(
                 map_create_attr->map_type,
-                nullptr,
+                map_create_attr->map_name,
                 map_create_attr->key_size,
                 map_create_attr->value_size,
                 map_create_attr->max_entries,
@@ -175,8 +283,7 @@ bpf(int cmd, union bpf_attr* attr, unsigned int size)
         }
         case BPF_OBJ_GET_INFO_BY_FD: {
             ExtensibleStruct<sys_bpf_obj_info_attr_t> info_by_fd_attr((void*)attr, (size_t)size);
-            return bpf_obj_get_info_by_fd(
-                info_by_fd_attr->bpf_fd, (void*)info_by_fd_attr->info, &info_by_fd_attr->info_len);
+            return obj_get_info_by_fd(&info_by_fd_attr);
         }
         case BPF_OBJ_PIN: {
             ExtensibleStruct<sys_bpf_obj_pin_attr_t> obj_pin_attr((void*)attr, (size_t)size);
@@ -212,10 +319,21 @@ bpf(int cmd, union bpf_attr* attr, unsigned int size)
                 .log_size = prog_load_attr->log_size,
                 .log_buf = (char*)prog_load_attr->log_buf,
             };
+            const char* name = prog_load_attr->prog_name;
+            size_t name_length;
+
+            if (!is_valid_name(name, &name_length)) {
+                return -EINVAL;
+            }
+
+            if (name_length == 0) {
+                // Disable using sha256 as object name.
+                name = "";
+            }
 
             return bpf_prog_load(
                 prog_load_attr->prog_type,
-                nullptr,
+                name,
                 (const char*)prog_load_attr->license,
                 (const struct bpf_insn*)prog_load_attr->insns,
                 prog_load_attr->insn_cnt,
@@ -262,3 +380,71 @@ bpf(int cmd, union bpf_attr* attr, unsigned int size)
         return -EINVAL;
     }
 }
+
+#define BPF_TO_SYS(field) sys->field = bpf->field
+#define BPF_TO_SYS_STR(field) strncpy_s(sys->field, sizeof(sys->field), bpf->field, _TRUNCATE)
+#define BPF_TO_SYS_MEM(field) memcpy(sys->field, bpf->field, sizeof(sys->field))
+#define SYS_TO_BPF(field) bpf->field = sys->field
+#define SYS_TO_BPF_STR(field) strncpy_s(bpf->field, sys->field, sizeof(bpf->field))
+#define SYS_TO_BPF_MEM(field) memcpy(bpf->field, sys->field, sizeof(bpf->field))
+
+static void
+convert_to_map_info(_Out_ struct bpf_map_info* bpf, _In_ const sys_bpf_map_info_t* sys)
+{
+    SYS_TO_BPF(type);
+    SYS_TO_BPF(id);
+    SYS_TO_BPF(key_size);
+    SYS_TO_BPF(value_size);
+    SYS_TO_BPF(max_entries);
+    SYS_TO_BPF(map_flags);
+    SYS_TO_BPF_STR(name);
+}
+
+static void
+convert_to_sys_map_info(_Out_ sys_bpf_map_info_t* sys, _In_ const struct bpf_map_info* bpf)
+{
+    BPF_TO_SYS(type);
+    BPF_TO_SYS(id);
+    BPF_TO_SYS(key_size);
+    BPF_TO_SYS(value_size);
+    BPF_TO_SYS(max_entries);
+    BPF_TO_SYS(map_flags);
+    BPF_TO_SYS_STR(name);
+}
+
+static void
+convert_to_prog_info(_Out_ struct bpf_prog_info* bpf, _In_ const sys_bpf_prog_info_t* sys)
+{
+    SYS_TO_BPF(type);
+    SYS_TO_BPF(id);
+    SYS_TO_BPF(nr_map_ids);
+    SYS_TO_BPF(map_ids);
+    SYS_TO_BPF_STR(name);
+}
+
+static void
+convert_to_sys_prog_info(_Out_ sys_bpf_prog_info_t* sys, _In_ const struct bpf_prog_info* bpf)
+{
+    *sys = {};
+    BPF_TO_SYS(type);
+    BPF_TO_SYS(id);
+    BPF_TO_SYS(nr_map_ids);
+    BPF_TO_SYS(map_ids);
+    BPF_TO_SYS_STR(name);
+}
+
+static void
+convert_to_link_info(_Out_ struct bpf_link_info* bpf, _In_ const sys_bpf_link_info_t* sys)
+{
+    SYS_TO_BPF(type);
+    SYS_TO_BPF(id);
+    SYS_TO_BPF(prog_id);
+}
+
+static void
+convert_to_sys_link_info(_Out_ sys_bpf_link_info_t* sys, _In_ const struct bpf_link_info* bpf)
+{
+    BPF_TO_SYS(type);
+    BPF_TO_SYS(id);
+    BPF_TO_SYS(prog_id);
+}