bpf tools: Improve libbpf error reporting
authorWang Nan <wangnan0@huawei.com>
Fri, 6 Nov 2015 13:49:37 +0000 (13:49 +0000)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 6 Nov 2015 18:52:41 +0000 (15:52 -0300)
In this patch, a series of libbpf specific error numbers and
libbpf_strerror() are introduced to help reporting errors.

Functions are updated to pass correct the error number through the
CHECK_ERR() macro.

All users of bpf_object__open{_buffer}() and bpf_program__title() in
perf are modified accordingly. In addition, due to the error codes
changing, bpf__strerror_load() is also modified to use them.

bpf__strerror_head() is also changed accordingly so it can parse libbpf
errors. bpf_loader_strerror() is introduced for that purpose, and will
be improved by the following patch.

load_program() is improved not to dump log buffer if it is empty. log
buffer is also used to deduce whether the error was caused by an invalid
program or other problem.

v1 -> v2:

 - Using macro for error code.

 - Fetch error message based on array index, eliminate for-loop.

 - Use log buffer to detect the reason of failure. 3 new error code
   are introduced to replace LIBBPF_ERRNO__LOAD.

In v1:

  # perf record -e ./test_ill_program.o ls
  event syntax error: './test_ill_program.o'
                       \___ Failed to load program: Validate your program and check 'license'/'version' sections in your object
  SKIP

  # perf record -e ./test_kversion_nomatch_program.o ls
  event syntax error: './test_kversion_nomatch_program.o'
                       \___ Failed to load program: Validate your program and check 'license'/'version' sections in your object
  SKIP

  # perf record -e ./test_big_program.o ls
  event syntax error: './test_big_program.o'
                       \___ Failed to load program: Validate your program and check 'license'/'version' sections in your object
  SKIP

  In v2:

  # perf record -e ./test_ill_program.o ls
  event syntax error: './test_ill_program.o'
                       \___ Kernel verifier blocks program loading
  SKIP

  # perf record -e ./test_kversion_nomatch_program.o
  event syntax error: './test_kversion_nomatch_program.o'
                       \___ Incorrect kernel version
  SKIP
  (Will be further improved by following patches)

  # perf record -e ./test_big_program.o
  event syntax error: './test_big_program.o'
                       \___ Program too big
  SKIP

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1446817783-86722-2-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/lib/bpf/libbpf.c
tools/lib/bpf/libbpf.h
tools/perf/tests/llvm.c
tools/perf/util/bpf-loader.c
tools/perf/util/parse-events.c

index 9f3c8cf9249b6d2e9cb87629f8a26237f98e2027..07b492d3dfaa16dd2a4901d5df4b7c418fc1de17 100644 (file)
@@ -61,6 +61,60 @@ void libbpf_set_print(libbpf_print_fn_t warn,
        __pr_debug = debug;
 }
 
+#define STRERR_BUFSIZE  128
+
+#define ERRNO_OFFSET(e)                ((e) - __LIBBPF_ERRNO__START)
+#define ERRCODE_OFFSET(c)      ERRNO_OFFSET(LIBBPF_ERRNO__##c)
+#define NR_ERRNO       (__LIBBPF_ERRNO__END - __LIBBPF_ERRNO__START)
+
+static const char *libbpf_strerror_table[NR_ERRNO] = {
+       [ERRCODE_OFFSET(LIBELF)]        = "Something wrong in libelf",
+       [ERRCODE_OFFSET(FORMAT)]        = "BPF object format invalid",
+       [ERRCODE_OFFSET(KVERSION)]      = "'version' section incorrect or lost",
+       [ERRCODE_OFFSET(ENDIAN)]        = "Endian missmatch",
+       [ERRCODE_OFFSET(INTERNAL)]      = "Internal error in libbpf",
+       [ERRCODE_OFFSET(RELOC)]         = "Relocation failed",
+       [ERRCODE_OFFSET(VERIFY)]        = "Kernel verifier blocks program loading",
+       [ERRCODE_OFFSET(PROG2BIG)]      = "Program too big",
+       [ERRCODE_OFFSET(KVER)]          = "Incorrect kernel version",
+};
+
+int libbpf_strerror(int err, char *buf, size_t size)
+{
+       if (!buf || !size)
+               return -1;
+
+       err = err > 0 ? err : -err;
+
+       if (err < __LIBBPF_ERRNO__START) {
+               int ret;
+
+               ret = strerror_r(err, buf, size);
+               buf[size - 1] = '\0';
+               return ret;
+       }
+
+       if (err < __LIBBPF_ERRNO__END) {
+               const char *msg;
+
+               msg = libbpf_strerror_table[ERRNO_OFFSET(err)];
+               snprintf(buf, size, "%s", msg);
+               buf[size - 1] = '\0';
+               return 0;
+       }
+
+       snprintf(buf, size, "Unknown libbpf error %d", err);
+       buf[size - 1] = '\0';
+       return -1;
+}
+
+#define CHECK_ERR(action, err, out) do {       \
+       err = action;                   \
+       if (err)                        \
+               goto out;               \
+} while(0)
+
+
 /* Copied from tools/perf/util/util.h */
 #ifndef zfree
 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
@@ -258,7 +312,7 @@ static struct bpf_object *bpf_object__new(const char *path,
        obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
        if (!obj) {
                pr_warning("alloc memory failed for %s\n", path);
-               return NULL;
+               return ERR_PTR(-ENOMEM);
        }
 
        strcpy(obj->path, path);
@@ -305,7 +359,7 @@ static int bpf_object__elf_init(struct bpf_object *obj)
 
        if (obj_elf_valid(obj)) {
                pr_warning("elf init: internal error\n");
-               return -EEXIST;
+               return -LIBBPF_ERRNO__LIBELF;
        }
 
        if (obj->efile.obj_buf_sz > 0) {
@@ -331,14 +385,14 @@ static int bpf_object__elf_init(struct bpf_object *obj)
        if (!obj->efile.elf) {
                pr_warning("failed to open %s as ELF file\n",
                                obj->path);
-               err = -EINVAL;
+               err = -LIBBPF_ERRNO__LIBELF;
                goto errout;
        }
 
        if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
                pr_warning("failed to get EHDR from %s\n",
                                obj->path);
-               err = -EINVAL;
+               err = -LIBBPF_ERRNO__FORMAT;
                goto errout;
        }
        ep = &obj->efile.ehdr;
@@ -346,7 +400,7 @@ static int bpf_object__elf_init(struct bpf_object *obj)
        if ((ep->e_type != ET_REL) || (ep->e_machine != 0)) {
                pr_warning("%s is not an eBPF object file\n",
                        obj->path);
-               err = -EINVAL;
+               err = -LIBBPF_ERRNO__FORMAT;
                goto errout;
        }
 
@@ -374,14 +428,14 @@ bpf_object__check_endianness(struct bpf_object *obj)
                        goto mismatch;
                break;
        default:
-               return -EINVAL;
+               return -LIBBPF_ERRNO__ENDIAN;
        }
 
        return 0;
 
 mismatch:
        pr_warning("Error: endianness mismatch.\n");
-       return -EINVAL;
+       return -LIBBPF_ERRNO__ENDIAN;
 }
 
 static int
@@ -402,7 +456,7 @@ bpf_object__init_kversion(struct bpf_object *obj,
 
        if (size != sizeof(kver)) {
                pr_warning("invalid kver section in %s\n", obj->path);
-               return -EINVAL;
+               return -LIBBPF_ERRNO__FORMAT;
        }
        memcpy(&kver, data, sizeof(kver));
        obj->kern_version = kver;
@@ -444,7 +498,7 @@ static int bpf_object__elf_collect(struct bpf_object *obj)
        if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
                pr_warning("failed to get e_shstrndx from %s\n",
                           obj->path);
-               return -EINVAL;
+               return -LIBBPF_ERRNO__FORMAT;
        }
 
        while ((scn = elf_nextscn(elf, scn)) != NULL) {
@@ -456,7 +510,7 @@ static int bpf_object__elf_collect(struct bpf_object *obj)
                if (gelf_getshdr(scn, &sh) != &sh) {
                        pr_warning("failed to get section header from %s\n",
                                   obj->path);
-                       err = -EINVAL;
+                       err = -LIBBPF_ERRNO__FORMAT;
                        goto out;
                }
 
@@ -464,7 +518,7 @@ static int bpf_object__elf_collect(struct bpf_object *obj)
                if (!name) {
                        pr_warning("failed to get section name from %s\n",
                                   obj->path);
-                       err = -EINVAL;
+                       err = -LIBBPF_ERRNO__FORMAT;
                        goto out;
                }
 
@@ -472,7 +526,7 @@ static int bpf_object__elf_collect(struct bpf_object *obj)
                if (!data) {
                        pr_warning("failed to get section data from %s(%s)\n",
                                   name, obj->path);
-                       err = -EINVAL;
+                       err = -LIBBPF_ERRNO__FORMAT;
                        goto out;
                }
                pr_debug("section %s, size %ld, link %d, flags %lx, type=%d\n",
@@ -495,7 +549,7 @@ static int bpf_object__elf_collect(struct bpf_object *obj)
                        if (obj->efile.symbols) {
                                pr_warning("bpf: multiple SYMTAB in %s\n",
                                           obj->path);
-                               err = -EEXIST;
+                               err = -LIBBPF_ERRNO__FORMAT;
                        } else
                                obj->efile.symbols = data;
                } else if ((sh.sh_type == SHT_PROGBITS) &&
@@ -504,7 +558,8 @@ static int bpf_object__elf_collect(struct bpf_object *obj)
                        err = bpf_object__add_program(obj, data->d_buf,
                                                      data->d_size, name, idx);
                        if (err) {
-                               char errmsg[128];
+                               char errmsg[STRERR_BUFSIZE];
+
                                strerror_r(-err, errmsg, sizeof(errmsg));
                                pr_warning("failed to alloc program %s (%s): %s",
                                           name, obj->path, errmsg);
@@ -576,7 +631,7 @@ bpf_program__collect_reloc(struct bpf_program *prog,
 
                if (!gelf_getrel(data, i, &rel)) {
                        pr_warning("relocation: failed to get %d reloc\n", i);
-                       return -EINVAL;
+                       return -LIBBPF_ERRNO__FORMAT;
                }
 
                insn_idx = rel.r_offset / sizeof(struct bpf_insn);
@@ -587,20 +642,20 @@ bpf_program__collect_reloc(struct bpf_program *prog,
                                 &sym)) {
                        pr_warning("relocation: symbol %"PRIx64" not found\n",
                                   GELF_R_SYM(rel.r_info));
-                       return -EINVAL;
+                       return -LIBBPF_ERRNO__FORMAT;
                }
 
                if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
                        pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
                                   insn_idx, insns[insn_idx].code);
-                       return -EINVAL;
+                       return -LIBBPF_ERRNO__RELOC;
                }
 
                map_idx = sym.st_value / sizeof(struct bpf_map_def);
                if (map_idx >= nr_maps) {
                        pr_warning("bpf relocation: map_idx %d large than %d\n",
                                   (int)map_idx, (int)nr_maps - 1);
-                       return -EINVAL;
+                       return -LIBBPF_ERRNO__RELOC;
                }
 
                prog->reloc_desc[i].insn_idx = insn_idx;
@@ -683,7 +738,7 @@ bpf_program__relocate(struct bpf_program *prog, int *map_fds)
                if (insn_idx >= (int)prog->insns_cnt) {
                        pr_warning("relocation out of range: '%s'\n",
                                   prog->section_name);
-                       return -ERANGE;
+                       return -LIBBPF_ERRNO__RELOC;
                }
                insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
                insns[insn_idx].imm = map_fds[map_idx];
@@ -721,7 +776,7 @@ static int bpf_object__collect_reloc(struct bpf_object *obj)
 
        if (!obj_elf_valid(obj)) {
                pr_warning("Internal error: elf object is closed\n");
-               return -EINVAL;
+               return -LIBBPF_ERRNO__INTERNAL;
        }
 
        for (i = 0; i < obj->efile.nr_reloc; i++) {
@@ -734,21 +789,21 @@ static int bpf_object__collect_reloc(struct bpf_object *obj)
 
                if (shdr->sh_type != SHT_REL) {
                        pr_warning("internal error at %d\n", __LINE__);
-                       return -EINVAL;
+                       return -LIBBPF_ERRNO__INTERNAL;
                }
 
                prog = bpf_object__find_prog_by_idx(obj, idx);
                if (!prog) {
                        pr_warning("relocation failed: no %d section\n",
                                   idx);
-                       return -ENOENT;
+                       return -LIBBPF_ERRNO__RELOC;
                }
 
                err = bpf_program__collect_reloc(prog, nr_maps,
                                                 shdr, data,
                                                 obj->efile.symbols);
                if (err)
-                       return -EINVAL;
+                       return err;
        }
        return 0;
 }
@@ -777,13 +832,23 @@ load_program(struct bpf_insn *insns, int insns_cnt,
                goto out;
        }
 
-       ret = -EINVAL;
+       ret = -LIBBPF_ERRNO__LOAD;
        pr_warning("load bpf program failed: %s\n", strerror(errno));
 
-       if (log_buf) {
+       if (log_buf && log_buf[0] != '\0') {
+               ret = -LIBBPF_ERRNO__VERIFY;
                pr_warning("-- BEGIN DUMP LOG ---\n");
                pr_warning("\n%s\n", log_buf);
                pr_warning("-- END LOG --\n");
+       } else {
+               if (insns_cnt >= BPF_MAXINSNS) {
+                       pr_warning("Program too large (%d insns), at most %d insns\n",
+                                  insns_cnt, BPF_MAXINSNS);
+                       ret = -LIBBPF_ERRNO__PROG2BIG;
+               } else if (log_buf) {
+                       pr_warning("log buffer is empty\n");
+                       ret = -LIBBPF_ERRNO__KVER;
+               }
        }
 
 out:
@@ -831,7 +896,7 @@ static int bpf_object__validate(struct bpf_object *obj)
        if (obj->kern_version == 0) {
                pr_warning("%s doesn't provide kernel version\n",
                           obj->path);
-               return -EINVAL;
+               return -LIBBPF_ERRNO__KVERSION;
        }
        return 0;
 }
@@ -840,32 +905,28 @@ static struct bpf_object *
 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz)
 {
        struct bpf_object *obj;
+       int err;
 
        if (elf_version(EV_CURRENT) == EV_NONE) {
                pr_warning("failed to init libelf for %s\n", path);
-               return NULL;
+               return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
        }
 
        obj = bpf_object__new(path, obj_buf, obj_buf_sz);
-       if (!obj)
-               return NULL;
+       if (IS_ERR(obj))
+               return obj;
 
-       if (bpf_object__elf_init(obj))
-               goto out;
-       if (bpf_object__check_endianness(obj))
-               goto out;
-       if (bpf_object__elf_collect(obj))
-               goto out;
-       if (bpf_object__collect_reloc(obj))
-               goto out;
-       if (bpf_object__validate(obj))
-               goto out;
+       CHECK_ERR(bpf_object__elf_init(obj), err, out);
+       CHECK_ERR(bpf_object__check_endianness(obj), err, out);
+       CHECK_ERR(bpf_object__elf_collect(obj), err, out);
+       CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
+       CHECK_ERR(bpf_object__validate(obj), err, out);
 
        bpf_object__elf_finish(obj);
        return obj;
 out:
        bpf_object__close(obj);
-       return NULL;
+       return ERR_PTR(err);
 }
 
 struct bpf_object *bpf_object__open(const char *path)
@@ -922,6 +983,8 @@ int bpf_object__unload(struct bpf_object *obj)
 
 int bpf_object__load(struct bpf_object *obj)
 {
+       int err;
+
        if (!obj)
                return -EINVAL;
 
@@ -931,18 +994,16 @@ int bpf_object__load(struct bpf_object *obj)
        }
 
        obj->loaded = true;
-       if (bpf_object__create_maps(obj))
-               goto out;
-       if (bpf_object__relocate(obj))
-               goto out;
-       if (bpf_object__load_progs(obj))
-               goto out;
+
+       CHECK_ERR(bpf_object__create_maps(obj), err, out);
+       CHECK_ERR(bpf_object__relocate(obj), err, out);
+       CHECK_ERR(bpf_object__load_progs(obj), err, out);
 
        return 0;
 out:
        bpf_object__unload(obj);
        pr_warning("failed to load object '%s'\n", obj->path);
-       return -EINVAL;
+       return err;
 }
 
 void bpf_object__close(struct bpf_object *obj)
@@ -990,7 +1051,7 @@ const char *
 bpf_object__get_name(struct bpf_object *obj)
 {
        if (!obj)
-               return NULL;
+               return ERR_PTR(-EINVAL);
        return obj->path;
 }
 
@@ -1043,7 +1104,7 @@ const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
                title = strdup(title);
                if (!title) {
                        pr_warning("failed to strdup program title\n");
-                       return NULL;
+                       return ERR_PTR(-ENOMEM);
                }
        }
 
index bc80af03c6f41b5ff976aa2124cc6c25dc6f021a..30a40e9fa503d34d867b98a26d955f85fbd74396 100644 (file)
 
 #include <stdio.h>
 #include <stdbool.h>
+#include <linux/err.h>
+
+enum libbpf_errno {
+       __LIBBPF_ERRNO__START = 4000,
+
+       /* Something wrong in libelf */
+       LIBBPF_ERRNO__LIBELF = __LIBBPF_ERRNO__START,
+       LIBBPF_ERRNO__FORMAT,   /* BPF object format invalid */
+       LIBBPF_ERRNO__KVERSION, /* Incorrect or no 'version' section */
+       LIBBPF_ERRNO__ENDIAN,   /* Endian missmatch */
+       LIBBPF_ERRNO__INTERNAL, /* Internal error in libbpf */
+       LIBBPF_ERRNO__RELOC,    /* Relocation failed */
+       LIBBPF_ERRNO__LOAD,     /* Load program failure for unknown reason */
+       LIBBPF_ERRNO__VERIFY,   /* Kernel verifier blocks program loading */
+       LIBBPF_ERRNO__PROG2BIG, /* Program too big */
+       LIBBPF_ERRNO__KVER,     /* Incorrect kernel version */
+       __LIBBPF_ERRNO__END,
+};
+
+int libbpf_strerror(int err, char *buf, size_t size);
 
 /*
  * In include/linux/compiler-gcc.h, __printf is defined. However
index 512d3620e9f9439b9bf81bbc076c0720e1d128e7..8f713f6d32f35b11ef5cbe1add73ac227e7124d7 100644 (file)
@@ -27,7 +27,7 @@ static int test__bpf_parsing(void *obj_buf, size_t obj_buf_sz)
        struct bpf_object *obj;
 
        obj = bpf_object__open_buffer(obj_buf, obj_buf_sz, NULL);
-       if (!obj)
+       if (IS_ERR(obj))
                return -1;
        bpf_object__close(obj);
        return 0;
index 0c5d174245bb19c69c78513f7fd7be6b5bf1a395..c46256b1f5fdec46d321f2d53b4f8c073e1bcf2d 100644 (file)
@@ -59,9 +59,9 @@ struct bpf_object *bpf__prepare_load(const char *filename, bool source)
        } else
                obj = bpf_object__open(filename);
 
-       if (!obj) {
+       if (IS_ERR(obj)) {
                pr_debug("bpf: failed to load %s\n", filename);
-               return ERR_PTR(-EINVAL);
+               return obj;
        }
 
        return obj;
@@ -96,9 +96,9 @@ config_bpf_program(struct bpf_program *prog)
        int err;
 
        config_str = bpf_program__title(prog, false);
-       if (!config_str) {
+       if (IS_ERR(config_str)) {
                pr_debug("bpf: unable to get title for program\n");
-               return -EINVAL;
+               return PTR_ERR(config_str);
        }
 
        priv = calloc(sizeof(*priv), 1);
@@ -308,13 +308,34 @@ int bpf__foreach_tev(struct bpf_object *obj,
        return 0;
 }
 
+static int
+bpf_loader_strerror(int err, char *buf, size_t size)
+{
+       char sbuf[STRERR_BUFSIZE];
+       const char *msg;
+
+       if (!buf || !size)
+               return -1;
+
+       err = err > 0 ? err : -err;
+
+       if (err >= __LIBBPF_ERRNO__START)
+               return libbpf_strerror(err, buf, size);
+
+       msg = strerror_r(err, sbuf, sizeof(sbuf));
+       snprintf(buf, size, "%s", msg);
+       buf[size - 1] = '\0';
+       return 0;
+}
+
 #define bpf__strerror_head(err, buf, size) \
        char sbuf[STRERR_BUFSIZE], *emsg;\
        if (!size)\
                return 0;\
        if (err < 0)\
                err = -err;\
-       emsg = strerror_r(err, sbuf, sizeof(sbuf));\
+       bpf_loader_strerror(err, sbuf, sizeof(sbuf));\
+       emsg = sbuf;\
        switch (err) {\
        default:\
                scnprintf(buf, size, "%s", emsg);\
@@ -345,8 +366,6 @@ int bpf__strerror_load(struct bpf_object *obj __maybe_unused,
                       int err, char *buf, size_t size)
 {
        bpf__strerror_head(err, buf, size);
-       bpf__strerror_entry(EINVAL, "%s: Are you root and runing a CONFIG_BPF_SYSCALL kernel?",
-                           emsg)
        bpf__strerror_end(buf, size);
        return 0;
 }
index bee60583839a7cdae0dd44f5c48dd818d53cb5e4..c75b25d5e28cb80f60b311a2d7e98a9f612e4530 100644 (file)
@@ -632,11 +632,11 @@ int parse_events_load_bpf(struct parse_events_evlist *data,
        struct bpf_object *obj;
 
        obj = bpf__prepare_load(bpf_file_name, source);
-       if (IS_ERR(obj) || !obj) {
+       if (IS_ERR(obj)) {
                char errbuf[BUFSIZ];
                int err;
 
-               err = obj ? PTR_ERR(obj) : -EINVAL;
+               err = PTR_ERR(obj);
 
                if (err == -ENOTSUP)
                        snprintf(errbuf, sizeof(errbuf),