tools: bpf: make use of reallocarray
authorJakub Kicinski <jakub.kicinski@netronome.com>
Tue, 10 Jul 2018 21:43:05 +0000 (14:43 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Wed, 11 Jul 2018 20:13:34 +0000 (22:13 +0200)
reallocarray() is a safer variant of realloc which checks for
multiplication overflow in case of array allocation.  Since it's
not available in Glibc < 2.26 import kernel's overflow.h and
add a static inline implementation when needed.  Use feature
detection to probe for existence of reallocarray.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Reviewed-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
tools/bpf/bpftool/Makefile
tools/bpf/bpftool/main.h
tools/bpf/bpftool/xlated_dumper.c
tools/build/feature/Makefile
tools/build/feature/test-reallocarray.c [new file with mode: 0644]
tools/include/linux/compiler-gcc.h
tools/include/linux/overflow.h [new file with mode: 0644]
tools/include/tools/libc_compat.h [new file with mode: 0644]
tools/lib/bpf/Makefile
tools/lib/bpf/libbpf.c

index 0911b00b25cc32fe633372a47db26ec95a46ceec..6c4830e188798277453ae0097b96b5886bc36c9e 100644 (file)
@@ -52,7 +52,7 @@ INSTALL ?= install
 RM ?= rm -f
 
 FEATURE_USER = .bpftool
-FEATURE_TESTS = libbfd disassembler-four-args
+FEATURE_TESTS = libbfd disassembler-four-args reallocarray
 FEATURE_DISPLAY = libbfd disassembler-four-args
 
 check_feat := 1
@@ -75,6 +75,10 @@ ifeq ($(feature-disassembler-four-args), 1)
 CFLAGS += -DDISASM_FOUR_ARGS_SIGNATURE
 endif
 
+ifeq ($(feature-reallocarray), 0)
+CFLAGS += -DCOMPAT_NEED_REALLOCARRAY
+endif
+
 include $(wildcard $(OUTPUT)*.d)
 
 all: $(OUTPUT)bpftool
index 15b6c49ae533799226b1976b5477e2d8822a766b..1e02e403169332a818cce3c8d9cfb93bceaa9d9c 100644 (file)
@@ -42,6 +42,7 @@
 #include <linux/compiler.h>
 #include <linux/kernel.h>
 #include <linux/hashtable.h>
+#include <tools/libc_compat.h>
 
 #include "json_writer.h"
 
index b97f1da60dd1d75f401501428f4c1bfdecdd69f3..3284759df98ad4f325b25f719db0cfef7ac9f61c 100644 (file)
@@ -35,6 +35,7 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#define _GNU_SOURCE
 #include <stdarg.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -66,9 +67,8 @@ void kernel_syms_load(struct dump_data *dd)
        while (!feof(fp)) {
                if (!fgets(buff, sizeof(buff), fp))
                        break;
-               tmp = realloc(dd->sym_mapping,
-                             (dd->sym_count + 1) *
-                             sizeof(*dd->sym_mapping));
+               tmp = reallocarray(dd->sym_mapping, dd->sym_count + 1,
+                                  sizeof(*dd->sym_mapping));
                if (!tmp) {
 out:
                        free(dd->sym_mapping);
index dac9563b54707c6cb65deb3e609ecc9785d4bf0c..0516259be70f071f2533496ead690d1ebd5ba3b8 100644 (file)
@@ -14,6 +14,7 @@ FILES=                                          \
          test-libaudit.bin                      \
          test-libbfd.bin                        \
          test-disassembler-four-args.bin        \
+         test-reallocarray.bin                 \
          test-liberty.bin                       \
          test-liberty-z.bin                     \
          test-cplus-demangle.bin                \
@@ -204,6 +205,9 @@ $(OUTPUT)test-libbfd.bin:
 $(OUTPUT)test-disassembler-four-args.bin:
        $(BUILD) -DPACKAGE='"perf"' -lbfd -lopcodes
 
+$(OUTPUT)test-reallocarray.bin:
+       $(BUILD)
+
 $(OUTPUT)test-liberty.bin:
        $(CC) $(CFLAGS) -Wall -Werror -o $@ test-libbfd.c -DPACKAGE='"perf"' $(LDFLAGS) -lbfd -ldl -liberty
 
diff --git a/tools/build/feature/test-reallocarray.c b/tools/build/feature/test-reallocarray.c
new file mode 100644 (file)
index 0000000..8170de3
--- /dev/null
@@ -0,0 +1,8 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <stdlib.h>
+
+int main(void)
+{
+       return !!reallocarray(NULL, 1, 1);
+}
index 70fe612957338c84ffeb61f03c358c6ca1e32789..0d35f18006a136b4578f2298243e3728a067c272 100644 (file)
@@ -36,3 +36,7 @@
 #endif
 #define __printf(a, b) __attribute__((format(printf, a, b)))
 #define __scanf(a, b)  __attribute__((format(scanf, a, b)))
+
+#if GCC_VERSION >= 50100
+#define COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW 1
+#endif
diff --git a/tools/include/linux/overflow.h b/tools/include/linux/overflow.h
new file mode 100644 (file)
index 0000000..8712ff7
--- /dev/null
@@ -0,0 +1,278 @@
+/* SPDX-License-Identifier: GPL-2.0 OR MIT */
+#ifndef __LINUX_OVERFLOW_H
+#define __LINUX_OVERFLOW_H
+
+#include <linux/compiler.h>
+
+/*
+ * In the fallback code below, we need to compute the minimum and
+ * maximum values representable in a given type. These macros may also
+ * be useful elsewhere, so we provide them outside the
+ * COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW block.
+ *
+ * It would seem more obvious to do something like
+ *
+ * #define type_min(T) (T)(is_signed_type(T) ? (T)1 << (8*sizeof(T)-1) : 0)
+ * #define type_max(T) (T)(is_signed_type(T) ? ((T)1 << (8*sizeof(T)-1)) - 1 : ~(T)0)
+ *
+ * Unfortunately, the middle expressions, strictly speaking, have
+ * undefined behaviour, and at least some versions of gcc warn about
+ * the type_max expression (but not if -fsanitize=undefined is in
+ * effect; in that case, the warning is deferred to runtime...).
+ *
+ * The slightly excessive casting in type_min is to make sure the
+ * macros also produce sensible values for the exotic type _Bool. [The
+ * overflow checkers only almost work for _Bool, but that's
+ * a-feature-not-a-bug, since people shouldn't be doing arithmetic on
+ * _Bools. Besides, the gcc builtins don't allow _Bool* as third
+ * argument.]
+ *
+ * Idea stolen from
+ * https://mail-index.netbsd.org/tech-misc/2007/02/05/0000.html -
+ * credit to Christian Biere.
+ */
+#define is_signed_type(type)       (((type)(-1)) < (type)1)
+#define __type_half_max(type) ((type)1 << (8*sizeof(type) - 1 - is_signed_type(type)))
+#define type_max(T) ((T)((__type_half_max(T) - 1) + __type_half_max(T)))
+#define type_min(T) ((T)((T)-type_max(T)-(T)1))
+
+
+#ifdef COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW
+/*
+ * For simplicity and code hygiene, the fallback code below insists on
+ * a, b and *d having the same type (similar to the min() and max()
+ * macros), whereas gcc's type-generic overflow checkers accept
+ * different types. Hence we don't just make check_add_overflow an
+ * alias for __builtin_add_overflow, but add type checks similar to
+ * below.
+ */
+#define check_add_overflow(a, b, d) ({         \
+       typeof(a) __a = (a);                    \
+       typeof(b) __b = (b);                    \
+       typeof(d) __d = (d);                    \
+       (void) (&__a == &__b);                  \
+       (void) (&__a == __d);                   \
+       __builtin_add_overflow(__a, __b, __d);  \
+})
+
+#define check_sub_overflow(a, b, d) ({         \
+       typeof(a) __a = (a);                    \
+       typeof(b) __b = (b);                    \
+       typeof(d) __d = (d);                    \
+       (void) (&__a == &__b);                  \
+       (void) (&__a == __d);                   \
+       __builtin_sub_overflow(__a, __b, __d);  \
+})
+
+#define check_mul_overflow(a, b, d) ({         \
+       typeof(a) __a = (a);                    \
+       typeof(b) __b = (b);                    \
+       typeof(d) __d = (d);                    \
+       (void) (&__a == &__b);                  \
+       (void) (&__a == __d);                   \
+       __builtin_mul_overflow(__a, __b, __d);  \
+})
+
+#else
+
+
+/* Checking for unsigned overflow is relatively easy without causing UB. */
+#define __unsigned_add_overflow(a, b, d) ({    \
+       typeof(a) __a = (a);                    \
+       typeof(b) __b = (b);                    \
+       typeof(d) __d = (d);                    \
+       (void) (&__a == &__b);                  \
+       (void) (&__a == __d);                   \
+       *__d = __a + __b;                       \
+       *__d < __a;                             \
+})
+#define __unsigned_sub_overflow(a, b, d) ({    \
+       typeof(a) __a = (a);                    \
+       typeof(b) __b = (b);                    \
+       typeof(d) __d = (d);                    \
+       (void) (&__a == &__b);                  \
+       (void) (&__a == __d);                   \
+       *__d = __a - __b;                       \
+       __a < __b;                              \
+})
+/*
+ * If one of a or b is a compile-time constant, this avoids a division.
+ */
+#define __unsigned_mul_overflow(a, b, d) ({            \
+       typeof(a) __a = (a);                            \
+       typeof(b) __b = (b);                            \
+       typeof(d) __d = (d);                            \
+       (void) (&__a == &__b);                          \
+       (void) (&__a == __d);                           \
+       *__d = __a * __b;                               \
+       __builtin_constant_p(__b) ?                     \
+         __b > 0 && __a > type_max(typeof(__a)) / __b : \
+         __a > 0 && __b > type_max(typeof(__b)) / __a;  \
+})
+
+/*
+ * For signed types, detecting overflow is much harder, especially if
+ * we want to avoid UB. But the interface of these macros is such that
+ * we must provide a result in *d, and in fact we must produce the
+ * result promised by gcc's builtins, which is simply the possibly
+ * wrapped-around value. Fortunately, we can just formally do the
+ * operations in the widest relevant unsigned type (u64) and then
+ * truncate the result - gcc is smart enough to generate the same code
+ * with and without the (u64) casts.
+ */
+
+/*
+ * Adding two signed integers can overflow only if they have the same
+ * sign, and overflow has happened iff the result has the opposite
+ * sign.
+ */
+#define __signed_add_overflow(a, b, d) ({      \
+       typeof(a) __a = (a);                    \
+       typeof(b) __b = (b);                    \
+       typeof(d) __d = (d);                    \
+       (void) (&__a == &__b);                  \
+       (void) (&__a == __d);                   \
+       *__d = (u64)__a + (u64)__b;             \
+       (((~(__a ^ __b)) & (*__d ^ __a))        \
+               & type_min(typeof(__a))) != 0;  \
+})
+
+/*
+ * Subtraction is similar, except that overflow can now happen only
+ * when the signs are opposite. In this case, overflow has happened if
+ * the result has the opposite sign of a.
+ */
+#define __signed_sub_overflow(a, b, d) ({      \
+       typeof(a) __a = (a);                    \
+       typeof(b) __b = (b);                    \
+       typeof(d) __d = (d);                    \
+       (void) (&__a == &__b);                  \
+       (void) (&__a == __d);                   \
+       *__d = (u64)__a - (u64)__b;             \
+       ((((__a ^ __b)) & (*__d ^ __a))         \
+               & type_min(typeof(__a))) != 0;  \
+})
+
+/*
+ * Signed multiplication is rather hard. gcc always follows C99, so
+ * division is truncated towards 0. This means that we can write the
+ * overflow check like this:
+ *
+ * (a > 0 && (b > MAX/a || b < MIN/a)) ||
+ * (a < -1 && (b > MIN/a || b < MAX/a) ||
+ * (a == -1 && b == MIN)
+ *
+ * The redundant casts of -1 are to silence an annoying -Wtype-limits
+ * (included in -Wextra) warning: When the type is u8 or u16, the
+ * __b_c_e in check_mul_overflow obviously selects
+ * __unsigned_mul_overflow, but unfortunately gcc still parses this
+ * code and warns about the limited range of __b.
+ */
+
+#define __signed_mul_overflow(a, b, d) ({                              \
+       typeof(a) __a = (a);                                            \
+       typeof(b) __b = (b);                                            \
+       typeof(d) __d = (d);                                            \
+       typeof(a) __tmax = type_max(typeof(a));                         \
+       typeof(a) __tmin = type_min(typeof(a));                         \
+       (void) (&__a == &__b);                                          \
+       (void) (&__a == __d);                                           \
+       *__d = (u64)__a * (u64)__b;                                     \
+       (__b > 0   && (__a > __tmax/__b || __a < __tmin/__b)) ||        \
+       (__b < (typeof(__b))-1  && (__a > __tmin/__b || __a < __tmax/__b)) || \
+       (__b == (typeof(__b))-1 && __a == __tmin);                      \
+})
+
+
+#define check_add_overflow(a, b, d)                                    \
+       __builtin_choose_expr(is_signed_type(typeof(a)),                \
+                       __signed_add_overflow(a, b, d),                 \
+                       __unsigned_add_overflow(a, b, d))
+
+#define check_sub_overflow(a, b, d)                                    \
+       __builtin_choose_expr(is_signed_type(typeof(a)),                \
+                       __signed_sub_overflow(a, b, d),                 \
+                       __unsigned_sub_overflow(a, b, d))
+
+#define check_mul_overflow(a, b, d)                                    \
+       __builtin_choose_expr(is_signed_type(typeof(a)),                \
+                       __signed_mul_overflow(a, b, d),                 \
+                       __unsigned_mul_overflow(a, b, d))
+
+
+#endif /* COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW */
+
+/**
+ * array_size() - Calculate size of 2-dimensional array.
+ *
+ * @a: dimension one
+ * @b: dimension two
+ *
+ * Calculates size of 2-dimensional array: @a * @b.
+ *
+ * Returns: number of bytes needed to represent the array or SIZE_MAX on
+ * overflow.
+ */
+static inline __must_check size_t array_size(size_t a, size_t b)
+{
+       size_t bytes;
+
+       if (check_mul_overflow(a, b, &bytes))
+               return SIZE_MAX;
+
+       return bytes;
+}
+
+/**
+ * array3_size() - Calculate size of 3-dimensional array.
+ *
+ * @a: dimension one
+ * @b: dimension two
+ * @c: dimension three
+ *
+ * Calculates size of 3-dimensional array: @a * @b * @c.
+ *
+ * Returns: number of bytes needed to represent the array or SIZE_MAX on
+ * overflow.
+ */
+static inline __must_check size_t array3_size(size_t a, size_t b, size_t c)
+{
+       size_t bytes;
+
+       if (check_mul_overflow(a, b, &bytes))
+               return SIZE_MAX;
+       if (check_mul_overflow(bytes, c, &bytes))
+               return SIZE_MAX;
+
+       return bytes;
+}
+
+static inline __must_check size_t __ab_c_size(size_t n, size_t size, size_t c)
+{
+       size_t bytes;
+
+       if (check_mul_overflow(n, size, &bytes))
+               return SIZE_MAX;
+       if (check_add_overflow(bytes, c, &bytes))
+               return SIZE_MAX;
+
+       return bytes;
+}
+
+/**
+ * struct_size() - Calculate size of structure with trailing array.
+ * @p: Pointer to the structure.
+ * @member: Name of the array member.
+ * @n: Number of elements in the array.
+ *
+ * Calculates size of memory needed for structure @p followed by an
+ * array of @n @member elements.
+ *
+ * Return: number of bytes needed or SIZE_MAX on overflow.
+ */
+#define struct_size(p, member, n)                                      \
+       __ab_c_size(n,                                                  \
+                   sizeof(*(p)->member) + __must_be_array((p)->member),\
+                   sizeof(*(p)))
+
+#endif /* __LINUX_OVERFLOW_H */
diff --git a/tools/include/tools/libc_compat.h b/tools/include/tools/libc_compat.h
new file mode 100644 (file)
index 0000000..664ced8
--- /dev/null
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: GPL-2.0+
+/* Copyright (C) 2018 Netronome Systems, Inc. */
+
+#ifndef __TOOLS_LIBC_COMPAT_H
+#define __TOOLS_LIBC_COMPAT_H
+
+#include <stdlib.h>
+#include <linux/overflow.h>
+
+#ifdef COMPAT_NEED_REALLOCARRAY
+static inline void *reallocarray(void *ptr, size_t nmemb, size_t size)
+{
+       size_t bytes;
+
+       if (unlikely(check_mul_overflow(nmemb, size, &bytes)))
+               return NULL;
+       return realloc(ptr, bytes);
+}
+#endif
+#endif
index 5390e7725e4306408bbe5078652508f09263b0df..7a8e4c98ef1a681922bb91745d5e99b3a5b25777 100644 (file)
@@ -66,7 +66,7 @@ ifndef VERBOSE
 endif
 
 FEATURE_USER = .libbpf
-FEATURE_TESTS = libelf libelf-getphdrnum libelf-mmap bpf
+FEATURE_TESTS = libelf libelf-getphdrnum libelf-mmap bpf reallocarray
 FEATURE_DISPLAY = libelf bpf
 
 INCLUDES = -I. -I$(srctree)/tools/include -I$(srctree)/tools/arch/$(ARCH)/include/uapi -I$(srctree)/tools/include/uapi -I$(srctree)/tools/perf
@@ -120,6 +120,10 @@ ifeq ($(feature-libelf-getphdrnum), 1)
   override CFLAGS += -DHAVE_ELF_GETPHDRNUM_SUPPORT
 endif
 
+ifeq ($(feature-reallocarray), 0)
+  override CFLAGS += -DCOMPAT_NEED_REALLOCARRAY
+endif
+
 # Append required CFLAGS
 override CFLAGS += $(EXTRA_WARNINGS)
 override CFLAGS += -Werror -Wall
index f732237610e56ef1ea46f8c6e2138d6bed312120..fd2c4433863d07fc1ced9499b928f9066985e577 100644 (file)
@@ -22,6 +22,7 @@
  * License along with this program; if not,  see <http://www.gnu.org/licenses>
  */
 
+#define _GNU_SOURCE
 #include <stdlib.h>
 #include <stdio.h>
 #include <stdarg.h>
@@ -41,6 +42,7 @@
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/vfs.h>
+#include <tools/libc_compat.h>
 #include <libelf.h>
 #include <gelf.h>
 
@@ -321,7 +323,7 @@ bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
        progs = obj->programs;
        nr_progs = obj->nr_programs;
 
-       progs = realloc(progs, sizeof(progs[0]) * (nr_progs + 1));
+       progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
        if (!progs) {
                /*
                 * In this case the original obj->programs
@@ -822,8 +824,8 @@ static int bpf_object__elf_collect(struct bpf_object *obj)
                                continue;
                        }
 
-                       reloc = realloc(reloc,
-                                       sizeof(*obj->efile.reloc) * nr_reloc);
+                       reloc = reallocarray(reloc, nr_reloc,
+                                            sizeof(*obj->efile.reloc));
                        if (!reloc) {
                                pr_warning("realloc failed\n");
                                err = -ENOMEM;
@@ -1115,7 +1117,7 @@ bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
                        return -LIBBPF_ERRNO__RELOC;
                }
                new_cnt = prog->insns_cnt + text->insns_cnt;
-               new_insn = realloc(prog->insns, new_cnt * sizeof(*insn));
+               new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
                if (!new_insn) {
                        pr_warning("oom in prog realloc\n");
                        return -ENOMEM;