The find utility uses a hardcoded value of 32 * 1024 as the limit of
the command-line length when calling 'find -exec ... {} +'. This results
in over 4 times more execve() calls than in coreutils' find.

This patch proposes to use the limit defined in system headers.

function                                             old     new   delta
bb_sc_arg_max                                          -      16     +16
find_main                                            480     492     +12
func_exec                                            152     156      +4
xargs_main                                           923     909     -14
------------------------------------------------------------------------------
(add/remove: 2/0 grow/shrink: 2/1 up/down: 32/-14)             Total: 18 bytes
   text    data     bss     dec     hex filename
 815849    4123    9504  829476   ca824 busybox_old
 815867    4123    9504  829494   ca836 busybox_unstripped

Signed-off-by: Bartosz Golaszewski <[email protected]>
---
 findutils/find.c  |  4 +++-
 findutils/xargs.c | 23 ++++++++---------------
 include/libbb.h   |  3 +++
 libbb/Kbuild.src  |  1 +
 libbb/sysconf.c   | 30 ++++++++++++++++++++++++++++++
 5 files changed, 45 insertions(+), 16 deletions(-)
 create mode 100644 libbb/sysconf.c

diff --git a/findutils/find.c b/findutils/find.c
index 8ac3da7..6ffcbab 100644
--- a/findutils/find.c
+++ b/findutils/find.c
@@ -419,6 +419,7 @@ struct globals {
        smallint need_print;
        smallint xdev_on;
        recurse_flags_t recurse_flags;
+       IF_FEATURE_FIND_EXEC_PLUS(long max_argv_len;)
 } FIX_ALIASING;
 #define G (*(struct globals*)&bb_common_bufsiz1)
 #define INIT_G() do { \
@@ -428,6 +429,7 @@ struct globals {
        /* we have to zero it out because of NOEXEC */ \
        memset(&G, 0, sizeof(G)); \
        IF_FEATURE_FIND_MAXDEPTH(G.minmaxdepth[1] = INT_MAX;) \
+       IF_FEATURE_FIND_EXEC_PLUS(G.max_argv_len = bb_sc_arg_max();) \
        G.need_print = 1; \
        G.recurse_flags = ACTION_RECURSE; \
 } while (0)
@@ -672,7 +674,7 @@ ACTF(exec)
                int rc = 0;
 
                /* If we have lots of files already, exec the command */
-               if (ap->file_len >= 32*1024)
+               if (ap->file_len >= G.max_argv_len)
                        rc = do_exec(ap, NULL);
 
                ap->file_len += strlen(fileName) + sizeof(char*) + 1;
diff --git a/findutils/xargs.c b/findutils/xargs.c
index 0ba5b56..67e7a66 100644
--- a/findutils/xargs.c
+++ b/findutils/xargs.c
@@ -523,11 +523,7 @@ int xargs_main(int argc, char **argv)
                argc++;
        }
 
-       /* -s NUM default. fileutils-4.4.2 uses 128k, but I heasitate
-        * to use such a big value - first need to change code to use
-        * growable buffer instead of fixed one.
-        */
-       n_max_chars = 32 * 1024;
+       n_max_chars = bb_sc_arg_max(); /* previously hardcoded to 32 * 1024 */
        /* Make smaller if system does not allow our default value.
         * The Open Group Base Specifications Issue 6:
         * "The xargs utility shall limit the command line length such that
@@ -536,16 +532,13 @@ int xargs_main(int argc, char **argv)
         * in the System Interfaces volume of IEEE Std 1003.1-2001)
         * shall not exceed {ARG_MAX}-2048 bytes".
         */
-       {
-               long arg_max = 0;
-#if defined _SC_ARG_MAX
-               arg_max = sysconf(_SC_ARG_MAX) - 2048;
-#elif defined ARG_MAX
-               arg_max = ARG_MAX - 2048;
-#endif
-               if (arg_max > 0 && n_max_chars > arg_max)
-                       n_max_chars = arg_max;
-       }
+
+       /* TODO Introduce a growable buffer and use bb_sc_arg_max() to
+        * determine a safe value for n_max_chars. Remove this check afterwards.
+        */
+       if (n_max_chars > (32 * 1024))
+               n_max_chars = 32 * 1024;
+
        if (opt & OPT_UPTO_SIZE) {
                n_max_chars = xatou_range(max_chars, 1, INT_MAX);
        }
diff --git a/include/libbb.h b/include/libbb.h
index a1a0dc1..3e645d6 100644
--- a/include/libbb.h
+++ b/include/libbb.h
@@ -731,6 +731,9 @@ extern void *xmalloc_open_read_close(const char *filename, 
size_t *maxsz_p) FAST
 /* Never returns NULL */
 extern void *xmalloc_xopen_read_close(const char *filename, size_t *maxsz_p) 
FAST_FUNC RETURNS_MALLOC;
 
+/* sysconf() wrappers */
+long bb_sc_arg_max(void) FAST_FUNC;
+
 #define SEAMLESS_COMPRESSION (0 \
  || ENABLE_FEATURE_SEAMLESS_XZ \
  || ENABLE_FEATURE_SEAMLESS_LZMA \
diff --git a/libbb/Kbuild.src b/libbb/Kbuild.src
index 6578d11..62680bd 100644
--- a/libbb/Kbuild.src
+++ b/libbb/Kbuild.src
@@ -92,6 +92,7 @@ lib-y += skip_whitespace.o
 lib-y += speed_table.o
 lib-y += str_tolower.o
 lib-y += strrstr.o
+lib-y += sysconf.o
 lib-y += time.o
 lib-y += trim.o
 lib-y += u_signal_names.o
diff --git a/libbb/sysconf.c b/libbb/sysconf.c
new file mode 100644
index 0000000..3635c43
--- /dev/null
+++ b/libbb/sysconf.c
@@ -0,0 +1,30 @@
+/* vi: set sw=4 ts=4: */
+/*
+ * Various system configuration helpers.
+ *
+ * Copyright (C) 2014 Bartosz Golaszewski <[email protected]>
+ *
+ * Licensed under GPLv2 or later, see file LICENSE in this source tree.
+ */
+
+#include "libbb.h"
+
+/*
+ * Return the maximum command line length.
+ *
+ * In case neither _SC_ARG_MAX nor ARG_MAX macros are defined
+ * we resort to using a small (ie. safe) default of 32 * 1024.
+ *
+ * POSIX suggests to substract 2048 bytes from sysconf(_SC_ARG_MAX)
+ * so that the process may safely modify its environment.
+ */
+long FAST_FUNC bb_sc_arg_max(void)
+{
+#if defined _SC_ARG_MAX
+       return sysconf(_SC_ARG_MAX) - 2048;
+#elif defined ARG_MAX
+       return ARG_MAX - 2048;
+#else
+       return 32 * 1024;
+#endif
+}
-- 
1.9.1

_______________________________________________
busybox mailing list
[email protected]
http://lists.busybox.net/mailman/listinfo/busybox

Reply via email to