Re: [PATCH][BZ #16398] Fix infinite loop in ftell when writing wide char data

[Siddhesh Poyarekar <siddhesh at redhat dot com>]

Thanks, this is what I finally committed.

Siddhesh


diff --git a/NEWS b/NEWS
index 7998c75..2c8f4f9 100644
--- a/NEWS
+++ b/NEWS
@@ -25,8 +25,8 @@ Version 2.19
   16150, 16151, 16153, 16167, 16169, 16172, 16195, 16214, 16245, 16271,
   16274, 16283, 16289, 16293, 16314, 16316, 16330, 16337, 16338, 16356,
   16365, 16366, 16369, 16372, 16375, 16379, 16384, 16385, 16386, 16387,
-  16390, 16394, 16400, 16407, 16408, 16414, 16430, 16431, 16453, 16474,
-  16506, 16510
+  16390, 16394, 16398, 16400, 16407, 16408, 16414, 16430, 16431, 16453,
+  16474, 16506, 16510
 
 * Slovenian translations for glibc messages have been contributed by the
   Translation Project's Slovenian team of translators.
diff --git a/libio/Makefile b/libio/Makefile
index 05432f4..747a779 100644
--- a/libio/Makefile
+++ b/libio/Makefile
@@ -60,7 +60,7 @@ tests = tst_swprintf tst_wprintf tst_swscanf tst_wscanf tst_getwc tst_putwc   \
 	tst-wmemstream1 tst-wmemstream2 \
 	bug-memstream1 bug-wmemstream1 \
 	tst-setvbuf1 tst-popen1 tst-fgetwc bug-wsetpos tst-fseek \
-	tst-fwrite-error
+	tst-fwrite-error tst-ftell-partial-wide
 ifeq (yes,$(build-shared))
 # Add test-fopenloc only if shared library is enabled since it depends on
 # shared localedata objects.
diff --git a/libio/tst-ftell-partial-wide.c b/libio/tst-ftell-partial-wide.c
new file mode 100644
index 0000000..3734e77
--- /dev/null
+++ b/libio/tst-ftell-partial-wide.c
@@ -0,0 +1,107 @@
+/* Verify that ftell does not go into an infinite loop when a conversion fails
+   due to insufficient space in the buffer.
+   Copyright (C) 2014 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <wchar.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <locale.h>
+#include <errno.h>
+#include <unistd.h>
+
+static int do_test (void);
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"
+
+/* Arbitrary number large enough so that the target buffer during conversion is
+   not large enough.  */
+#define STRING_SIZE (1400)
+#define NSTRINGS (2)
+
+static int
+do_test (void)
+{
+  FILE *fp = NULL;
+  wchar_t *inputs[NSTRINGS] = {NULL};
+  int ret = 1;
+
+  if (setlocale (LC_ALL, "en_US.UTF-8") == NULL)
+    {
+      printf ("Cannot set en_US.UTF-8 locale.\n");
+      goto out;
+    }
+
+
+  /* Generate input from one character, chosen because it has an odd number of
+     bytes in UTF-8, making it easier to reproduce the problem:
+
+     NAME    Hiragana letter GO
+     CHAR    ã
+     UTF-8   E38194
+     UCS     3054
+     MARC-8  692434  */
+  wchar_t seed = L'ã';
+  for (int i = 0; i < NSTRINGS; i++)
+    {
+      inputs[i] = malloc (STRING_SIZE * sizeof (wchar_t));
+      if (inputs[i] == NULL)
+	{
+	  printf ("Failed to allocate memory for inputs: %m\n");
+	  goto out;
+	}
+      wmemset (inputs[i], seed, STRING_SIZE - 1);
+      inputs[i][STRING_SIZE - 1] = L'\0';
+    }
+
+  char *filename;
+  int fd = create_temp_file ("tst-fseek-wide-partial.out", &filename);
+
+  if (fd == -1)
+    {
+      printf ("create_temp_file: %m\n");
+      goto out;
+    }
+
+  fp = fdopen (fd, "w+");
+  if (fp == NULL)
+    {
+      printf ("fopen: %m\n");
+      close (fd);
+      goto out;
+    }
+
+  for (int i = 0; i < NSTRINGS; i++)
+    {
+      printf ("offset: %ld\n", ftell (fp));
+      if (fputws (inputs[i], fp) == -1)
+	{
+	  perror ("fputws");
+	  goto out;
+	}
+    }
+  ret = 0;
+
+out:
+  if (fp != NULL)
+    fclose (fp);
+  for (int i = 0; i < NSTRINGS; i++)
+    free (inputs[i]);
+
+  return ret;
+}
diff --git a/libio/wfileops.c b/libio/wfileops.c
index 87d3cdc..877fc1f 100644
--- a/libio/wfileops.c
+++ b/libio/wfileops.c
@@ -715,7 +715,7 @@ _IO_wfile_seekoff (fp, offset, dir, mode)
 		       - fp->_wide_data->_IO_write_base) / clen;
 	  else
 	    {
-	      enum __codecvt_result status;
+	      enum __codecvt_result status = __codecvt_ok;
 	      delta = (fp->_wide_data->_IO_write_ptr
 		       - fp->_wide_data->_IO_write_base);
 	      const wchar_t *write_base = fp->_wide_data->_IO_write_base;
@@ -728,9 +728,12 @@ _IO_wfile_seekoff (fp, offset, dir, mode)
 		 flush buffers for every ftell.  */
 	      do
 		{
-		  /* Ugh, no point trying to avoid the flush.  Just do it
-		     and go back to how it was with the read mode.  */
-		  if (delta > 0 && new_write_ptr == fp->_IO_buf_end)
+		  /* There is not enough space in the buffer to do the entire
+		     conversion, so there is no point trying to avoid the
+		     buffer flush.  Just do it and go back to how it was with
+		     the read mode.  */
+		  if (status == __codecvt_partial
+		      || (delta > 0 && new_write_ptr == fp->_IO_buf_end))
 		    {
 		      if (_IO_switch_to_wget_mode (fp))
 			return WEOF;