Re: bug#35137: [df] incorrect parsing of /proc/self/mountinfo with \r in mount path

[Zbigniew Jędrzejewski-Szmek]

Hi,

I don't know this codebase, so can't comment on the patch, but the
same bug in util-linux was solved by ditching scanf.

https://github.com/karelzak/util-linux/commit/e902609400a861dbdb47d5c3eb98b951530bf01d
https://github.com/karelzak/util-linux/commit/e3782bf6776dcef329b09f4324e1be680f690f3c

Zbyszek


On Tue, Apr 09, 2019 at 07:15:04PM -0700, Paul Eggert wrote:
> Bernhard Voelker wrote:
> 
> +/* Find the next white space in STR, terminate the string there in place,
> +   and return that position.  Otherwise return NULL.  */
> +
> +static char *
> +terminate_at_blank (char const *str)
> +{
> +  char *s = NULL;
> +  if ((s = strchr (str, ' ')) != NULL)
> +*s = '\0';
> +  return s;
> +}
> 
> Since the function modifies its argument, the argument type should
> be char *, not char const *. Also, the code has an assignment in an
> 'if' conditional and the comment is not quite right. Better is:
> 
>   /* Find the next space in STR, terminate the string there in place,
>  and return that position.  Otherwise return NULL.  */
> 
>   static char *
>   terminate_at_blank (char *str)
>   {
> char *s = strchr (str, ' ');
> if (s)
>   *s = '\0';
> return s;
>   }
> 
> >+if (! (blank = terminate_at_blank (mntroot)))
> >+  continue;
> 
> Avoid assignments in 'if' conditionals. Better is:
> 
> blank = terminate_at_blank (target);
> if (! blank)
>   continue;
> 
> +if (*source == ' ')
> +  {
> +/* The source is an empty string, which is e.g. allowed for
> +   tmpfs: "mount -t tmpfs '' /mnt".  */
> +*source = '\0';
> +  }
> +else
> +  {
> +if (! (blank = terminate_at_blank (source)))
> +  continue;
> +  }
> 
> Since 'blank' is not used later, surely these 11 lines of code can
> be simplified to 2 lines:
> 
> if (! terminate_at_blank (source))
>   continue;
> 
> >+int mntroot_s;
> >+char *mntroot, *blank, *target, *dash, *fstype, *source;
> 
> I suggest using C99-style declaration-after-statement style rather
> than this old-fashioned C89-style declarations-at-start-of-block
> style, just for the changed part of the code anyway.
> 

Re: bug#35137: [df] incorrect parsing of /proc/self/mountinfo with \r in mount path

[Bernhard Voelker]

On 4/10/19 4:15 AM, Paul Eggert wrote:
> Bernhard Voelker wrote:
> 
> +/* Find the next white space in STR, terminate the string there in place,
> +   and return that position.  Otherwise return NULL.  */
> +
> +static char *
> +terminate_at_blank (char const *str)
> +{
> +  char *s = NULL;
> +  if ((s = strchr (str, ' ')) != NULL)
> +*s = '\0';
> +  return s;
> +}
> 
> Since the function modifies its argument, the argument type should be char *, 
> not char const *. Also, the code has an assignment in an 'if' conditional and 
> the comment is not quite right. Better is:
> 
>/* Find the next space in STR, terminate the string there in place,
>   and return that position.  Otherwise return NULL.  */
> 
>static char *
>terminate_at_blank (char *str)
>{
>  char *s = strchr (str, ' ');
>  if (s)
>*s = '\0';
>  return s;
>}
> 
>> +if (! (blank = terminate_at_blank (mntroot)))
>> +  continue;
> 
> Avoid assignments in 'if' conditionals. Better is:
> 
>  blank = terminate_at_blank (target);
>  if (! blank)
>continue;
> 
> +if (*source == ' ')
> +  {
> +/* The source is an empty string, which is e.g. allowed for
> +   tmpfs: "mount -t tmpfs '' /mnt".  */
> +*source = '\0';
> +  }
> +else
> +  {
> +if (! (blank = terminate_at_blank (source)))
> +  continue;
> +  }
> 
> Since 'blank' is not used later, surely these 11 lines of code can be 
> simplified 
> to 2 lines:
> 
>  if (! terminate_at_blank (source))
>continue;
> 
>> +int mntroot_s;
>> +char *mntroot, *blank, *target, *dash, *fstype, *source;
> 
> I suggest using C99-style declaration-after-statement style rather than this 
> old-fashioned C89-style declarations-at-start-of-block style, just for the 
> changed part of the code anyway.

Thanks for the review.
Pushed with all these changes at:
https://git.sv.gnu.org/cgit/gnulib.git/commit/?id=eb8278fefa

For coreutils, I'll push the (attached) gnulib update with a NEWS entry soon,
and then work on tests.

Have a nice day,
Berny

>From b938a1badf672f8168daf71fd751e947877c9fc7 Mon Sep 17 00:00:00 2001
From: Bernhard Voelker 
Date: Wed, 10 Apr 2019 09:13:27 +0200
Subject: [PATCH] gnulib: update to the latest

* gnulib: Update to latest, mainly for:
  > mountlist: make parsing /proc/self/mountinfo more robust
* NEWS: Mention the fix.

Fixes https://bugs.gnu.org/33468
---
 NEWS   | 7 +++
 gnulib | 2 +-
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/NEWS b/NEWS
index 6844228be..132f2a0f3 100644
--- a/NEWS
+++ b/NEWS
@@ -4,6 +4,13 @@ GNU coreutils NEWS-*- outline -*-
 
 ** Bug fixes
 
+  df now correctly parses the /proc/self/mountinfo file for unusual entries
+  with '\r' in a field value ("mount -t tmpfs tmpfs /foo$'\r'bar"),
+  when the source field is empty ('mount -t tmpfs "" /mnt'), and when the
+  filesystem type contains characters like blank which need escaping.
+  [bugs introduced in coreutils-8.24 with the introduction of reading
+   the /proc/self/mountinfo file]
+
   factor again outputs immediately when stdout is a tty but stdin is not.
   [bug introduced in coreutils-8.24]
 
diff --git a/gnulib b/gnulib
index 188d87b05..eb8278fef 16
--- a/gnulib
+++ b/gnulib
@@ -1 +1 @@
-Subproject commit 188d87b05190690d6f8b0577ec65ef221a711d08
+Subproject commit eb8278fefa0bbf2a53b706bffb2c99ccfe5d7bd4
-- 
2.21.0

Re: bug#35137: [df] incorrect parsing of /proc/self/mountinfo with \r in mount path

[Paul Eggert]

Bernhard Voelker wrote:

+/* Find the next white space in STR, terminate the string there in place,
+   and return that position.  Otherwise return NULL.  */
+
+static char *
+terminate_at_blank (char const *str)
+{
+  char *s = NULL;
+  if ((s = strchr (str, ' ')) != NULL)
+*s = '\0';
+  return s;
+}

Since the function modifies its argument, the argument type should be char *, 
not char const *. Also, the code has an assignment in an 'if' conditional and 
the comment is not quite right. Better is:


  /* Find the next space in STR, terminate the string there in place,
 and return that position.  Otherwise return NULL.  */

  static char *
  terminate_at_blank (char *str)
  {
char *s = strchr (str, ' ');
if (s)
  *s = '\0';
return s;
  }


+if (! (blank = terminate_at_blank (mntroot)))
+  continue;


Avoid assignments in 'if' conditionals. Better is:

blank = terminate_at_blank (target);
if (! blank)
  continue;

+if (*source == ' ')
+  {
+/* The source is an empty string, which is e.g. allowed for
+   tmpfs: "mount -t tmpfs '' /mnt".  */
+*source = '\0';
+  }
+else
+  {
+if (! (blank = terminate_at_blank (source)))
+  continue;
+  }

Since 'blank' is not used later, surely these 11 lines of code can be simplified 
to 2 lines:


if (! terminate_at_blank (source))
  continue;


+int mntroot_s;
+char *mntroot, *blank, *target, *dash, *fstype, *source;


I suggest using C99-style declaration-after-statement style rather than this 
old-fashioned C89-style declarations-at-start-of-block style, just for the 
changed part of the code anyway.

Re: bug#35137: [df] incorrect parsing of /proc/self/mountinfo with \r in mount path

[Bernhard Voelker]

[adding gnulib, coming from https://bugs.gnu.org/35137]

On 4/5/19 9:01 AM, Bernhard Voelker wrote:
> On 4/4/19 9:52 AM, Zbigniew Jędrzejewski-Szmek wrote:
>> See https://github.com/systemd/systemd/issues/12018 and
>> https://github.com/karelzak/util-linux/issues/780 for additional context.
>>
>> $ mkdir "$(echo -e foo\\rbar)"
>> $ sudo mount -t tmpfs tmpfs foo^Mbar/
>> $ cat -v /proc/self/mountinfo|grep foo
>> 865 39 0:59 / /tmp/foo^Mbar rw,relatime shared:462 - tmpfs tmpfs rw,seclabel
>> $ df -h | grep foo
>> $ df -h /tmp/foo$'\r'bar
>> Filesystem  Size  Used Avail Use% Mounted on
>> -   3.9G 0  3.9G   0% /tmp/foo?bar
>>
>> When asked to show all filesystems, the mount point is not shown at all.
>> When asked to show just that one, df parses the mount point correctly,
>> but it gets the filesystem type wrong.
> 
> Thanks for the report.
> 
> I see the issue is not yet solved in util-linux as well.
> For coreutils, the fix is in gnulib.  Parsing the line is starting
> to get ugly ... dirty patch attached.
> 
> This also caters for the issue that df(1) totally skips a file system
> if the source is an empty string which is allowed for e.g. tmpfs:
> 
>   $ mount -t tmpfs '' /mnt
>   $ df -h | grep mnt

The attached is a patch for gnulib to fix the reported and two other issues
when parsing /proc/self/mountinfo.
Any objections?

Meanwhile, I'm working a test for coreutils' df.

FWIW: Karel fixed this issue in util-linux yesterday:
https://git.kernel.org/pub/scm/utils/util-linux/util-linux.git/commit/?id=86673b3a46

Have a nice day,
Berny
>From 15be8e6bcb6203fdb91d4a01fb65c2d4ef76995e Mon Sep 17 00:00:00 2001
From: Bernhard Voelker 
Date: Tue, 9 Apr 2019 22:30:16 +0200
Subject: [PATCH] mountlist: make parsing /proc/self/mountinfo more robust
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Cater for the following issues with mountinfo parsing (the first
one was reported by Zbigniew Jędrzejewski-Szmek 
in ).

1. The fields source, target, mntroot and fstype may contain characters
like '\r'; sscanf(3) fails to read such values with the %s format
specifier because it would stop at such characters.
Example: "mount -t tmpfs tmpfs /foo^Mbar".
The only true separator in that file is the ' ' character.

2. The source field may be an empty string, which happens e.g. with
"mount -t tmpfs '' /target".

3. The fstype field may contain mangled characters as well which need
unescaping.

* lib/mountlist.c (terminate_at_blank): Add utility function.
(read_file_system_list): In the block trying to read the mountinfo file,
avoid using sscanf(3) with %s format; instead, parse the above fields
separated by white spaces one by one.
Handle the case when the source field is an empty string.
Unescape the fstype field.
---
 ChangeLog   | 22 +
 lib/mountlist.c | 83 +++--
 2 files changed, 74 insertions(+), 31 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 87d95ed99..e049e81a0 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,25 @@
+2019-04-09  Bernhard Voelker  
+
+	mountlist: make parsing /proc/self/mountinfo more robust
+	Cater for the following issues with mountinfo parsing (the first
+	one was reported by Zbigniew Jędrzejewski-Szmek 
+	in ).
+	1. The fields source, target, mntroot and fstype may contain characters
+	like '\r'; sscanf(3) fails to read such values with the %s format
+	specifier because it would stop at such characters.
+	Example: "mount -t tmpfs tmpfs /foo^Mbar".
+	The only true separator in that file is the ' ' character.
+	2. The source field may be an empty string, which happens e.g. with
+	"mount -t tmpfs '' /target".
+	3. The fstype field may contain mangled characters as well which need
+	unescaping.
+	* lib/mountlist.c (terminate_at_blank): Add utility function.
+	(read_file_system_list): In the block trying to read the mountinfo file,
+	avoid using sscanf(3) with %s format; instead, parse the above fields
+	separated by white spaces one by one.
+	Handle the case when the source field is an empty string.
+	Unescape the fstype field.
+
 2019-04-09  Bruno Haible  
 
 	openmp: Add workaround for 32-bit programs on AIX.
diff --git a/lib/mountlist.c b/lib/mountlist.c
index 9b54a2cf7..9e01d13f4 100644
--- a/lib/mountlist.c
+++ b/lib/mountlist.c
@@ -418,6 +418,18 @@ unescape_tab (char *str)
 str[j++] = str[i];
 }
 }
+
+/* Find the next white space in STR, terminate the string there in place,
+   and return that position.  Otherwise return NULL.  */
+
+static char *
+terminate_at_blank (char const *str)
+{
+  char *s = NULL;
+  if ((s = strchr (str, ' ')) != NULL)
+*s = '\0';
+  return s;
+}
 #endif
 
 /* Return a list of the currently mounted file systems, or NULL on error.
@@ -453,56 +465,65 @@ read_file_system_list (bool need_fs_type)
 while (getline (, _size, fp) != -1)
   {
 unsigned