[PATCH] gdb: Fix "target file /proc/.../cmdline contained unexpected null characters"

[PATCH] gdb: Fix "target file /proc/.../cmdline contained unexpected null characters"

[Ilya Leoshkevich]

cmdline is read with target_fileio_read_stralloc(), which warns on
seeing null characters.  However, it's perfectly valid for cmdline to
contain \0s, so switch to target_fileio_read_alloc().
---
 gdb/linux-tdep.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
index b5eee5e108c..96cbe8e5520 100644
--- a/gdb/linux-tdep.c
+++ b/gdb/linux-tdep.c
@@ -1902,15 +1902,22 @@ linux_fill_prpsinfo (struct elf_internal_linux_prpsinfo *p)
   pid = inferior_ptid.pid ();
   xsnprintf (filename, sizeof (filename), "/proc/%d/cmdline", (int) pid);
   /* The full name of the program which generated the corefile.  */
-  gdb::unique_xmalloc_ptr<char> fname
-    = target_fileio_read_stralloc (NULL, filename);
+  gdb_byte *buf = NULL;
+  size_t buf_len = target_fileio_read_alloc (NULL, filename, &buf);
+  gdb::unique_xmalloc_ptr<char> fname ((char *)buf);
 
-  if (fname == NULL || fname.get ()[0] == '\0')
+  if (buf_len < 1 || fname.get ()[0] == '\0')
     {
       /* No program name was read, so we won't be able to retrieve more
 	 information about the process.  */
       return 0;
     }
+  if (fname.get ()[buf_len - 1] != '\0')
+    {
+      warning (_("target file %s "
+		 "does not contain a trailing null character"),
+	       filename);
+    }
 
   memset (p, 0, sizeof (*p));
 
-- 
2.40.1

Re: [PATCH] gdb: Fix "target file /proc/.../cmdline contained unexpected null characters"

[Bruno Larsen]

On 22/06/2023 01:14, Ilya Leoshkevich via Gdb-patches wrote:
> cmdline is read with target_fileio_read_stralloc(), which warns on
> seeing null characters.  However, it's perfectly valid for cmdline to
> contain \0s, so switch to target_fileio_read_alloc().

Hi! Thanks for working on this.

 From what I understand, GDB commit messages should be written so that 
they make sense even if you don't read the title, so having a slightly 
more descriptive message would be nice.
> ---
>   gdb/linux-tdep.c | 13 ++++++++++---
>   1 file changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
> index b5eee5e108c..96cbe8e5520 100644
> --- a/gdb/linux-tdep.c
> +++ b/gdb/linux-tdep.c
> @@ -1902,15 +1902,22 @@ linux_fill_prpsinfo (struct elf_internal_linux_prpsinfo *p)
>     pid = inferior_ptid.pid ();
>     xsnprintf (filename, sizeof (filename), "/proc/%d/cmdline", (int) pid);
>     /* The full name of the program which generated the corefile.  */
> -  gdb::unique_xmalloc_ptr<char> fname
> -    = target_fileio_read_stralloc (NULL, filename);
> +  gdb_byte *buf = NULL;
> +  size_t buf_len = target_fileio_read_alloc (NULL, filename, &buf);
> +  gdb::unique_xmalloc_ptr<char> fname ((char *)buf);
>   

Looking at the code on linux_info_proc, I see that this approach is 
already used, so it looks like a good idea to me. However, in there, the 
string is slightly sanitized, so that '\0' in the middle of the string 
are turned into ' ', and the final character is always set to '\0'. I 
think you could probably do the same here.

> -  if (fname == NULL || fname.get ()[0] == '\0')
> +  if (buf_len < 1 || fname.get ()[0] == '\0')
>       {
>         /* No program name was read, so we won't be able to retrieve more
>   	 information about the process.  */
>         return 0;
>       }
> +  if (fname.get ()[buf_len - 1] != '\0')
If you don't do the same, at least here the last character should be 
changed. Its pretty dangerous to allow it to pass since like 5 lines 
later this is used as a C-string, so you could get read-past-the-end and 
other nasty problems.

-- 
Cheers,
Bruno

> +    {
> +      warning (_("target file %s "
> +		 "does not contain a trailing null character"),
> +	       filename);
> +    }
>   
>     memset (p, 0, sizeof (*p));
>   

Re: [PATCH] gdb: Fix "target file /proc/.../cmdline contained unexpected null characters"

[Ilya Leoshkevich]

On Tue, 2023-07-18 at 15:03 +0200, Bruno Larsen wrote:
> On 22/06/2023 01:14, Ilya Leoshkevich via Gdb-patches wrote:
> > cmdline is read with target_fileio_read_stralloc(), which warns on
> > seeing null characters.  However, it's perfectly valid for cmdline
> > to
> > contain \0s, so switch to target_fileio_read_alloc().
> 
> Hi! Thanks for working on this.

Thanks for the review!

> 
>  From what I understand, GDB commit messages should be written so
> that 
> they make sense even if you don't read the title, so having a
> slightly 
> more descriptive message would be nice.

Will do.

> > ---
> >   gdb/linux-tdep.c | 13 ++++++++++---
> >   1 file changed, 10 insertions(+), 3 deletions(-)
> > 
> > diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
> > index b5eee5e108c..96cbe8e5520 100644
> > --- a/gdb/linux-tdep.c
> > +++ b/gdb/linux-tdep.c
> > @@ -1902,15 +1902,22 @@ linux_fill_prpsinfo (struct
> > elf_internal_linux_prpsinfo *p)
> >     pid = inferior_ptid.pid ();
> >     xsnprintf (filename, sizeof (filename), "/proc/%d/cmdline",
> > (int) pid);
> >     /* The full name of the program which generated the corefile. 
> > */
> > -  gdb::unique_xmalloc_ptr<char> fname
> > -    = target_fileio_read_stralloc (NULL, filename);
> > +  gdb_byte *buf = NULL;
> > +  size_t buf_len = target_fileio_read_alloc (NULL, filename,
> > &buf);
> > +  gdb::unique_xmalloc_ptr<char> fname ((char *)buf);
> >   
> 
> Looking at the code on linux_info_proc, I see that this approach is 
> already used, so it looks like a good idea to me. However, in there,
> the 
> string is slightly sanitized, so that '\0' in the middle of the
> string 
> are turned into ' ', and the final character is always set to '\0'. I
> think you could probably do the same here.

Here we are interested only in the filename, so ending the C string
at the first '\0' is what we need.

> > -  if (fname == NULL || fname.get ()[0] == '\0')
> > +  if (buf_len < 1 || fname.get ()[0] == '\0')
> >       {
> >         /* No program name was read, so we won't be able to
> > retrieve more
> >          information about the process.  */
> >         return 0;
> >       }
> > +  if (fname.get ()[buf_len - 1] != '\0')
> If you don't do the same, at least here the last character should be 
> changed. Its pretty dangerous to allow it to pass since like 5 lines 
> later this is used as a C-string, so you could get read-past-the-end
> and 
> other nasty problems.

Oh, right - I think I should just return here. /proc/.../cmdline not
having a trailing '\0' must be a kernel bug anyway.