Re: [PATCH 2/3] proc: Update get proc_pid_cmdline() to use mm.h helpers
On Fri, Dec 13, 2013 at 9:23 AM, Stephen Smalley <sds(a)tycho.nsa.gov> wrote:
On 12/02/2013 04:10 PM, William Roberts wrote:
> Re-factor proc_pid_cmdline() to use get_cmdline_length() and
> copy_cmdline() helpers from mm.h
>
> Signed-off-by: William Roberts <wroberts(a)tresys.com>
> ---
> fs/proc/base.c | 35 ++++++++++-------------------------
> 1 file changed, 10 insertions(+), 25 deletions(-)
>
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 03c8d74..fb4eda5 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -203,37 +203,22 @@ static int proc_root_link(struct dentry *dentry, struct path
*path)
> static int proc_pid_cmdline(struct task_struct *task, char * buffer)
> {
> int res = 0;
> - unsigned int len;
> + unsigned int len = 0;
Why? You set len below before first use, so this is redundant.
Yep you're right.
> struct mm_struct *mm = get_task_mm(task);
> if (!mm)
> - goto out;
> - if (!mm->arg_end)
> - goto out_mm; /* Shh! No looking before we're done */
> + return 0;
Equivalent to goto out in the original code, so why change it? Don't
make unnecessary changes.
Also, I think the get_task_mm() ought to move into the helper (or all of
proc_pid_cmdline() should just become the helper). In what situation
will you not be calling get_task_mm() and mmput() around every call to
the helper?
Again my thought on this is to reduce get_task_mm() and mmput() calls.
How expensive are they, etc. However, just to recap the other email. If we
move to saying the audit cache of this can be capped at PATH_MAX even
if it results in some wasted memory, we can just take the original procfs code
and add a length argument.
>
> - len = mm->arg_end - mm->arg_start;
> -
> + len = get_cmdline_length(mm);
> + if (!len)
> + goto mm_out;
Could be moved into the helper. Not sure how the inline function helps
readability or maintainability.
Sure... mostly for readability.
> +
> + /*The caller of this allocates a page */
> if (len > PAGE_SIZE)
> len = PAGE_SIZE;
If the capping of len is handled by the caller, then pass an int to your
helper rather than an unsigned int to avoid problems later with
access_process_vm().
Ok... just weird that lengths are signed to me. when will you ever have negative
space?
> -out_mm:
> +
> + res = copy_cmdline(task, mm, buffer, len);
> +mm_out:
> mmput(mm);
Odd style. If there is only one exit path, just call it out; if there
are two, keep them as out_mm and out.
Yes their is only 1 jmp label. Your right, this is odd.
> -out:
> return res;
> }
>
>
--
Respectfully,
William C Roberts