On Mon, May 26, 2014 at 10:41 AM, Djalal Harouni <tixxdz@xxxxxxxxxx> wrote:
> On Mon, May 26, 2014 at 10:01:20AM -0700, Andy Lutomirski wrote:
>> On Mon, May 26, 2014 at 6:27 AM, Djalal Harouni <tixxdz@xxxxxxxxxx> wrote:
>> > This patch is preparation, it adds a couple of helpers to read data and
>> > to get the cached permission checks during that ->read().
>> >
>> > Currently INF entries share the same code, they do not implement
>> > specific ->open(), only ->read() coupled with callback calls. Doing
>> > permission checks during ->open() will not work and will only disturb
>> > the INF entries that do not need permission checks. Yes not all the INF
>> > entries need checks, the ones that need protections are listed below:
>> > /proc/<pid>/wchan
>> > /proc/<pid>/syscall
>> > /proc/<pid>/{auxv,io} (will be handled in next patches).
>> >
>> > So to have proper permission checks convert this INF entries to REG
>> > entries and use their open() and read() handlers to implement these
>> > checks. To achieve this we add the following helpers:
>> >
>> > * proc_read_from_buffer() a wrapper around simple_read_from_buffer(), it
>> > makes sure that count <= PROC_BLOCK_SIZE (3*1024)
>> >
>> > * pid_entry_read(): it will get a free page and pass it to the specific
>> > /proc/<pid>/$entry handler (callback). The handler is of the same
>> > types of the previous INF handlers, it will only receive an extra
>> > "permitted" argument that contains the cached permission check that
>> > was performed during ->open().
>> >
>> > The handler is of type:
>> > typedef int (*proc_read_fn_t)(char *page,
>> > struct task_struct *task, int permitted);
>>
>> [...]
>>
>> This strikes me as *way* too complicated. Why not just change
>> proc_info_file_operations to *always* cache may_ptrace on open?
> Not all the INF entries need permission checks during open:
> The one that need it:
> /proc/<pid>/{wchan|syscall} converted in this series.
> /proc/<pid>/{auxv|io} (will be converted in next series)
>
> The ones that do not need it:
> /proc/<pid>/{limite|cmdline|shedstat|oom_score|...}
>
> There is no reason to do it for these entries, and if you do it you will
> also have to passe the cached permission checks, so I don't think that
> you want to modify all the INF handlers especially the ones that do not
> need it to take an extra argument.
...but you don't have to pass the checks, because nothing will care
whether you passed them.
> Then you will also modify this:
> in file fs/proc/internal.h the:
> union proc_op {
> ...
> int (*proc_read)(struct task_struct *task, char *page);
> ...
> }
>
> And then you will follow in all other places and handlers... and modify
> the definition of INF entries... It's much more complicated...
>
But the final result will be much less messy and have less duplicated
code. Your patch seems like it duplicates a bunch of code and I
suspect that Al Viro will dislike it if he reads it.
--Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
