Re: [Patch] Fix up return value from mq_open()

[Jiri Slaby]

On 09/30/2011 04:15 AM, Doug Ledford wrote:
> Commit d40dcdb ipc/mqueue.c: fix mq_open() return value
> changed the return value for mq_open() to return
> 
> EMFILE - The process already has the maximum number of files and
>                message queues open.
> 
> instead of
> 
> ENOMEM - Insufficient memory.
> 
> While the portion of d40dcdb that introduced using ERR_PTR() to
> propagate the condition were helpful, the change to the type of error
> was not.  The specific test that was failing was a memory space test.
> It can fail if there are too many message queues open and this one would
> cause us to exceed the tasks counter for memory bytes capacity (aka, int
> wrap) or it can fail if the total number of bytes already allocated plus
> this allocation will exceed the user's RLIMIT for POSIX message queues.

Yes, then it should not definitely return ENOMEM, right? You exceeded a
limit, not ran out of memory. At least for the latter case.

>  However, depending on the values passed in via an attribute struct,
> this could happen on the 1st file or the 100th file. There is no direct
> relationship to file count in this test.  There is only a direct
> correlation to memory usage.  Therefore return the code to giving ENOMEM
> instead of EMFILE.
...
> commit b80e4058e2b4234fb33a9fd84b4ce288e8ba65cb
> Author: Doug Ledford <dledford@xxxxxxxxxx>
> Date:   Thu Sep 29 19:02:47 2011 -0400
> 
>     ipc/mqueue: make the error returns match the conditions
>     
>     We should not return -EMFILE when the problem is memory usage.

Sorry, I still don't udnerstand, how this test:
  u->mq_bytes + mq_bytes > task_rlimit(p, RLIMIT_MSGQUEUE)
is about running out of memory. I see only a test for exceeding some
quota. Should this be changed, change it to something like ENOSPC.

> diff --git a/ipc/mqueue.c b/ipc/mqueue.c
> index 0474ddb..3e53604 100644
> --- a/ipc/mqueue.c
> +++ b/ipc/mqueue.c
> @@ -165,7 +165,7 @@ static struct inode *mqueue_get_inode(struct super_block *sb,
>  		    u->mq_bytes + mq_bytes > task_rlimit(p, RLIMIT_MSGQUEUE)) {
>  			spin_unlock(&mq_lock);
>  			/* mqueue_evict_inode() releases info->messages */
> -			ret = -EMFILE;
> +			ret = -ENOMEM;

The code now looks like:
int ret = -ENOMEM;
...
if (sth)
  ret = -ENOMEM;
...
return ret;

which is not good.

thanks,
-- 
js
suse labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/