[PATCH] Small bug fix for aio

[PATCH] Small bug fix for aio

[Mingming Cao]

[-- Attachment #1: Type: text/plain, Size: 199 bytes --]

Hi,

Here is a trivial patch fixed a bug in ioctx_alloc(). If
aio_setup_ring() failed, ioctx_alloc() should pass the return error from
aio_setup_ring() back to sys_io_setup().

Please apply. Thanks.

[-- Attachment #2: aiofix.patch --]
[-- Type: text/x-patch, Size: 887 bytes --]

diff -urNp linux-2.5.68/fs/aio.c 2568/fs/aio.c
--- linux-2.5.68/fs/aio.c	Sat Apr 19 19:49:25 2003
+++ 2568/fs/aio.c	Tue Apr 22 17:52:21 2003
@@ -204,6 +204,7 @@ static struct kioctx *ioctx_alloc(unsign
 {
 	struct mm_struct *mm;
 	struct kioctx *ctx;
+	int ret = 0;
 
 	/* Prevent overflows */
 	if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
@@ -233,7 +234,8 @@ static struct kioctx *ioctx_alloc(unsign
 	INIT_LIST_HEAD(&ctx->run_list);
 	INIT_WORK(&ctx->wq, aio_kick_handler, ctx);
 
-	if (aio_setup_ring(ctx) < 0)
+	ret = aio_setup_ring(ctx);
+	if (unlikely(ret < 0))
 		goto out_freectx;
 
 	/* limit the number of system wide aios */
@@ -259,7 +261,7 @@ out_cleanup:
 
 out_freectx:
 	kmem_cache_free(kioctx_cachep, ctx);
-	ctx = ERR_PTR(-ENOMEM);
+	ctx = ERR_PTR(ret);
 
 	dprintk("aio: error allocating ioctx %p\n", ctx);
 	return ctx;

Re: [PATCH] Small bug fix for aio

[Benjamin LaHaise]

On Tue, Apr 22, 2003 at 06:54:51PM -0700, Mingming Cao wrote:
> Here is a trivial patch fixed a bug in ioctx_alloc(). If
> aio_setup_ring() failed, ioctx_alloc() should pass the return error from
> aio_setup_ring() back to sys_io_setup().

This particular case was intentional: -ENOMEM really is the right return 
code when the ring cannot be allocated.  Mixing it up with the potential 
return codes from do_mmap or other functions might result in a very 
nonsensical return.  Perhaps someone can provide an arguement in favour 
of propagating the value, but I think it really can only mean "you asked 
the kernel to allocate too much memory or did something stupidly buggy with 
threads".  Cheers,

		-ben
-- 
Junk email?  <a href="mailto:aart@kvack.org">aart@kvack.org</a>

Re: [PATCH] Small bug fix for aio

[Mingming Cao]

On Wed, 2003-04-23 at 20:14, Benjamin LaHaise wrote:
> On Tue, Apr 22, 2003 at 06:54:51PM -0700, Mingming Cao wrote:
> > Here is a trivial patch fixed a bug in ioctx_alloc(). If
> > aio_setup_ring() failed, ioctx_alloc() should pass the return error from
> > aio_setup_ring() back to sys_io_setup().
> 
> This particular case was intentional: -ENOMEM really is the right return 
> code when the ring cannot be allocated.  Mixing it up with the potential 
> return codes from do_mmap or other functions might result in a very 
> nonsensical return.  Perhaps someone can provide an arguement in favour 
> of propagating the value, but I think it really can only mean "you asked 
> the kernel to allocate too much memory or did something stupidly buggy with 
> threads".  Cheers,
> 
> 		-ben

Ok, then I think -ENOMEM makes more sense. Thanks for clarifying.


Mingming