There is absolutely no reason to crash the kernel when we have a
perfectly good return value already available to use for conveying
failure status.
Let's return an error code instead of crashing the kernel: that sounds
like a much better plan.
Signed-off-by: Nick Bowler <[email protected]>
---
lib/scatterlist.c | 3 ++-
1 files changed, 2 insertions(+), 1 deletions(-)
diff --git a/lib/scatterlist.c b/lib/scatterlist.c
index 3675452b23ca..11ecaf000696 100644
--- a/lib/scatterlist.c
+++ b/lib/scatterlist.c
@@ -248,7 +248,8 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents,
unsigned int left;
#ifndef ARCH_HAS_SG_CHAIN
- BUG_ON(nents > max_ents);
+ if (WARN_ON_ONCE(nents > max_ents))
+ return -E2BIG;
#endif
memset(table, 0, sizeof(*table));
--
1.7.8.6
On Thu, 1 Nov 2012 15:03:00 -0400
Nick Bowler <[email protected]> wrote:
> There is absolutely no reason to crash the kernel when we have a
> perfectly good return value already available to use for conveying
> failure status.
Yes, I suppose that's true. I don't see a case for BUGging the kernel
here.
> Let's return an error code instead of crashing the kernel: that sounds
> like a much better plan.
>
> Signed-off-by: Nick Bowler <[email protected]>
> ---
> lib/scatterlist.c | 3 ++-
> 1 files changed, 2 insertions(+), 1 deletions(-)
>
> diff --git a/lib/scatterlist.c b/lib/scatterlist.c
> index 3675452b23ca..11ecaf000696 100644
> --- a/lib/scatterlist.c
> +++ b/lib/scatterlist.c
> @@ -248,7 +248,8 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents,
> unsigned int left;
>
> #ifndef ARCH_HAS_SG_CHAIN
> - BUG_ON(nents > max_ents);
> + if (WARN_ON_ONCE(nents > max_ents))
> + return -E2BIG;
> #endif
OK, pet peeve: if this E2BIG gets returned to userspace, our poor user
will look it up and see "Argument list too long; used when the
arguments passed to a new program being executed with one of the exec
functions occupy too much memory space". He then gets to spend half a
day reviewing his code's exec() callsites!
See? Although the error's name sounds like a nice match to the
internal state, it isn't really a match at all and our use of it is
misleading.
Unfortunately there is no EKERNELSCREWEDUP, so we usually use EINVAL.
On 2012-11-14 13:05 -0800, Andrew Morton wrote:
> On Thu, 1 Nov 2012 15:03:00 -0400
> Nick Bowler <[email protected]> wrote:
>
> > There is absolutely no reason to crash the kernel when we have a
> > perfectly good return value already available to use for conveying
> > failure status.
>
> Yes, I suppose that's true. I don't see a case for BUGging the kernel
> here.
[...]
> > - BUG_ON(nents > max_ents);
> > + if (WARN_ON_ONCE(nents > max_ents))
> > + return -E2BIG;
> > #endif
>
> OK, pet peeve: if this E2BIG gets returned to userspace, our poor user
> will look it up and see "Argument list too long; used when the
> arguments passed to a new program being executed with one of the exec
> functions occupy too much memory space". He then gets to spend half a
> day reviewing his code's exec() callsites!
>
> See? Although the error's name sounds like a nice match to the
> internal state, it isn't really a match at all and our use of it is
> misleading.
>
> Unfortunately there is no EKERNELSCREWEDUP,
Well, maybe we should add it! :P
> so we usually use EINVAL.
Fair enough. I will prepare v2. But perhaps EOPNOTSUPP would be a
better fit?
Thanks,
--
Nick Bowler, Elliptic Technologies (http://www.elliptictech.com/)
On Wed, Nov 14, 2012 at 11:15 PM, Nick Bowler <[email protected]> wrote:
> On 2012-11-14 13:05 -0800, Andrew Morton wrote:
>> On Thu, 1 Nov 2012 15:03:00 -0400
>> Nick Bowler <[email protected]> wrote:
>>
>> > There is absolutely no reason to crash the kernel when we have a
>> > perfectly good return value already available to use for conveying
>> > failure status.
>>
>> Yes, I suppose that's true. I don't see a case for BUGging the kernel
>> here.
> [...]
>> > - BUG_ON(nents > max_ents);
>> > + if (WARN_ON_ONCE(nents > max_ents))
>> > + return -E2BIG;
>> > #endif
>>
>> OK, pet peeve: if this E2BIG gets returned to userspace, our poor user
>> will look it up and see "Argument list too long; used when the
>> arguments passed to a new program being executed with one of the exec
>> functions occupy too much memory space". He then gets to spend half a
>> day reviewing his code's exec() callsites!
>>
>> See? Although the error's name sounds like a nice match to the
>> internal state, it isn't really a match at all and our use of it is
>> misleading.
>>
>> Unfortunately there is no EKERNELSCREWEDUP,
>
> Well, maybe we should add it! :P
>
>> so we usually use EINVAL.
>
> Fair enough. I will prepare v2. But perhaps EOPNOTSUPP would be a
> better fit?
IMHO this would be even more confusing...
--
Thanks,
//richard
On Wed, 14 Nov 2012 17:15:44 -0500
Nick Bowler <[email protected]> wrote:
> > Unfortunately there is no EKERNELSCREWEDUP,
>
> Well, maybe we should add it! :P
Well. Given how frequently it will be used, it would need to be a much
shorter identifier. Perhaps "EEK". Or just "E".
> > so we usually use EINVAL.
>
> Fair enough. I will prepare v2. But perhaps EOPNOTSUPP would be a
> better fit?
I have queued a delta:
From: Andrew Morton <[email protected]>
Subject: scatterlist-dont-bug-when-we-can-trivially-return-a-proper-error-fix
s/E2BIG/EINVAL/
Cc: Nick Bowler <[email protected]>
Cc: Maxim Levitsky <[email protected]>
Cc: Tejun Heo <[email protected]>
Cc: Jens Axboe <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
---
lib/scatterlist.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff -puN lib/scatterlist.c~scatterlist-dont-bug-when-we-can-trivially-return-a-proper-error-fix lib/scatterlist.c
--- a/lib/scatterlist.c~scatterlist-dont-bug-when-we-can-trivially-return-a-proper-error-fix
+++ a/lib/scatterlist.c
@@ -249,7 +249,7 @@ int __sg_alloc_table(struct sg_table *ta
#ifndef ARCH_HAS_SG_CHAIN
if (WARN_ON_ONCE(nents > max_ents))
- return -E2BIG;
+ return -EINVAL;
#endif
memset(table, 0, sizeof(*table));
_