Skip to contentSkip to conversation

aio: Fix type of iterator variable in do_io_submit()

PATCH In review Eric Biggers 3 participants 4 messages Apr 22, 2014 – Apr 23, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 aio: Fix type of iterator variable in do_io_submit()

do_io_submit() iterated over the userspace iocb structure pointers using a variable i of type 'int'. This was wrong since 'nr', the number of iocb structure pointers, could potentially be up to LONG_MAX / sizeof(struct iocb *). Fix it (and also remove the unnecessary initialization to 0).

---
 fs/aio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/aio.c +1 -1
index 12a3de0e..4c96af7 100644
@@ -1441,7 +1441,7 @@ long do_io_submit(aio_context_t ctx_id, long nr,
 {
 	struct kioctx *ctx;
 	long ret = 0;
-	int i = 0;
+	long i;
 	struct blk_plug plug;
 
 	if (unlikely(nr < 0))
Signature · 7 lines
1.9.2

--
To unsubscribe, send a message with 'unsubscribe linux-aio' in
the body to [email protected].  For more info on Linux AIO,
see: http://www.kvack.org/aio/
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
#1 lore raw
Quoted 5 lines · Eric Biggers

do_io_submit() iterated over the userspace iocb structure pointers using a variable i of type 'int'. This was wrong since 'nr', the number of iocb structure pointers, could potentially be up to LONG_MAX / sizeof(struct iocb *). Fix it (and also remove the unnecessary initialization to 0).

You're not wrong, but do we *really* want users to be able to submit 144115188075855872 I/Os with a single system call? How about limiting them to a single billion? Given that they have to allocate 64GB of *control* data structures to submit this many I/Os, I think this will be sufficient for many years to come.

--
To unsubscribe, send a message with 'unsubscribe linux-aio' in the body to [email protected]. For more info on Linux AIO,
see: http://www.kvack.org/aio/
Don't email: <a href=mailto:"[email protected]">[email protected]</a>

maintainer
#2 lore raw
Quoted 7 lines · Matthew Wilcox
Quoted 5 lines · Eric Biggers

do_io_submit() iterated over the userspace iocb structure pointers using a variable i of type 'int'. This was wrong since 'nr', the number of iocb structure pointers, could potentially be up to LONG_MAX / sizeof(struct iocb *). Fix it (and also remove the unnecessary initialization to 0).

Quoted 5 lines

You're not wrong, but do we *really* want users to be able to submit 144115188075855872 I/Os with a single system call? How about limiting them to a single billion? Given that they have to allocate 64GB of *control* data structures to submit this many I/Os, I think this will be sufficient for many years to come.

Practically speaking, this change has no effect. The io_submit() syscall will exit far before we even hit INT_MAX because of the limits on the number of iocbs.

-ben

Signature · 7 lines
"Thought is the essence of where you are now."

--
To unsubscribe, send a message with 'unsubscribe linux-aio' in
the body to [email protected].  For more info on Linux AIO,
see: http://www.kvack.org/aio/
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
#3 lore raw
Quoted 3 lines · Benjamin LaHaise

Practically speaking, this change has no effect. The io_submit() syscall will exit far before we even hit INT_MAX because of the limits on the number of iocbs.

Yes it looks like it doesn't actually make a difference due to the default 'aio-max-nr' limit of 1048576 (although you actually can submit just over twice this many). In my opinion this change should still be made so that the correctness of the code doesn't rely on that nonlocal assumption, however. And/or the explicit limit of (LONG_MAX / sizeof(struct iocb *)) elements could be changed to something lower, as Matthew suggested.

Eric

--
To unsubscribe, send a message with 'unsubscribe linux-aio' in the body to [email protected]. For more info on Linux AIO,
see: http://www.kvack.org/aio/
Don't email: <a href=mailto:"[email protected]">[email protected]</a>