This reference at the end of rb_get_reader_page() was causing off-by-one
writes to the prev pointer of the page after the reader page when that
page is the head page, and therefore the reader page has the RB_PAGE_HEAD
flag in its list.next pointer. This eventually results in a GPF in a
subsequent call to rb_set_head_page() (usually from rb_get_reader_page())
when that prev pointer is dereferenced. The dereferenced register would
characteristically have an address that appears shifted left by one byte
(eg, ffxxxxxxxxxxxxyy instead of ffffxxxxxxxxxxxx) due to being written at
an address one byte too high.
Signed-off-by: David Sharp <[email protected]>
---
kernel/trace/ring_buffer.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 2326b04..d5b7308 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -2906,7 +2906,7 @@ rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
*
* Now make the new head point back to the reader page.
*/
- reader->list.next->prev = &cpu_buffer->reader_page->list;
+ rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
/* Finally update the reader page to the new head */
--
1.6.5.3
Just a FYI, it's best to send me email to my [email protected]
account. I don't always read my messages on my redhat account.
On Wed, 2010-01-06 at 17:12 -0800, David Sharp wrote:
> This reference at the end of rb_get_reader_page() was causing off-by-one
> writes to the prev pointer of the page after the reader page when that
> page is the head page, and therefore the reader page has the RB_PAGE_HEAD
> flag in its list.next pointer. This eventually results in a GPF in a
> subsequent call to rb_set_head_page() (usually from rb_get_reader_page())
> when that prev pointer is dereferenced. The dereferenced register would
> characteristically have an address that appears shifted left by one byte
> (eg, ffxxxxxxxxxxxxyy instead of ffffxxxxxxxxxxxx) due to being written at
> an address one byte too high.
>
> Signed-off-by: David Sharp <[email protected]>
> ---
> kernel/trace/ring_buffer.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
> index 2326b04..d5b7308 100644
> --- a/kernel/trace/ring_buffer.c
> +++ b/kernel/trace/ring_buffer.c
> @@ -2906,7 +2906,7 @@ rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
> *
> * Now make the new head point back to the reader page.
> */
> - reader->list.next->prev = &cpu_buffer->reader_page->list;
> + rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
> rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
Wow, this is a nasty little race. The writer had to move the header page
forward by one between the time we swapped to now, which is the command
just outside the diff.
>
> /* Finally update the reader page to the new head */
We probably should add a change at the top too, just to be safe.
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 2326b04..edefe3b 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -2869,7 +2869,7 @@ rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
* Splice the empty reader page into the list around the head.
*/
reader = rb_set_head_page(cpu_buffer);
- cpu_buffer->reader_page->list.next = reader->list.next;
+ cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
cpu_buffer->reader_page->list.prev = reader->list.prev;
/*
Thanks, I'll apply this.
-- Steve
Commit-ID: 5ded3dc6a3c7549b36a8ac27bbd81b33756a2c29
Gitweb: http://git.kernel.org/tip/5ded3dc6a3c7549b36a8ac27bbd81b33756a2c29
Author: David Sharp <[email protected]>
AuthorDate: Wed, 6 Jan 2010 17:12:07 -0800
Committer: Steven Rostedt <[email protected]>
CommitDate: Wed, 6 Jan 2010 20:38:25 -0500
ring-buffer: Wrap a list.next reference with rb_list_head()
This reference at the end of rb_get_reader_page() was causing off-by-one
writes to the prev pointer of the page after the reader page when that
page is the head page, and therefore the reader page has the RB_PAGE_HEAD
flag in its list.next pointer. This eventually results in a GPF in a
subsequent call to rb_set_head_page() (usually from rb_get_reader_page())
when that prev pointer is dereferenced. The dereferenced register would
characteristically have an address that appears shifted left by one byte
(eg, ffxxxxxxxxxxxxyy instead of ffffxxxxxxxxxxxx) due to being written at
an address one byte too high.
Signed-off-by: David Sharp <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Steven Rostedt <[email protected]>
---
kernel/trace/ring_buffer.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 2326b04..d5b7308 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -2906,7 +2906,7 @@ rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
*
* Now make the new head point back to the reader page.
*/
- reader->list.next->prev = &cpu_buffer->reader_page->list;
+ rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
/* Finally update the reader page to the new head */