2009-04-03 07:31:19

by Li Zefan

[permalink] [raw]
Subject: [PATCH] blktrace: fix a bug in blk_msg_write()

This is another long-standing bug.


================


(console 1)
# echo -n 'a' > /sys/kernel/debug/block/sda/msg
(console 2)
# blktrace -d /dev/sda -a pc -o - | blkparse -i -
8,0 0 0 0.000000000 0 m N a������@��

We should terminate the msg buffer with '\0'.

Signed-off-by: Li Zefan <[email protected]>
---
kernel/trace/blktrace.c | 5 +++--
1 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index c0ef70d..2bf341f 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -327,10 +327,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
char *msg;
struct blk_trace *bt;

- if (count > BLK_TN_MAX_MSG)
+ if (count >= BLK_TN_MAX_MSG)
return -EINVAL;

- msg = kmalloc(count, GFP_KERNEL);
+ msg = kmalloc(count + 1, GFP_KERNEL);
if (msg == NULL)
return -ENOMEM;

@@ -338,6 +338,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
kfree(msg);
return -EFAULT;
}
+ msg[count] = '\0';

bt = filp->private_data;
__trace_note_message(bt, "%s", msg);
--
1.5.4.rc3


2009-04-03 11:18:22

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATCH] blktrace: fix a bug in blk_msg_write()


* Li Zefan <[email protected]> wrote:

> This is another long-standing bug.
>
> ================
>
>
> (console 1)
> # echo -n 'a' > /sys/kernel/debug/block/sda/msg
> (console 2)
> # blktrace -d /dev/sda -a pc -o - | blkparse -i -
> 8,0 0 0 0.000000000 0 m N a������@��
>
> We should terminate the msg buffer with '\0'.
>
> Signed-off-by: Li Zefan <[email protected]>
> ---
> kernel/trace/blktrace.c | 5 +++--
> 1 files changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> index c0ef70d..2bf341f 100644
> --- a/kernel/trace/blktrace.c
> +++ b/kernel/trace/blktrace.c
> @@ -327,10 +327,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> char *msg;
> struct blk_trace *bt;
>
> - if (count > BLK_TN_MAX_MSG)
> + if (count >= BLK_TN_MAX_MSG)
> return -EINVAL;
>
> - msg = kmalloc(count, GFP_KERNEL);
> + msg = kmalloc(count + 1, GFP_KERNEL);
> if (msg == NULL)
> return -ENOMEM;
>
> @@ -338,6 +338,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> kfree(msg);
> return -EFAULT;
> }
> + msg[count] = '\0';
>
> bt = filp->private_data;
> __trace_note_message(bt, "%s", msg);

Nice fix!

Jens, do you ack this and should i add a Cc: <[email protected]>
backporting tag?

This fix applies fine to v2.6.29 after:

sed 's/kernel\/trace\/blktrace.c/block\/blktrace.c'

Ingo

2009-04-03 12:27:26

by Jens Axboe

[permalink] [raw]
Subject: Re: [PATCH] blktrace: fix a bug in blk_msg_write()

On Fri, Apr 03 2009, Ingo Molnar wrote:
>
> * Li Zefan <[email protected]> wrote:
>
> > This is another long-standing bug.
> >
> > ================
> >
> >
> > (console 1)
> > # echo -n 'a' > /sys/kernel/debug/block/sda/msg
> > (console 2)
> > # blktrace -d /dev/sda -a pc -o - | blkparse -i -
> > 8,0 0 0 0.000000000 0 m N a??????????????????@??????
> >
> > We should terminate the msg buffer with '\0'.
> >
> > Signed-off-by: Li Zefan <[email protected]>
> > ---
> > kernel/trace/blktrace.c | 5 +++--
> > 1 files changed, 3 insertions(+), 2 deletions(-)
> >
> > diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> > index c0ef70d..2bf341f 100644
> > --- a/kernel/trace/blktrace.c
> > +++ b/kernel/trace/blktrace.c
> > @@ -327,10 +327,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > char *msg;
> > struct blk_trace *bt;
> >
> > - if (count > BLK_TN_MAX_MSG)
> > + if (count >= BLK_TN_MAX_MSG)
> > return -EINVAL;
> >
> > - msg = kmalloc(count, GFP_KERNEL);
> > + msg = kmalloc(count + 1, GFP_KERNEL);
> > if (msg == NULL)
> > return -ENOMEM;
> >
> > @@ -338,6 +338,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > kfree(msg);
> > return -EFAULT;
> > }
> > + msg[count] = '\0';
> >
> > bt = filp->private_data;
> > __trace_note_message(bt, "%s", msg);
>
> Nice fix!
>
> Jens, do you ack this and should i add a Cc: <[email protected]>
> backporting tag?
>
> This fix applies fine to v2.6.29 after:
>
> sed 's/kernel\/trace\/blktrace.c/block\/blktrace.c'

This is identical to the following patch from Carl Henrik Lunde, which
he just ported to -tip. It has existed before, as a patch to 2.6.29-git,
but was unfortunately dropped.

So is this a coincidence? If not, Carl Henrik Lunde should be credited.

>From f116bce73e04d01614fd28ea678e0953cc321155 Mon Sep 17 00:00:00 2001
From: Carl Henrik Lunde <[email protected]>
Date: Wed, 25 Mar 2009 14:23:16 +0100
Subject: [PATCH] blktrace: NUL-terminate user space messages

Make sure messages from user space are NUL-terminated strings,
otherwise we could dump random memory to the block trace file.
Additionally, I've limited the message to BLK_TN_MAX_MSG-1 characters,
because the last character would be stripped by vscnprintf anyway.

Signed-off-by: Carl Henrik Lunde <[email protected]>
---
kernel/trace/blktrace.c | 5 +++--
1 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index d43cdac..75eb345 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -311,10 +311,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
char *msg;
struct blk_trace *bt;

- if (count > BLK_TN_MAX_MSG)
+ if (count > BLK_TN_MAX_MSG - 1)
return -EINVAL;

- msg = kmalloc(count, GFP_KERNEL);
+ msg = kmalloc(count + 1, GFP_KERNEL);
if (msg == NULL)
return -ENOMEM;

@@ -323,6 +323,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
return -EFAULT;
}

+ msg[count] = '\0';
bt = filp->private_data;
__trace_note_message(bt, "%s", msg);
kfree(msg);
--
1.6.2.rc0.90.g0753

>
> Ingo

--
Jens Axboe

2009-04-03 12:52:42

by Li Zefan

[permalink] [raw]
Subject: [tip:tracing/blktrace-v2] blktrace: small cleanup in blk_msg_write()

Commit-ID: 7635b03adf3d7b84da7649b81efa91e6ebf11b85
Gitweb: http://git.kernel.org/tip/7635b03adf3d7b84da7649b81efa91e6ebf11b85
Author: Li Zefan <[email protected]>
AuthorDate: Fri, 3 Apr 2009 15:31:34 +0800
Committer: Ingo Molnar <[email protected]>
CommitDate: Fri, 3 Apr 2009 14:48:11 +0200

blktrace: small cleanup in blk_msg_write()

Signed-off-by: Li Zefan <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: "Alan D. Brunelle" <[email protected]>
Cc: Jens Axboe <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>


---
kernel/trace/blktrace.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index a400b86..73d7860 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -327,7 +327,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
char *msg;
struct blk_trace *bt;

- if (count > BLK_TN_MAX_MSG - 1)
+ if (count >= BLK_TN_MAX_MSG)
return -EINVAL;

msg = kmalloc(count + 1, GFP_KERNEL);

2009-04-03 12:53:39

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATCH] blktrace: fix a bug in blk_msg_write()


* Jens Axboe <[email protected]> wrote:

> On Fri, Apr 03 2009, Ingo Molnar wrote:
> >
> > * Li Zefan <[email protected]> wrote:
> >
> > > This is another long-standing bug.
> > >
> > > ================
> > >
> > >
> > > (console 1)
> > > # echo -n 'a' > /sys/kernel/debug/block/sda/msg
> > > (console 2)
> > > # blktrace -d /dev/sda -a pc -o - | blkparse -i -
> > > 8,0 0 0 0.000000000 0 m N a??????????????????@??????
> > >
> > > We should terminate the msg buffer with '\0'.
> > >
> > > Signed-off-by: Li Zefan <[email protected]>
> > > ---
> > > kernel/trace/blktrace.c | 5 +++--
> > > 1 files changed, 3 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> > > index c0ef70d..2bf341f 100644
> > > --- a/kernel/trace/blktrace.c
> > > +++ b/kernel/trace/blktrace.c
> > > @@ -327,10 +327,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > > char *msg;
> > > struct blk_trace *bt;
> > >
> > > - if (count > BLK_TN_MAX_MSG)
> > > + if (count >= BLK_TN_MAX_MSG)
> > > return -EINVAL;
> > >
> > > - msg = kmalloc(count, GFP_KERNEL);
> > > + msg = kmalloc(count + 1, GFP_KERNEL);
> > > if (msg == NULL)
> > > return -ENOMEM;
> > >
> > > @@ -338,6 +338,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > > kfree(msg);
> > > return -EFAULT;
> > > }
> > > + msg[count] = '\0';
> > >
> > > bt = filp->private_data;
> > > __trace_note_message(bt, "%s", msg);
> >
> > Nice fix!
> >
> > Jens, do you ack this and should i add a Cc: <[email protected]>
> > backporting tag?
> >
> > This fix applies fine to v2.6.29 after:
> >
> > sed 's/kernel\/trace\/blktrace.c/block\/blktrace.c'
>
> This is identical to the following patch from Carl Henrik Lunde, which
> he just ported to -tip. It has existed before, as a patch to 2.6.29-git,
> but was unfortunately dropped.
>
> So is this a coincidence? If not, Carl Henrik Lunde should be credited.
>
> >From f116bce73e04d01614fd28ea678e0953cc321155 Mon Sep 17 00:00:00 2001
> From: Carl Henrik Lunde <[email protected]>
> Date: Wed, 25 Mar 2009 14:23:16 +0100
> Subject: [PATCH] blktrace: NUL-terminate user space messages
>
> Make sure messages from user space are NUL-terminated strings,
> otherwise we could dump random memory to the block trace file.
> Additionally, I've limited the message to BLK_TN_MAX_MSG-1 characters,
> because the last character would be stripped by vscnprintf anyway.
>
> Signed-off-by: Carl Henrik Lunde <[email protected]>

Sure, i've picked up Carl's version of the fix instead, it dates
back to early February!

( Li Zefan's fix was a tiny notch cleaner - i picked that cleanup up
on top of Carl's patch separately. )

Thanks,

Ingo

2009-04-03 12:55:57

by Carl Henrik Lunde

[permalink] [raw]
Subject: [tip:tracing/blktrace-v2] blktrace: NUL-terminate user space messages

Commit-ID: a4b3ada83d06554d307dd54abdc62b2e5648264a
Gitweb: http://git.kernel.org/tip/a4b3ada83d06554d307dd54abdc62b2e5648264a
Author: Carl Henrik Lunde <[email protected]>
AuthorDate: Fri, 3 Apr 2009 14:27:15 +0200
Committer: Ingo Molnar <[email protected]>
CommitDate: Fri, 3 Apr 2009 14:46:22 +0200

blktrace: NUL-terminate user space messages

Impact: fix corrupted blkparse output

Make sure messages from user space are NUL-terminated strings,
otherwise we could dump random memory to the block trace file.

Additionally, I've limited the message to BLK_TN_MAX_MSG-1
characters, because the last character would be stripped by
vscnprintf anyway.

Signed-off-by: Carl Henrik Lunde <[email protected]>
Cc: Li Zefan <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: "Alan D. Brunelle" <[email protected]>
Cc: Steven Rostedt <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>


---
kernel/trace/blktrace.c | 5 +++--
1 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index 947c5b3..a400b86 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -327,10 +327,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
char *msg;
struct blk_trace *bt;

- if (count > BLK_TN_MAX_MSG)
+ if (count > BLK_TN_MAX_MSG - 1)
return -EINVAL;

- msg = kmalloc(count, GFP_KERNEL);
+ msg = kmalloc(count + 1, GFP_KERNEL);
if (msg == NULL)
return -ENOMEM;

@@ -339,6 +339,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
return -EFAULT;
}

+ msg[count] = '\0';
bt = filp->private_data;
__trace_note_message(bt, "%s", msg);
kfree(msg);

2009-04-03 13:00:32

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATCH] blktrace: fix a bug in blk_msg_write()


* Ingo Molnar <[email protected]> wrote:

>
> * Jens Axboe <[email protected]> wrote:
>
> > On Fri, Apr 03 2009, Ingo Molnar wrote:
> > >
> > > * Li Zefan <[email protected]> wrote:
> > >
> > > > This is another long-standing bug.
> > > >
> > > > ================
> > > >
> > > >
> > > > (console 1)
> > > > # echo -n 'a' > /sys/kernel/debug/block/sda/msg
> > > > (console 2)
> > > > # blktrace -d /dev/sda -a pc -o - | blkparse -i -
> > > > 8,0 0 0 0.000000000 0 m N a??????????????????@??????
> > > >
> > > > We should terminate the msg buffer with '\0'.
> > > >
> > > > Signed-off-by: Li Zefan <[email protected]>
> > > > ---
> > > > kernel/trace/blktrace.c | 5 +++--
> > > > 1 files changed, 3 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> > > > index c0ef70d..2bf341f 100644
> > > > --- a/kernel/trace/blktrace.c
> > > > +++ b/kernel/trace/blktrace.c
> > > > @@ -327,10 +327,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > > > char *msg;
> > > > struct blk_trace *bt;
> > > >
> > > > - if (count > BLK_TN_MAX_MSG)
> > > > + if (count >= BLK_TN_MAX_MSG)
> > > > return -EINVAL;
> > > >
> > > > - msg = kmalloc(count, GFP_KERNEL);
> > > > + msg = kmalloc(count + 1, GFP_KERNEL);
> > > > if (msg == NULL)
> > > > return -ENOMEM;
> > > >
> > > > @@ -338,6 +338,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > > > kfree(msg);
> > > > return -EFAULT;
> > > > }
> > > > + msg[count] = '\0';
> > > >
> > > > bt = filp->private_data;
> > > > __trace_note_message(bt, "%s", msg);
> > >
> > > Nice fix!
> > >
> > > Jens, do you ack this and should i add a Cc: <[email protected]>
> > > backporting tag?
> > >
> > > This fix applies fine to v2.6.29 after:
> > >
> > > sed 's/kernel\/trace\/blktrace.c/block\/blktrace.c'
> >
> > This is identical to the following patch from Carl Henrik Lunde, which
> > he just ported to -tip. It has existed before, as a patch to 2.6.29-git,
> > but was unfortunately dropped.
> >
> > So is this a coincidence? If not, Carl Henrik Lunde should be credited.
> >
> > >From f116bce73e04d01614fd28ea678e0953cc321155 Mon Sep 17 00:00:00 2001
> > From: Carl Henrik Lunde <[email protected]>
> > Date: Wed, 25 Mar 2009 14:23:16 +0100
> > Subject: [PATCH] blktrace: NUL-terminate user space messages
> >
> > Make sure messages from user space are NUL-terminated strings,
> > otherwise we could dump random memory to the block trace file.
> > Additionally, I've limited the message to BLK_TN_MAX_MSG-1 characters,
> > because the last character would be stripped by vscnprintf anyway.
> >
> > Signed-off-by: Carl Henrik Lunde <[email protected]>
>
> Sure, i've picked up Carl's version of the fix instead, it dates
> back to early February!

I've also checked the linux-btrace archives a few months back and
nothing else seems to be missing - are you aware of any other
missing items?

Ingo

2009-04-03 13:03:28

by Jens Axboe

[permalink] [raw]
Subject: Re: [PATCH] blktrace: fix a bug in blk_msg_write()

On Fri, Apr 03 2009, Ingo Molnar wrote:
>
> * Ingo Molnar <[email protected]> wrote:
>
> >
> > * Jens Axboe <[email protected]> wrote:
> >
> > > On Fri, Apr 03 2009, Ingo Molnar wrote:
> > > >
> > > > * Li Zefan <[email protected]> wrote:
> > > >
> > > > > This is another long-standing bug.
> > > > >
> > > > > ================
> > > > >
> > > > >
> > > > > (console 1)
> > > > > # echo -n 'a' > /sys/kernel/debug/block/sda/msg
> > > > > (console 2)
> > > > > # blktrace -d /dev/sda -a pc -o - | blkparse -i -
> > > > > 8,0 0 0 0.000000000 0 m N a??????????????????@??????
> > > > >
> > > > > We should terminate the msg buffer with '\0'.
> > > > >
> > > > > Signed-off-by: Li Zefan <[email protected]>
> > > > > ---
> > > > > kernel/trace/blktrace.c | 5 +++--
> > > > > 1 files changed, 3 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> > > > > index c0ef70d..2bf341f 100644
> > > > > --- a/kernel/trace/blktrace.c
> > > > > +++ b/kernel/trace/blktrace.c
> > > > > @@ -327,10 +327,10 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > > > > char *msg;
> > > > > struct blk_trace *bt;
> > > > >
> > > > > - if (count > BLK_TN_MAX_MSG)
> > > > > + if (count >= BLK_TN_MAX_MSG)
> > > > > return -EINVAL;
> > > > >
> > > > > - msg = kmalloc(count, GFP_KERNEL);
> > > > > + msg = kmalloc(count + 1, GFP_KERNEL);
> > > > > if (msg == NULL)
> > > > > return -ENOMEM;
> > > > >
> > > > > @@ -338,6 +338,7 @@ static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
> > > > > kfree(msg);
> > > > > return -EFAULT;
> > > > > }
> > > > > + msg[count] = '\0';
> > > > >
> > > > > bt = filp->private_data;
> > > > > __trace_note_message(bt, "%s", msg);
> > > >
> > > > Nice fix!
> > > >
> > > > Jens, do you ack this and should i add a Cc: <[email protected]>
> > > > backporting tag?
> > > >
> > > > This fix applies fine to v2.6.29 after:
> > > >
> > > > sed 's/kernel\/trace\/blktrace.c/block\/blktrace.c'
> > >
> > > This is identical to the following patch from Carl Henrik Lunde, which
> > > he just ported to -tip. It has existed before, as a patch to 2.6.29-git,
> > > but was unfortunately dropped.
> > >
> > > So is this a coincidence? If not, Carl Henrik Lunde should be credited.
> > >
> > > >From f116bce73e04d01614fd28ea678e0953cc321155 Mon Sep 17 00:00:00 2001
> > > From: Carl Henrik Lunde <[email protected]>
> > > Date: Wed, 25 Mar 2009 14:23:16 +0100
> > > Subject: [PATCH] blktrace: NUL-terminate user space messages
> > >
> > > Make sure messages from user space are NUL-terminated strings,
> > > otherwise we could dump random memory to the block trace file.
> > > Additionally, I've limited the message to BLK_TN_MAX_MSG-1 characters,
> > > because the last character would be stripped by vscnprintf anyway.
> > >
> > > Signed-off-by: Carl Henrik Lunde <[email protected]>
> >
> > Sure, i've picked up Carl's version of the fix instead, it dates
> > back to early February!
>
> I've also checked the linux-btrace archives a few months back and
> nothing else seems to be missing - are you aware of any other
> missing items?

Nope, I think Carls patch was the only pending thing I had from the
archives.

--
Jens Axboe