[v8,1/7] net: introduce helper sendpage_ok() in include/linux/net.h

Commit Message

Coly Li Sept. 25, 2020, 3:01 p.m. UTC

The original problem was from nvme-over-tcp code, who mistakenly uses
kernel_sendpage() to send pages allocated by __get_free_pages() without
__GFP_COMP flag. Such pages don't have refcount (page_count is 0) on
tail pages, sending them by kernel_sendpage() may trigger a kernel panic
from a corrupted kernel heap, because these pages are incorrectly freed
in network stack as page_count 0 pages.

This patch introduces a helper sendpage_ok(), it returns true if the
checking page,
- is not slab page: PageSlab(page) is false.
- has page refcount: page_count(page) is not zero

All drivers who want to send page to remote end by kernel_sendpage()
may use this helper to check whether the page is OK. If the helper does
not return true, the driver should try other non sendpage method (e.g.
sock_no_sendpage()) to handle the page.

Signed-off-by: Coly Li <[e-mail]>
Cc: Chaitanya Kulkarni <[e-mail]>
Cc: Christoph Hellwig <[e-mail]>
Cc: Hannes Reinecke <[e-mail]>
Cc: Jan Kara <[e-mail]>
Cc: Jens Axboe <[e-mail]>
Cc: Mikhail Skorzhinskii <[e-mail]>
Cc: Philipp Reisner <[e-mail]>
Cc: Sagi Grimberg <[e-mail]>
Cc: Vlastimil Babka <[e-mail]>
Cc: [e-mail]
---
 include/linux/net.h | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

Comments

Greg KH Sept. 25, 2020, 3:18 p.m. UTC | #1

On Fri, Sep 25, 2020 at 11:01:13PM +0800, Coly Li wrote:
> The original problem was from nvme-over-tcp code, who mistakenly uses
> kernel_sendpage() to send pages allocated by __get_free_pages() without
> __GFP_COMP flag. Such pages don't have refcount (page_count is 0) on
> tail pages, sending them by kernel_sendpage() may trigger a kernel panic
> from a corrupted kernel heap, because these pages are incorrectly freed
> in network stack as page_count 0 pages.
> 
> This patch introduces a helper sendpage_ok(), it returns true if the
> checking page,
> - is not slab page: PageSlab(page) is false.
> - has page refcount: page_count(page) is not zero
> 
> All drivers who want to send page to remote end by kernel_sendpage()
> may use this helper to check whether the page is OK. If the helper does
> not return true, the driver should try other non sendpage method (e.g.
> sock_no_sendpage()) to handle the page.
> 
> Signed-off-by: Coly Li <[e-mail]>
> Cc: Chaitanya Kulkarni <[e-mail]>
> Cc: Christoph Hellwig <[e-mail]>
> Cc: Hannes Reinecke <[e-mail]>
> Cc: Jan Kara <[e-mail]>
> Cc: Jens Axboe <[e-mail]>
> Cc: Mikhail Skorzhinskii <[e-mail]>
> Cc: Philipp Reisner <[e-mail]>
> Cc: Sagi Grimberg <[e-mail]>
> Cc: Vlastimil Babka <[e-mail]>
> Cc: [e-mail]
> ---
>  include/linux/net.h | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/include/linux/net.h b/include/linux/net.h
> index d48ff1180879..05db8690f67e 100644
> --- a/include/linux/net.h
> +++ b/include/linux/net.h
> @@ -21,6 +21,7 @@
>  #include <linux/rcupdate.h>
>  #include <linux/once.h>
>  #include <linux/fs.h>
> +#include <linux/mm.h>
>  #include <linux/sockptr.h>
>  
>  #include <uapi/linux/net.h>
> @@ -286,6 +287,21 @@ do {									\
>  #define net_get_random_once_wait(buf, nbytes)			\
>  	get_random_once_wait((buf), (nbytes))
>  
> +/*
> + * E.g. XFS meta- & log-data is in slab pages, or bcache meta
> + * data pages, or other high order pages allocated by
> + * __get_free_pages() without __GFP_COMP, which have a page_count
> + * of 0 and/or have PageSlab() set. We cannot use send_page for
> + * those, as that does get_page(); put_page(); and would cause
> + * either a VM_BUG directly, or __page_cache_release a page that
> + * would actually still be referenced by someone, leading to some
> + * obscure delayed Oops somewhere else.
> + */
> +static inline bool sendpage_ok(struct page *page)
> +{
> +	return  !PageSlab(page) && page_count(page) >= 1;

Do you have one extra ' ' after "return" there?

And this feels like a mm thing, why put it in net.h and not mm.h?

thanks,

greg k-h

Coly Li Sept. 26, 2020, 1:28 p.m. UTC | #2

On 2020/9/25 23:18, Greg KH wrote:
> On Fri, Sep 25, 2020 at 11:01:13PM +0800, Coly Li wrote:
>> The original problem was from nvme-over-tcp code, who mistakenly uses
>> kernel_sendpage() to send pages allocated by __get_free_pages() without
>> __GFP_COMP flag. Such pages don't have refcount (page_count is 0) on
>> tail pages, sending them by kernel_sendpage() may trigger a kernel panic
>> from a corrupted kernel heap, because these pages are incorrectly freed
>> in network stack as page_count 0 pages.
>>
>> This patch introduces a helper sendpage_ok(), it returns true if the
>> checking page,
>> - is not slab page: PageSlab(page) is false.
>> - has page refcount: page_count(page) is not zero
>>
>> All drivers who want to send page to remote end by kernel_sendpage()
>> may use this helper to check whether the page is OK. If the helper does
>> not return true, the driver should try other non sendpage method (e.g.
>> sock_no_sendpage()) to handle the page.
>>
>> Signed-off-by: Coly Li <[e-mail]>
>> Cc: Chaitanya Kulkarni <[e-mail]>
>> Cc: Christoph Hellwig <[e-mail]>
>> Cc: Hannes Reinecke <[e-mail]>
>> Cc: Jan Kara <[e-mail]>
>> Cc: Jens Axboe <[e-mail]>
>> Cc: Mikhail Skorzhinskii <[e-mail]>
>> Cc: Philipp Reisner <[e-mail]>
>> Cc: Sagi Grimberg <[e-mail]>
>> Cc: Vlastimil Babka <[e-mail]>
>> Cc: [e-mail]
>> ---
>>  include/linux/net.h | 16 ++++++++++++++++
>>  1 file changed, 16 insertions(+)
>>
>> diff --git a/include/linux/net.h b/include/linux/net.h
>> index d48ff1180879..05db8690f67e 100644
>> --- a/include/linux/net.h
>> +++ b/include/linux/net.h
>> @@ -21,6 +21,7 @@
>>  #include <linux/rcupdate.h>
>>  #include <linux/once.h>
>>  #include <linux/fs.h>
>> +#include <linux/mm.h>
>>  #include <linux/sockptr.h>
>>  
>>  #include <uapi/linux/net.h>
>> @@ -286,6 +287,21 @@ do {									\
>>  #define net_get_random_once_wait(buf, nbytes)			\
>>  	get_random_once_wait((buf), (nbytes))
>>  
>> +/*
>> + * E.g. XFS meta- & log-data is in slab pages, or bcache meta
>> + * data pages, or other high order pages allocated by
>> + * __get_free_pages() without __GFP_COMP, which have a page_count
>> + * of 0 and/or have PageSlab() set. We cannot use send_page for
>> + * those, as that does get_page(); put_page(); and would cause
>> + * either a VM_BUG directly, or __page_cache_release a page that
>> + * would actually still be referenced by someone, leading to some
>> + * obscure delayed Oops somewhere else.
>> + */
>> +static inline bool sendpage_ok(struct page *page)
>> +{
>> +	return  !PageSlab(page) && page_count(page) >= 1;
> 
> Do you have one extra ' ' after "return" there?

It should be fixed in next version.

> 
> And this feels like a mm thing, why put it in net.h and not mm.h?

This check is specific for kernel_sendpage(), so I want to place it
closer to where kernel_sendpage() is declared.

And indeed there was similar discussion about why this helper is not in
mm code in v5 series. Christoph supported to place sendpage_ok() in
net.h, an uncompleted piece of his opinion was "It is not a mm bug, it
is a networking quirk."

Thanks.

Coly Li

Greg KH Sept. 27, 2020, 12:21 p.m. UTC | #3

On Sat, Sep 26, 2020 at 09:28:03PM +0800, Coly Li wrote:
> On 2020/9/25 23:18, Greg KH wrote:
> > On Fri, Sep 25, 2020 at 11:01:13PM +0800, Coly Li wrote:
> >> The original problem was from nvme-over-tcp code, who mistakenly uses
> >> kernel_sendpage() to send pages allocated by __get_free_pages() without
> >> __GFP_COMP flag. Such pages don't have refcount (page_count is 0) on
> >> tail pages, sending them by kernel_sendpage() may trigger a kernel panic
> >> from a corrupted kernel heap, because these pages are incorrectly freed
> >> in network stack as page_count 0 pages.
> >>
> >> This patch introduces a helper sendpage_ok(), it returns true if the
> >> checking page,
> >> - is not slab page: PageSlab(page) is false.
> >> - has page refcount: page_count(page) is not zero
> >>
> >> All drivers who want to send page to remote end by kernel_sendpage()
> >> may use this helper to check whether the page is OK. If the helper does
> >> not return true, the driver should try other non sendpage method (e.g.
> >> sock_no_sendpage()) to handle the page.
> >>
> >> Signed-off-by: Coly Li <[e-mail]>
> >> Cc: Chaitanya Kulkarni <[e-mail]>
> >> Cc: Christoph Hellwig <[e-mail]>
> >> Cc: Hannes Reinecke <[e-mail]>
> >> Cc: Jan Kara <[e-mail]>
> >> Cc: Jens Axboe <[e-mail]>
> >> Cc: Mikhail Skorzhinskii <[e-mail]>
> >> Cc: Philipp Reisner <[e-mail]>
> >> Cc: Sagi Grimberg <[e-mail]>
> >> Cc: Vlastimil Babka <[e-mail]>
> >> Cc: [e-mail]
> >> ---
> >>  include/linux/net.h | 16 ++++++++++++++++
> >>  1 file changed, 16 insertions(+)
> >>
> >> diff --git a/include/linux/net.h b/include/linux/net.h
> >> index d48ff1180879..05db8690f67e 100644
> >> --- a/include/linux/net.h
> >> +++ b/include/linux/net.h
> >> @@ -21,6 +21,7 @@
> >>  #include <linux/rcupdate.h>
> >>  #include <linux/once.h>
> >>  #include <linux/fs.h>
> >> +#include <linux/mm.h>
> >>  #include <linux/sockptr.h>
> >>  
> >>  #include <uapi/linux/net.h>
> >> @@ -286,6 +287,21 @@ do {									\
> >>  #define net_get_random_once_wait(buf, nbytes)			\
> >>  	get_random_once_wait((buf), (nbytes))
> >>  
> >> +/*
> >> + * E.g. XFS meta- & log-data is in slab pages, or bcache meta
> >> + * data pages, or other high order pages allocated by
> >> + * __get_free_pages() without __GFP_COMP, which have a page_count
> >> + * of 0 and/or have PageSlab() set. We cannot use send_page for
> >> + * those, as that does get_page(); put_page(); and would cause
> >> + * either a VM_BUG directly, or __page_cache_release a page that
> >> + * would actually still be referenced by someone, leading to some
> >> + * obscure delayed Oops somewhere else.
> >> + */
> >> +static inline bool sendpage_ok(struct page *page)
> >> +{
> >> +	return  !PageSlab(page) && page_count(page) >= 1;
> > 
> > Do you have one extra ' ' after "return" there?
> 
> It should be fixed in next version.
> 
> > 
> > And this feels like a mm thing, why put it in net.h and not mm.h?
> 
> This check is specific for kernel_sendpage(), so I want to place it
> closer to where kernel_sendpage() is declared.
> 
> And indeed there was similar discussion about why this helper is not in
> mm code in v5 series. Christoph supported to place sendpage_ok() in
> net.h, an uncompleted piece of his opinion was "It is not a mm bug, it
> is a networking quirk."

Ah, nevermind then, sorry for the noise :)

greg k-h

Patch

diff --git a/include/linux/net.h b/include/linux/net.h
index d48ff1180879..05db8690f67e 100644
--- a/include/linux/net.h
+++ b/include/linux/net.h
@@ -21,6 +21,7 @@ 
 #include <linux/rcupdate.h>
 #include <linux/once.h>
 #include <linux/fs.h>
+#include <linux/mm.h>
 #include <linux/sockptr.h>
 
 #include <uapi/linux/net.h>
@@ -286,6 +287,21 @@  do {									\
 #define net_get_random_once_wait(buf, nbytes)			\
 	get_random_once_wait((buf), (nbytes))
 
+/*
+ * E.g. XFS meta- & log-data is in slab pages, or bcache meta
+ * data pages, or other high order pages allocated by
+ * __get_free_pages() without __GFP_COMP, which have a page_count
+ * of 0 and/or have PageSlab() set. We cannot use send_page for
+ * those, as that does get_page(); put_page(); and would cause
+ * either a VM_BUG directly, or __page_cache_release a page that
+ * would actually still be referenced by someone, leading to some
+ * obscure delayed Oops somewhere else.
+ */
+static inline bool sendpage_ok(struct page *page)
+{
+	return  !PageSlab(page) && page_count(page) >= 1;
+}
+
 int kernel_sendmsg(struct socket *sock, struct msghdr *msg, struct kvec *vec,
 		   size_t num, size_t len);
 int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,