[PATCH v2 0/3] Few bug fixes and Convert to pin_user_pages*()

[PATCH v2 0/3] Few bug fixes and Convert to pin_user_pages*()

[Souptick Joarder]

This series contains few clean up, minor bug fixes and
Convert get_user_pages() to pin_user_pages().

I'm compile tested this, but unable to run-time test,
so any testing help is much appriciated.

v2:
	Addressed few review comments and compile issue.
	Patch[1/2] from v1 split into 2 in v2.

Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Paul Durrant <xadimgnik@gmail.com>

Souptick Joarder (3):
  xen/privcmd: Corrected error handling path
  xen/privcmd: Mark pages as dirty
  xen/privcmd: Convert get_user_pages*() to      pin_user_pages*()

 drivers/xen/privcmd.c | 32 ++++++++++++++------------------
 1 file changed, 14 insertions(+), 18 deletions(-)

-- 
1.9.1

[PATCH v2 1/3] xen/privcmd: Corrected error handling path

[Souptick Joarder]

Previously, if lock_pages() end up partially mapping pages, it used
to return -ERRNO due to which unlock_pages() have to go through
each pages[i] till *nr_pages* to validate them. This can be avoided
by passing correct number of partially mapped pages & -ERRNO separately,
while returning from lock_pages() due to error.

With this fix unlock_pages() doesn't need to validate pages[i] till
*nr_pages* for error scenario and few condition checks can be ignored.

Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Paul Durrant <xadimgnik@gmail.com>
---
 drivers/xen/privcmd.c | 31 +++++++++++++++----------------
 1 file changed, 15 insertions(+), 16 deletions(-)

diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
index a250d11..33677ea 100644
--- a/drivers/xen/privcmd.c
+++ b/drivers/xen/privcmd.c
@@ -580,13 +580,13 @@ static long privcmd_ioctl_mmap_batch(
 
 static int lock_pages(
 	struct privcmd_dm_op_buf kbufs[], unsigned int num,
-	struct page *pages[], unsigned int nr_pages)
+	struct page *pages[], unsigned int nr_pages, unsigned int *pinned)
 {
 	unsigned int i;
+	int page_count = 0;
 
 	for (i = 0; i < num; i++) {
 		unsigned int requested;
-		int pinned;
 
 		requested = DIV_ROUND_UP(
 			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
@@ -594,14 +594,15 @@ static int lock_pages(
 		if (requested > nr_pages)
 			return -ENOSPC;
 
-		pinned = get_user_pages_fast(
+		page_count = get_user_pages_fast(
 			(unsigned long) kbufs[i].uptr,
 			requested, FOLL_WRITE, pages);
-		if (pinned < 0)
-			return pinned;
+		if (page_count < 0)
+			return page_count;
 
-		nr_pages -= pinned;
-		pages += pinned;
+		*pinned += page_count;
+		nr_pages -= page_count;
+		pages += page_count;
 	}
 
 	return 0;
@@ -611,13 +612,8 @@ static void unlock_pages(struct page *pages[], unsigned int nr_pages)
 {
 	unsigned int i;
 
-	if (!pages)
-		return;
-
-	for (i = 0; i < nr_pages; i++) {
-		if (pages[i])
-			put_page(pages[i]);
-	}
+	for (i = 0; i < nr_pages; i++)
+		put_page(pages[i]);
 }
 
 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
@@ -630,6 +626,7 @@ static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
 	struct xen_dm_op_buf *xbufs = NULL;
 	unsigned int i;
 	long rc;
+	unsigned int pinned = 0;
 
 	if (copy_from_user(&kdata, udata, sizeof(kdata)))
 		return -EFAULT;
@@ -683,9 +680,11 @@ static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
 		goto out;
 	}
 
-	rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
-	if (rc)
+	rc = lock_pages(kbufs, kdata.num, pages, nr_pages, &pinned);
+	if (rc < 0) {
+		nr_pages = pinned;
 		goto out;
+	}
 
 	for (i = 0; i < kdata.num; i++) {
 		set_xen_guest_handle(xbufs[i].h, kbufs[i].uptr);
-- 
1.9.1

[PATCH v2 2/3] xen/privcmd: Mark pages as dirty

[Souptick Joarder]

pages need to be marked as dirty before unpinned it in
unlock_pages() which was oversight. This is fixed now.

Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Suggested-by: John Hubbard <jhubbard@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Paul Durrant <xadimgnik@gmail.com>
---
 drivers/xen/privcmd.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
index 33677ea..f6c1543 100644
--- a/drivers/xen/privcmd.c
+++ b/drivers/xen/privcmd.c
@@ -612,8 +612,11 @@ static void unlock_pages(struct page *pages[], unsigned int nr_pages)
 {
 	unsigned int i;
 
-	for (i = 0; i < nr_pages; i++)
+	for (i = 0; i < nr_pages; i++) {
+		if (!PageDirty(pages[i]))
+			set_page_dirty_lock(pages[i]);
 		put_page(pages[i]);
+	}
 }
 
 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
-- 
1.9.1

[PATCH v2 3/3] xen/privcmd: Convert get_user_pages*() to pin_user_pages*()

[Souptick Joarder]

In 2019, we introduced pin_user_pages*() and now we are converting
get_user_pages*() to the new API as appropriate. [1] & [2] could
be referred for more information. This is case 5 as per document [1].

[1] Documentation/core-api/pin_user_pages.rst

[2] "Explicit pinning of user-space pages":
        https://lwn.net/Articles/807108/

Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Paul Durrant <xadimgnik@gmail.com>
---
 drivers/xen/privcmd.c | 10 ++--------
 1 file changed, 2 insertions(+), 8 deletions(-)

diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
index f6c1543..5c5cd24 100644
--- a/drivers/xen/privcmd.c
+++ b/drivers/xen/privcmd.c
@@ -594,7 +594,7 @@ static int lock_pages(
 		if (requested > nr_pages)
 			return -ENOSPC;
 
-		page_count = get_user_pages_fast(
+		page_count = pin_user_pages_fast(
 			(unsigned long) kbufs[i].uptr,
 			requested, FOLL_WRITE, pages);
 		if (page_count < 0)
@@ -610,13 +610,7 @@ static int lock_pages(
 
 static void unlock_pages(struct page *pages[], unsigned int nr_pages)
 {
-	unsigned int i;
-
-	for (i = 0; i < nr_pages; i++) {
-		if (!PageDirty(pages[i]))
-			set_page_dirty_lock(pages[i]);
-		put_page(pages[i]);
-	}
+	unpin_user_pages_dirty_lock(pages, nr_pages, true);
 }
 
 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
-- 
1.9.1

Re: [PATCH v2 1/3] xen/privcmd: Corrected error handling path

[Jürgen Groß]

On 06.07.20 20:16, Souptick Joarder wrote:
> Previously, if lock_pages() end up partially mapping pages, it used
> to return -ERRNO due to which unlock_pages() have to go through
> each pages[i] till *nr_pages* to validate them. This can be avoided
> by passing correct number of partially mapped pages & -ERRNO separately,
> while returning from lock_pages() due to error.
> 
> With this fix unlock_pages() doesn't need to validate pages[i] till
> *nr_pages* for error scenario and few condition checks can be ignored.
> 
> Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
> Cc: John Hubbard <jhubbard@nvidia.com>
> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> Cc: Paul Durrant <xadimgnik@gmail.com>
> ---
>   drivers/xen/privcmd.c | 31 +++++++++++++++----------------
>   1 file changed, 15 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
> index a250d11..33677ea 100644
> --- a/drivers/xen/privcmd.c
> +++ b/drivers/xen/privcmd.c
> @@ -580,13 +580,13 @@ static long privcmd_ioctl_mmap_batch(
>   
>   static int lock_pages(
>   	struct privcmd_dm_op_buf kbufs[], unsigned int num,
> -	struct page *pages[], unsigned int nr_pages)
> +	struct page *pages[], unsigned int nr_pages, unsigned int *pinned)
>   {
>   	unsigned int i;
> +	int page_count = 0;

Initial value shouldn't be needed, and ...

>   
>   	for (i = 0; i < num; i++) {
>   		unsigned int requested;
> -		int pinned;

... you could move the declaration here.

With that done you can add my

Reviewed-by: Juergen Gross <jgross@suse.com>


Juergen

Re: [PATCH v2 2/3] xen/privcmd: Mark pages as dirty

[Jürgen Groß]

On 06.07.20 20:16, Souptick Joarder wrote:
> pages need to be marked as dirty before unpinned it in
> unlock_pages() which was oversight. This is fixed now.
> 
> Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
> Suggested-by: John Hubbard <jhubbard@nvidia.com>
> Cc: John Hubbard <jhubbard@nvidia.com>
> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> Cc: Paul Durrant <xadimgnik@gmail.com>
> ---
>   drivers/xen/privcmd.c | 5 ++++-
>   1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
> index 33677ea..f6c1543 100644
> --- a/drivers/xen/privcmd.c
> +++ b/drivers/xen/privcmd.c
> @@ -612,8 +612,11 @@ static void unlock_pages(struct page *pages[], unsigned int nr_pages)
>   {
>   	unsigned int i;
>   
> -	for (i = 0; i < nr_pages; i++)
> +	for (i = 0; i < nr_pages; i++) {
> +		if (!PageDirty(pages[i]))
> +			set_page_dirty_lock(pages[i]);

With put_page() directly following I think you should be able to use
set_page_dirty() instead, as there is obviously a reference to the page
existing.

>   		put_page(pages[i]);
> +	}
>   }
>   
>   static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
> 

Juergen

Re: [PATCH v2 2/3] xen/privcmd: Mark pages as dirty

[Souptick Joarder]

On Tue, Jul 7, 2020 at 3:08 PM Jürgen Groß <jgross@suse.com> wrote:
>
> On 06.07.20 20:16, Souptick Joarder wrote:
> > pages need to be marked as dirty before unpinned it in
> > unlock_pages() which was oversight. This is fixed now.
> >
> > Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
> > Suggested-by: John Hubbard <jhubbard@nvidia.com>
> > Cc: John Hubbard <jhubbard@nvidia.com>
> > Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> > Cc: Paul Durrant <xadimgnik@gmail.com>
> > ---
> >   drivers/xen/privcmd.c | 5 ++++-
> >   1 file changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
> > index 33677ea..f6c1543 100644
> > --- a/drivers/xen/privcmd.c
> > +++ b/drivers/xen/privcmd.c
> > @@ -612,8 +612,11 @@ static void unlock_pages(struct page *pages[], unsigned int nr_pages)
> >   {
> >       unsigned int i;
> >
> > -     for (i = 0; i < nr_pages; i++)
> > +     for (i = 0; i < nr_pages; i++) {
> > +             if (!PageDirty(pages[i]))
> > +                     set_page_dirty_lock(pages[i]);
>
> With put_page() directly following I think you should be able to use
> set_page_dirty() instead, as there is obviously a reference to the page
> existing.

Patch [3/3] will convert above codes to use unpin_user_pages_dirty_lock()
which internally do the same check. So I thought to keep linux-stable and
linux-next code in sync. John had a similar concern [1] and later agreed to keep
this check.

Shall I keep this check ?  No ?

[1] https://lore.kernel.org/xen-devel/a750e5e5-fd5d-663b-c5fd-261d7c939ba7@nvidia.com/

>
> >               put_page(pages[i]);
> > +     }
> >   }
> >
> >   static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
> >
>
> Juergen

Re: [PATCH v2 1/3] xen/privcmd: Corrected error handling path

[Souptick Joarder]

On Tue, Jul 7, 2020 at 3:05 PM Jürgen Groß <jgross@suse.com> wrote:
>
> On 06.07.20 20:16, Souptick Joarder wrote:
> > Previously, if lock_pages() end up partially mapping pages, it used
> > to return -ERRNO due to which unlock_pages() have to go through
> > each pages[i] till *nr_pages* to validate them. This can be avoided
> > by passing correct number of partially mapped pages & -ERRNO separately,
> > while returning from lock_pages() due to error.
> >
> > With this fix unlock_pages() doesn't need to validate pages[i] till
> > *nr_pages* for error scenario and few condition checks can be ignored.
> >
> > Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
> > Cc: John Hubbard <jhubbard@nvidia.com>
> > Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> > Cc: Paul Durrant <xadimgnik@gmail.com>
> > ---
> >   drivers/xen/privcmd.c | 31 +++++++++++++++----------------
> >   1 file changed, 15 insertions(+), 16 deletions(-)
> >
> > diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
> > index a250d11..33677ea 100644
> > --- a/drivers/xen/privcmd.c
> > +++ b/drivers/xen/privcmd.c
> > @@ -580,13 +580,13 @@ static long privcmd_ioctl_mmap_batch(
> >
> >   static int lock_pages(
> >       struct privcmd_dm_op_buf kbufs[], unsigned int num,
> > -     struct page *pages[], unsigned int nr_pages)
> > +     struct page *pages[], unsigned int nr_pages, unsigned int *pinned)
> >   {
> >       unsigned int i;
> > +     int page_count = 0;
>
> Initial value shouldn't be needed, and ...
>
> >
> >       for (i = 0; i < num; i++) {
> >               unsigned int requested;
> > -             int pinned;
>
> ... you could move the declaration here.
>
> With that done you can add my
>
> Reviewed-by: Juergen Gross <jgross@suse.com>

Ok. But does it going make any difference other than limiting scope ?

>
>
> Juergen

Re: [PATCH v2 2/3] xen/privcmd: Mark pages as dirty

[Jürgen Groß]

On 07.07.20 13:30, Souptick Joarder wrote:
> On Tue, Jul 7, 2020 at 3:08 PM Jürgen Groß <jgross@suse.com> wrote:
>>
>> On 06.07.20 20:16, Souptick Joarder wrote:
>>> pages need to be marked as dirty before unpinned it in
>>> unlock_pages() which was oversight. This is fixed now.
>>>
>>> Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
>>> Suggested-by: John Hubbard <jhubbard@nvidia.com>
>>> Cc: John Hubbard <jhubbard@nvidia.com>
>>> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
>>> Cc: Paul Durrant <xadimgnik@gmail.com>
>>> ---
>>>    drivers/xen/privcmd.c | 5 ++++-
>>>    1 file changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
>>> index 33677ea..f6c1543 100644
>>> --- a/drivers/xen/privcmd.c
>>> +++ b/drivers/xen/privcmd.c
>>> @@ -612,8 +612,11 @@ static void unlock_pages(struct page *pages[], unsigned int nr_pages)
>>>    {
>>>        unsigned int i;
>>>
>>> -     for (i = 0; i < nr_pages; i++)
>>> +     for (i = 0; i < nr_pages; i++) {
>>> +             if (!PageDirty(pages[i]))
>>> +                     set_page_dirty_lock(pages[i]);
>>
>> With put_page() directly following I think you should be able to use
>> set_page_dirty() instead, as there is obviously a reference to the page
>> existing.
> 
> Patch [3/3] will convert above codes to use unpin_user_pages_dirty_lock()
> which internally do the same check. So I thought to keep linux-stable and
> linux-next code in sync. John had a similar concern [1] and later agreed to keep
> this check.
> 
> Shall I keep this check ?  No ?
> 
> [1] https://lore.kernel.org/xen-devel/a750e5e5-fd5d-663b-c5fd-261d7c939ba7@nvidia.com/

I wasn't referring to checking PageDirty(), but to the use of
set_page_dirty_lock().

Looking at the comment just before the implementation of
set_page_dirty_lock() suggests that it is fine to use set_page_dirty()
instead (so not calling lock_page()).

Only the transition from get_user_pages_fast() to pin_user_pages_fast()
requires to use the locked version IMO.


Juergen

Re: [PATCH v2 1/3] xen/privcmd: Corrected error handling path

[Jürgen Groß]

On 07.07.20 13:40, Souptick Joarder wrote:
> On Tue, Jul 7, 2020 at 3:05 PM Jürgen Groß <jgross@suse.com> wrote:
>>
>> On 06.07.20 20:16, Souptick Joarder wrote:
>>> Previously, if lock_pages() end up partially mapping pages, it used
>>> to return -ERRNO due to which unlock_pages() have to go through
>>> each pages[i] till *nr_pages* to validate them. This can be avoided
>>> by passing correct number of partially mapped pages & -ERRNO separately,
>>> while returning from lock_pages() due to error.
>>>
>>> With this fix unlock_pages() doesn't need to validate pages[i] till
>>> *nr_pages* for error scenario and few condition checks can be ignored.
>>>
>>> Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
>>> Cc: John Hubbard <jhubbard@nvidia.com>
>>> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
>>> Cc: Paul Durrant <xadimgnik@gmail.com>
>>> ---
>>>    drivers/xen/privcmd.c | 31 +++++++++++++++----------------
>>>    1 file changed, 15 insertions(+), 16 deletions(-)
>>>
>>> diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
>>> index a250d11..33677ea 100644
>>> --- a/drivers/xen/privcmd.c
>>> +++ b/drivers/xen/privcmd.c
>>> @@ -580,13 +580,13 @@ static long privcmd_ioctl_mmap_batch(
>>>
>>>    static int lock_pages(
>>>        struct privcmd_dm_op_buf kbufs[], unsigned int num,
>>> -     struct page *pages[], unsigned int nr_pages)
>>> +     struct page *pages[], unsigned int nr_pages, unsigned int *pinned)
>>>    {
>>>        unsigned int i;
>>> +     int page_count = 0;
>>
>> Initial value shouldn't be needed, and ...
>>
>>>
>>>        for (i = 0; i < num; i++) {
>>>                unsigned int requested;
>>> -             int pinned;
>>
>> ... you could move the declaration here.
>>
>> With that done you can add my
>>
>> Reviewed-by: Juergen Gross <jgross@suse.com>
> 
> Ok. But does it going make any difference other than limiting scope ?

Dropping the initializer surely does, and in the end page_count just
replaces the former pinned variable, so why would we want to widen the
scope with this patch?


Juergen

Re: [PATCH v2 3/3] xen/privcmd: Convert get_user_pages*() to pin_user_pages*()

[Jürgen Groß]

On 06.07.20 20:16, Souptick Joarder wrote:
> In 2019, we introduced pin_user_pages*() and now we are converting
> get_user_pages*() to the new API as appropriate. [1] & [2] could
> be referred for more information. This is case 5 as per document [1].
> 
> [1] Documentation/core-api/pin_user_pages.rst
> 
> [2] "Explicit pinning of user-space pages":
>          https://lwn.net/Articles/807108/
> 
> Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
> Cc: John Hubbard <jhubbard@nvidia.com>
> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> Cc: Paul Durrant <xadimgnik@gmail.com>

Reviewed-by: Juergen Gross <jgross@suse.com>


Juergen

Re: [PATCH v2 2/3] xen/privcmd: Mark pages as dirty

[John Hubbard]

On 2020-07-07 04:43, Jürgen Groß wrote:
> On 07.07.20 13:30, Souptick Joarder wrote:
>> On Tue, Jul 7, 2020 at 3:08 PM Jürgen Groß <jgross@suse.com> wrote:
...
>>>> diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
>>>> index 33677ea..f6c1543 100644
>>>> --- a/drivers/xen/privcmd.c
>>>> +++ b/drivers/xen/privcmd.c
>>>> @@ -612,8 +612,11 @@ static void unlock_pages(struct page *pages[], unsigned int nr_pages)
>>>>    {
>>>>        unsigned int i;
>>>>
>>>> -     for (i = 0; i < nr_pages; i++)
>>>> +     for (i = 0; i < nr_pages; i++) {
>>>> +             if (!PageDirty(pages[i]))
>>>> +                     set_page_dirty_lock(pages[i]);
>>>
>>> With put_page() directly following I think you should be able to use
>>> set_page_dirty() instead, as there is obviously a reference to the page
>>> existing.
>>
>> Patch [3/3] will convert above codes to use unpin_user_pages_dirty_lock()
>> which internally do the same check. So I thought to keep linux-stable and
>> linux-next code in sync. John had a similar concern [1] and later agreed to keep
>> this check.
>>
>> Shall I keep this check ?  No ?

It doesn't matter *too* much, because patch 3/3 fixes up everything by
changing it all to unpin_user_pages_dirty_lock(). However, there is something
to be said for having correct interim patches, too. :)  Details:

>>
>> [1] https://lore.kernel.org/xen-devel/a750e5e5-fd5d-663b-c5fd-261d7c939ba7@nvidia.com/
> 
> I wasn't referring to checking PageDirty(), but to the use of
> set_page_dirty_lock().
> 
> Looking at the comment just before the implementation of
> set_page_dirty_lock() suggests that it is fine to use set_page_dirty()
> instead (so not calling lock_page()).


no no, that's a misreading of the comment. Unless this xen/privcmd code has
somehow taken a reference on page->mapping->host (which I do *not* think is
the case), then it is still racy to call set_page_dirty() here. Instead,
set_page_dirty_lock() should be used.


> 
> Only the transition from get_user_pages_fast() to pin_user_pages_fast()
> requires to use the locked version IMO.
> 

That's a different misunderstanding. :) pin_user_pages*() APIs are meant to be
functionally drop-in replacements for get_user_pages*(). Internally,
pin_user_pages*() functions do some additional tracking, but from a caller's
perspective, it should look the same. In other words, there is nothing
about pin_user_pages_fast() that requires set_page_dirty_lock() upon release.
The reason set_page_dirty_lock() was chosen is that there are very few
(none at all?) call sites that need to release and dirty a page, that also meet
the requirements to safely call set_page_dirty().

That's why there is a unpin_user_pages_dirty_lock(), but there is not a
corresponding unpin_user_pages_dirty() call: the latter has not been required
so far, even though the call site conversions are nearly done.


thanks,
-- 
John Hubbard
NVIDIA

Re: [PATCH v2 1/3] xen/privcmd: Corrected error handling path

[Souptick Joarder]

On Tue, Jul 7, 2020 at 5:15 PM Jürgen Groß <jgross@suse.com> wrote:
>
> On 07.07.20 13:40, Souptick Joarder wrote:
> > On Tue, Jul 7, 2020 at 3:05 PM Jürgen Groß <jgross@suse.com> wrote:
> >>
> >> On 06.07.20 20:16, Souptick Joarder wrote:
> >>> Previously, if lock_pages() end up partially mapping pages, it used
> >>> to return -ERRNO due to which unlock_pages() have to go through
> >>> each pages[i] till *nr_pages* to validate them. This can be avoided
> >>> by passing correct number of partially mapped pages & -ERRNO separately,
> >>> while returning from lock_pages() due to error.
> >>>
> >>> With this fix unlock_pages() doesn't need to validate pages[i] till
> >>> *nr_pages* for error scenario and few condition checks can be ignored.
> >>>
> >>> Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
> >>> Cc: John Hubbard <jhubbard@nvidia.com>
> >>> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> >>> Cc: Paul Durrant <xadimgnik@gmail.com>
> >>> ---
> >>>    drivers/xen/privcmd.c | 31 +++++++++++++++----------------
> >>>    1 file changed, 15 insertions(+), 16 deletions(-)
> >>>
> >>> diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
> >>> index a250d11..33677ea 100644
> >>> --- a/drivers/xen/privcmd.c
> >>> +++ b/drivers/xen/privcmd.c
> >>> @@ -580,13 +580,13 @@ static long privcmd_ioctl_mmap_batch(
> >>>
> >>>    static int lock_pages(
> >>>        struct privcmd_dm_op_buf kbufs[], unsigned int num,
> >>> -     struct page *pages[], unsigned int nr_pages)
> >>> +     struct page *pages[], unsigned int nr_pages, unsigned int *pinned)
> >>>    {
> >>>        unsigned int i;
> >>> +     int page_count = 0;
> >>
> >> Initial value shouldn't be needed, and ...
> >>
> >>>
> >>>        for (i = 0; i < num; i++) {
> >>>                unsigned int requested;
> >>> -             int pinned;
> >>
> >> ... you could move the declaration here.
> >>
> >> With that done you can add my
> >>
> >> Reviewed-by: Juergen Gross <jgross@suse.com>
> >
> > Ok. But does it going make any difference other than limiting scope ?
>
> Dropping the initializer surely does, and in the end page_count just
> replaces the former pinned variable, so why would we want to widen the
> scope with this patch?

Agree, no reason to move it up. Will change it in v3.

>
>
> Juergen

Re: [PATCH v2 2/3] xen/privcmd: Mark pages as dirty

[Jürgen Groß]

On 07.07.20 21:30, John Hubbard wrote:
> On 2020-07-07 04:43, Jürgen Groß wrote:
>> On 07.07.20 13:30, Souptick Joarder wrote:
>>> On Tue, Jul 7, 2020 at 3:08 PM Jürgen Groß <jgross@suse.com> wrote:
> ...
>>>>> diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
>>>>> index 33677ea..f6c1543 100644
>>>>> --- a/drivers/xen/privcmd.c
>>>>> +++ b/drivers/xen/privcmd.c
>>>>> @@ -612,8 +612,11 @@ static void unlock_pages(struct page *pages[], 
>>>>> unsigned int nr_pages)
>>>>>    {
>>>>>        unsigned int i;
>>>>>
>>>>> -     for (i = 0; i < nr_pages; i++)
>>>>> +     for (i = 0; i < nr_pages; i++) {
>>>>> +             if (!PageDirty(pages[i]))
>>>>> +                     set_page_dirty_lock(pages[i]);
>>>>
>>>> With put_page() directly following I think you should be able to use
>>>> set_page_dirty() instead, as there is obviously a reference to the page
>>>> existing.
>>>
>>> Patch [3/3] will convert above codes to use 
>>> unpin_user_pages_dirty_lock()
>>> which internally do the same check. So I thought to keep linux-stable 
>>> and
>>> linux-next code in sync. John had a similar concern [1] and later 
>>> agreed to keep
>>> this check.
>>>
>>> Shall I keep this check ?  No ?
> 
> It doesn't matter *too* much, because patch 3/3 fixes up everything by
> changing it all to unpin_user_pages_dirty_lock(). However, there is 
> something
> to be said for having correct interim patches, too. :)  Details:
> 
>>>
>>> [1] 
>>> https://lore.kernel.org/xen-devel/a750e5e5-fd5d-663b-c5fd-261d7c939ba7@nvidia.com/ 
>>>
>>
>> I wasn't referring to checking PageDirty(), but to the use of
>> set_page_dirty_lock().
>>
>> Looking at the comment just before the implementation of
>> set_page_dirty_lock() suggests that it is fine to use set_page_dirty()
>> instead (so not calling lock_page()).
> 
> 
> no no, that's a misreading of the comment. Unless this xen/privcmd code has
> somehow taken a reference on page->mapping->host (which I do *not* think is
> the case), then it is still racy to call set_page_dirty() here. Instead,
> set_page_dirty_lock() should be used.

Ah, okay. Thanks for the clarification.

So you can add my

Reviewed-by: Juergen Gross <jgross@suse.com>


Juergen