[PATCH 0/5] v2 xfs refcount conversions

[PATCH 0/5] v2 xfs refcount conversions

[Elena Reshetova]

Hi Dave,

Following our previous conversation I have updated each
patch with a highlight of potential problematic places
with regards to the memory ordering guarantees, as well
as link to the document I am working on to clarify the
matter.

While I didn't see any issues with these 5 particular
conversions, of course I don't understand code anywhere
good enough to make the final claim that "everything is
100% ok".

Please let me know what kind of other clarification or
anything is needed that can help you to make the needed
verification.

Best Regards,
Elena.



Changes in v2:
  - each patch now has a clarification on potential
    problematic places with regards to the memory ordering
    guarantees as well as reference to a hopefully soon to be
    merged more extensive documentation on the
    differences between atomic_t and refcount_t on ordering.
  - Rebase on linux-next master.


This series, for xfs, replaces atomic_t reference
counters with the new refcount_t type and API (see include/linux/refcount.h).
By doing this we prevent intentional or accidental
underflows or overflows that can lead to use-after-free vulnerabilities.

The patches are fully independent and can be cherry-picked separately.
If there are no objections to the patches, please merge them via respective trees.

Elena Reshetova (5):
  fs, xfs: convert xfs_bui_log_item.bui_refcount from atomic_t to
    refcount_t
  fs, xfs: convert xfs_efi_log_item.efi_refcount from atomic_t to
    refcount_t
  fs, xfs: convert xlog_ticket.t_ref from atomic_t to refcount_t
  fs, xfs: convert xfs_cui_log_item.cui_refcount from atomic_t to
    refcount_t
  fs, xfs: convert xfs_rui_log_item.rui_refcount from atomic_t to
    refcount_t

 fs/xfs/xfs_bmap_item.c     |  6 +++---
 fs/xfs/xfs_bmap_item.h     |  2 +-
 fs/xfs/xfs_extfree_item.c  |  6 +++---
 fs/xfs/xfs_extfree_item.h  |  2 +-
 fs/xfs/xfs_linux.h         |  1 +
 fs/xfs/xfs_log.c           | 10 +++++-----
 fs/xfs/xfs_log_priv.h      |  2 +-
 fs/xfs/xfs_refcount_item.c |  6 +++---
 fs/xfs/xfs_refcount_item.h |  2 +-
 fs/xfs/xfs_rmap_item.c     |  6 +++---
 fs/xfs/xfs_rmap_item.h     |  2 +-
 11 files changed, 23 insertions(+), 22 deletions(-)

-- 
2.7.4

[PATCH 1/5] fs, xfs: convert xfs_bui_log_item.bui_refcount from atomic_t to refcount_t

[Elena Reshetova]

atomic_t variables are currently used to implement reference
counters with the following properties:
 - counter is initialized to 1 using atomic_set()
 - a resource is freed upon counter reaching zero
 - once counter reaches zero, its further
   increments aren't allowed
 - counter schema uses basic atomic operations
   (set, inc, inc_not_zero, dec_and_test, etc.)

Such atomic variables should be converted to a newly provided
refcount_t type and API that prevents accidental counter overflows
and underflows. This is important since overflows and underflows
can lead to use-after-free situation and be exploitable.

The variable xfs_bui_log_item.bui_refcount is used as pure reference counter.
Convert it to refcount_t and fix up the operations.

**Important note for maintainers:

Some functions from refcount_t API defined in lib/refcount.c
have different memory ordering guarantees than their atomic
counterparts.
The full comparison can be seen in
https://lkml.org/lkml/2017/11/15/57 and it is hopefully soon
in state to be merged to the documentation tree.
Normally the differences should not matter since refcount_t provides
enough guarantees to satisfy the refcounting use cases, but in
some rare cases it might matter.
Please double check that you don't have some undocumented
memory guarantees for this variable usage.

For the xfs_bui_log_item.bui_refcount it might make a difference
in following places:
 - xfs_bui_release(): decrement in refcount_dec_and_test() only
   provides RELEASE ordering and control dependency on success
   vs. fully ordered atomic counterpart

Suggested-by: Kees Cook <keescook@chromium.org>
Reviewed-by: David Windsor <dwindsor@gmail.com>
Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
---
 fs/xfs/xfs_bmap_item.c | 6 +++---
 fs/xfs/xfs_bmap_item.h | 2 +-
 fs/xfs/xfs_linux.h     | 1 +
 3 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/fs/xfs/xfs_bmap_item.c b/fs/xfs/xfs_bmap_item.c
index dd136f7..0f2e3d8 100644
--- a/fs/xfs/xfs_bmap_item.c
+++ b/fs/xfs/xfs_bmap_item.c
@@ -201,7 +201,7 @@ xfs_bui_init(
 	buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
 	buip->bui_format.bui_id = (uintptr_t)(void *)buip;
 	atomic_set(&buip->bui_next_extent, 0);
-	atomic_set(&buip->bui_refcount, 2);
+	refcount_set(&buip->bui_refcount, 2);
 
 	return buip;
 }
@@ -217,8 +217,8 @@ void
 xfs_bui_release(
 	struct xfs_bui_log_item	*buip)
 {
-	ASSERT(atomic_read(&buip->bui_refcount) > 0);
-	if (atomic_dec_and_test(&buip->bui_refcount)) {
+	ASSERT(refcount_read(&buip->bui_refcount) > 0);
+	if (refcount_dec_and_test(&buip->bui_refcount)) {
 		xfs_trans_ail_remove(&buip->bui_item, SHUTDOWN_LOG_IO_ERROR);
 		xfs_bui_item_free(buip);
 	}
diff --git a/fs/xfs/xfs_bmap_item.h b/fs/xfs/xfs_bmap_item.h
index c867daa..7048b14 100644
--- a/fs/xfs/xfs_bmap_item.h
+++ b/fs/xfs/xfs_bmap_item.h
@@ -61,7 +61,7 @@ struct kmem_zone;
  */
 struct xfs_bui_log_item {
 	struct xfs_log_item		bui_item;
-	atomic_t			bui_refcount;
+	refcount_t			bui_refcount;
 	atomic_t			bui_next_extent;
 	unsigned long			bui_flags;	/* misc flags */
 	struct xfs_bui_log_format	bui_format;
diff --git a/fs/xfs/xfs_linux.h b/fs/xfs/xfs_linux.h
index 99562ec..31658ac 100644
--- a/fs/xfs/xfs_linux.h
+++ b/fs/xfs/xfs_linux.h
@@ -20,6 +20,7 @@
 
 #include <linux/types.h>
 #include <linux/uuid.h>
+#include <linux/refcount.h>
 
 /*
  * Kernel specific type declarations for XFS
-- 
2.7.4

[PATCH 2/5] fs, xfs: convert xfs_efi_log_item.efi_refcount from atomic_t to refcount_t

[Elena Reshetova]

atomic_t variables are currently used to implement reference
counters with the following properties:
 - counter is initialized to 1 using atomic_set()
 - a resource is freed upon counter reaching zero
 - once counter reaches zero, its further
   increments aren't allowed
 - counter schema uses basic atomic operations
   (set, inc, inc_not_zero, dec_and_test, etc.)

Such atomic variables should be converted to a newly provided
refcount_t type and API that prevents accidental counter overflows
and underflows. This is important since overflows and underflows
can lead to use-after-free situation and be exploitable.

The variable xfs_efi_log_item.efi_refcount is used as pure reference counter.
Convert it to refcount_t and fix up the operations.

**Important note for maintainers:

Some functions from refcount_t API defined in lib/refcount.c
have different memory ordering guarantees than their atomic
counterparts.
The full comparison can be seen in
https://lkml.org/lkml/2017/11/15/57 and it is hopefully soon
in state to be merged to the documentation tree.
Normally the differences should not matter since refcount_t provides
enough guarantees to satisfy the refcounting use cases, but in
some rare cases it might matter.
Please double check that you don't have some undocumented
memory guarantees for this variable usage.

For the xfs_efi_log_item.efi_refcount it might make a difference
in following places:
 - xfs_efi_release(): decrement in refcount_dec_and_test() only
   provides RELEASE ordering and control dependency on success
   vs. fully ordered atomic counterpart

Suggested-by: Kees Cook <keescook@chromium.org>
Reviewed-by: David Windsor <dwindsor@gmail.com>
Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
---
 fs/xfs/xfs_extfree_item.c | 6 +++---
 fs/xfs/xfs_extfree_item.h | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/fs/xfs/xfs_extfree_item.c b/fs/xfs/xfs_extfree_item.c
index 44f8c54..dfe2bcb 100644
--- a/fs/xfs/xfs_extfree_item.c
+++ b/fs/xfs/xfs_extfree_item.c
@@ -220,7 +220,7 @@ xfs_efi_init(
 	efip->efi_format.efi_nextents = nextents;
 	efip->efi_format.efi_id = (uintptr_t)(void *)efip;
 	atomic_set(&efip->efi_next_extent, 0);
-	atomic_set(&efip->efi_refcount, 2);
+	refcount_set(&efip->efi_refcount, 2);
 
 	return efip;
 }
@@ -290,8 +290,8 @@ void
 xfs_efi_release(
 	struct xfs_efi_log_item	*efip)
 {
-	ASSERT(atomic_read(&efip->efi_refcount) > 0);
-	if (atomic_dec_and_test(&efip->efi_refcount)) {
+	ASSERT(refcount_read(&efip->efi_refcount) > 0);
+	if (refcount_dec_and_test(&efip->efi_refcount)) {
 		xfs_trans_ail_remove(&efip->efi_item, SHUTDOWN_LOG_IO_ERROR);
 		xfs_efi_item_free(efip);
 	}
diff --git a/fs/xfs/xfs_extfree_item.h b/fs/xfs/xfs_extfree_item.h
index a32c794..fadf736 100644
--- a/fs/xfs/xfs_extfree_item.h
+++ b/fs/xfs/xfs_extfree_item.h
@@ -64,7 +64,7 @@ struct kmem_zone;
  */
 typedef struct xfs_efi_log_item {
 	xfs_log_item_t		efi_item;
-	atomic_t		efi_refcount;
+	refcount_t		efi_refcount;
 	atomic_t		efi_next_extent;
 	unsigned long		efi_flags;	/* misc flags */
 	xfs_efi_log_format_t	efi_format;
-- 
2.7.4

[PATCH 3/5] fs, xfs: convert xlog_ticket.t_ref from atomic_t to refcount_t

[Elena Reshetova]

atomic_t variables are currently used to implement reference
counters with the following properties:
 - counter is initialized to 1 using atomic_set()
 - a resource is freed upon counter reaching zero
 - once counter reaches zero, its further
   increments aren't allowed
 - counter schema uses basic atomic operations
   (set, inc, inc_not_zero, dec_and_test, etc.)

Such atomic variables should be converted to a newly provided
refcount_t type and API that prevents accidental counter overflows
and underflows. This is important since overflows and underflows
can lead to use-after-free situation and be exploitable.

The variable xlog_ticket.t_ref is used as pure reference counter.
Convert it to refcount_t and fix up the operations.

**Important note for maintainers:

Some functions from refcount_t API defined in lib/refcount.c
have different memory ordering guarantees than their atomic
counterparts.
The full comparison can be seen in
https://lkml.org/lkml/2017/11/15/57 and it is hopefully soon
in state to be merged to the documentation tree.
Normally the differences should not matter since refcount_t provides
enough guarantees to satisfy the refcounting use cases, but in
some rare cases it might matter.
Please double check that you don't have some undocumented
memory guarantees for this variable usage.

For the xlog_ticket.t_ref it might make a difference
in following places:
 - xfs_log_ticket_put(): decrement in refcount_dec_and_test() only
   provides RELEASE ordering and control dependency on success
   vs. fully ordered atomic counterpart

Suggested-by: Kees Cook <keescook@chromium.org>
Reviewed-by: David Windsor <dwindsor@gmail.com>
Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
---
 fs/xfs/xfs_log.c      | 10 +++++-----
 fs/xfs/xfs_log_priv.h |  2 +-
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
index a503af9..6e1528f 100644
--- a/fs/xfs/xfs_log.c
+++ b/fs/xfs/xfs_log.c
@@ -3600,8 +3600,8 @@ void
 xfs_log_ticket_put(
 	xlog_ticket_t	*ticket)
 {
-	ASSERT(atomic_read(&ticket->t_ref) > 0);
-	if (atomic_dec_and_test(&ticket->t_ref))
+	ASSERT(refcount_read(&ticket->t_ref) > 0);
+	if (refcount_dec_and_test(&ticket->t_ref))
 		kmem_zone_free(xfs_log_ticket_zone, ticket);
 }
 
@@ -3609,8 +3609,8 @@ xlog_ticket_t *
 xfs_log_ticket_get(
 	xlog_ticket_t	*ticket)
 {
-	ASSERT(atomic_read(&ticket->t_ref) > 0);
-	atomic_inc(&ticket->t_ref);
+	ASSERT(refcount_read(&ticket->t_ref) > 0);
+	refcount_inc(&ticket->t_ref);
 	return ticket;
 }
 
@@ -3732,7 +3732,7 @@ xlog_ticket_alloc(
 
 	unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes);
 
-	atomic_set(&tic->t_ref, 1);
+	refcount_set(&tic->t_ref, 1);
 	tic->t_task		= current;
 	INIT_LIST_HEAD(&tic->t_queue);
 	tic->t_unit_res		= unit_res;
diff --git a/fs/xfs/xfs_log_priv.h b/fs/xfs/xfs_log_priv.h
index 1299759..7482145 100644
--- a/fs/xfs/xfs_log_priv.h
+++ b/fs/xfs/xfs_log_priv.h
@@ -168,7 +168,7 @@ typedef struct xlog_ticket {
 	struct list_head   t_queue;	 /* reserve/write queue */
 	struct task_struct *t_task;	 /* task that owns this ticket */
 	xlog_tid_t	   t_tid;	 /* transaction identifier	 : 4  */
-	atomic_t	   t_ref;	 /* ticket reference count       : 4  */
+	refcount_t	   t_ref;	 /* ticket reference count       : 4  */
 	int		   t_curr_res;	 /* current reservation in bytes : 4  */
 	int		   t_unit_res;	 /* unit reservation in bytes    : 4  */
 	char		   t_ocnt;	 /* original count		 : 1  */
-- 
2.7.4

[PATCH 4/5] fs, xfs: convert xfs_cui_log_item.cui_refcount from atomic_t to refcount_t

[Elena Reshetova]

atomic_t variables are currently used to implement reference
counters with the following properties:
 - counter is initialized to 1 using atomic_set()
 - a resource is freed upon counter reaching zero
 - once counter reaches zero, its further
   increments aren't allowed
 - counter schema uses basic atomic operations
   (set, inc, inc_not_zero, dec_and_test, etc.)

Such atomic variables should be converted to a newly provided
refcount_t type and API that prevents accidental counter overflows
and underflows. This is important since overflows and underflows
can lead to use-after-free situation and be exploitable.

The variable xfs_cui_log_item.cui_refcount is used as pure reference counter.
Convert it to refcount_t and fix up the operations.

**Important note for maintainers:

Some functions from refcount_t API defined in lib/refcount.c
have different memory ordering guarantees than their atomic
counterparts.
The full comparison can be seen in
https://lkml.org/lkml/2017/11/15/57 and it is hopefully soon
in state to be merged to the documentation tree.
Normally the differences should not matter since refcount_t provides
enough guarantees to satisfy the refcounting use cases, but in
some rare cases it might matter.
Please double check that you don't have some undocumented
memory guarantees for this variable usage.

For the xfs_cui_log_item.cui_refcount it might make a difference
in following places:
 - xfs_cui_release(): decrement in refcount_dec_and_test() only
   provides RELEASE ordering and control dependency on success
   vs. fully ordered atomic counterpart

Suggested-by: Kees Cook <keescook@chromium.org>
Reviewed-by: David Windsor <dwindsor@gmail.com>
Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
---
 fs/xfs/xfs_refcount_item.c | 6 +++---
 fs/xfs/xfs_refcount_item.h | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/fs/xfs/xfs_refcount_item.c b/fs/xfs/xfs_refcount_item.c
index 8f2e2fa..004d002 100644
--- a/fs/xfs/xfs_refcount_item.c
+++ b/fs/xfs/xfs_refcount_item.c
@@ -205,7 +205,7 @@ xfs_cui_init(
 	cuip->cui_format.cui_nextents = nextents;
 	cuip->cui_format.cui_id = (uintptr_t)(void *)cuip;
 	atomic_set(&cuip->cui_next_extent, 0);
-	atomic_set(&cuip->cui_refcount, 2);
+	refcount_set(&cuip->cui_refcount, 2);
 
 	return cuip;
 }
@@ -221,8 +221,8 @@ void
 xfs_cui_release(
 	struct xfs_cui_log_item	*cuip)
 {
-	ASSERT(atomic_read(&cuip->cui_refcount) > 0);
-	if (atomic_dec_and_test(&cuip->cui_refcount)) {
+	ASSERT(refcount_read(&cuip->cui_refcount) > 0);
+	if (refcount_dec_and_test(&cuip->cui_refcount)) {
 		xfs_trans_ail_remove(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR);
 		xfs_cui_item_free(cuip);
 	}
diff --git a/fs/xfs/xfs_refcount_item.h b/fs/xfs/xfs_refcount_item.h
index 5b74ddd..abc0377 100644
--- a/fs/xfs/xfs_refcount_item.h
+++ b/fs/xfs/xfs_refcount_item.h
@@ -63,7 +63,7 @@ struct kmem_zone;
  */
 struct xfs_cui_log_item {
 	struct xfs_log_item		cui_item;
-	atomic_t			cui_refcount;
+	refcount_t			cui_refcount;
 	atomic_t			cui_next_extent;
 	unsigned long			cui_flags;	/* misc flags */
 	struct xfs_cui_log_format	cui_format;
-- 
2.7.4

[PATCH 5/5] fs, xfs: convert xfs_rui_log_item.rui_refcount from atomic_t to refcount_t

[Elena Reshetova]

atomic_t variables are currently used to implement reference
counters with the following properties:
 - counter is initialized to 1 using atomic_set()
 - a resource is freed upon counter reaching zero
 - once counter reaches zero, its further
   increments aren't allowed
 - counter schema uses basic atomic operations
   (set, inc, inc_not_zero, dec_and_test, etc.)

Such atomic variables should be converted to a newly provided
refcount_t type and API that prevents accidental counter overflows
and underflows. This is important since overflows and underflows
can lead to use-after-free situation and be exploitable.

The variable xfs_rui_log_item.rui_refcount is used as pure reference counter.
Convert it to refcount_t and fix up the operations.

**Important note for maintainers:

Some functions from refcount_t API defined in lib/refcount.c
have different memory ordering guarantees than their atomic
counterparts.
The full comparison can be seen in
https://lkml.org/lkml/2017/11/15/57 and it is hopefully soon
in state to be merged to the documentation tree.
Normally the differences should not matter since refcount_t provides
enough guarantees to satisfy the refcounting use cases, but in
some rare cases it might matter.
Please double check that you don't have some undocumented
memory guarantees for this variable usage.

For the xfs_rui_log_item.rui_refcount it might make a difference
in following places:
 - xfs_rui_release(): decrement in refcount_dec_and_test() only
   provides RELEASE ordering and control dependency on success
   vs. fully ordered atomic counterpart

Suggested-by: Kees Cook <keescook@chromium.org>
Reviewed-by: David Windsor <dwindsor@gmail.com>
Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
---
 fs/xfs/xfs_rmap_item.c | 6 +++---
 fs/xfs/xfs_rmap_item.h | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/fs/xfs/xfs_rmap_item.c b/fs/xfs/xfs_rmap_item.c
index f3b139c..f914829f 100644
--- a/fs/xfs/xfs_rmap_item.c
+++ b/fs/xfs/xfs_rmap_item.c
@@ -204,7 +204,7 @@ xfs_rui_init(
 	ruip->rui_format.rui_nextents = nextents;
 	ruip->rui_format.rui_id = (uintptr_t)(void *)ruip;
 	atomic_set(&ruip->rui_next_extent, 0);
-	atomic_set(&ruip->rui_refcount, 2);
+	refcount_set(&ruip->rui_refcount, 2);
 
 	return ruip;
 }
@@ -243,8 +243,8 @@ void
 xfs_rui_release(
 	struct xfs_rui_log_item	*ruip)
 {
-	ASSERT(atomic_read(&ruip->rui_refcount) > 0);
-	if (atomic_dec_and_test(&ruip->rui_refcount)) {
+	ASSERT(refcount_read(&ruip->rui_refcount) > 0);
+	if (refcount_dec_and_test(&ruip->rui_refcount)) {
 		xfs_trans_ail_remove(&ruip->rui_item, SHUTDOWN_LOG_IO_ERROR);
 		xfs_rui_item_free(ruip);
 	}
diff --git a/fs/xfs/xfs_rmap_item.h b/fs/xfs/xfs_rmap_item.h
index 340c968..1e425a9 100644
--- a/fs/xfs/xfs_rmap_item.h
+++ b/fs/xfs/xfs_rmap_item.h
@@ -64,7 +64,7 @@ struct kmem_zone;
  */
 struct xfs_rui_log_item {
 	struct xfs_log_item		rui_item;
-	atomic_t			rui_refcount;
+	refcount_t			rui_refcount;
 	atomic_t			rui_next_extent;
 	unsigned long			rui_flags;	/* misc flags */
 	struct xfs_rui_log_format	rui_format;
-- 
2.7.4