rbd: prevent open of image being unmapped

Define a new rbd device flags field, manipulated using atomic bit operations. Replace the use of the current "exists" flag with a bit in this new "flags" field.

Signed-off-by Alex Elder <[email protected]>

---
  drivers/block/rbd.c |   17 ++++++++++++-----
  1 file changed, 12 insertions(+), 5 deletions(-)

drivers/block/rbd.c +9 -2

index 02002b1..9eb1631 100644
@@ -232,7 +232,7 @@ struct rbd_device {
  	spinlock_t		lock;		/* queue lock */
 
  	struct rbd_image_header	header;
-	atomic_t		exists;
+	unsigned long		flags;
  	struct rbd_spec		*spec;
 
  	char			*header_name;

@@ -260,6 +260,12 @@ struct rbd_device {
  	unsigned long		open_count;
  };
 
+/* Flag bits for rbd_dev->flags */
+
+enum rbd_dev_flags {
+	rbd_dev_flag_exists,	/* mapped snapshot has not been deleted */
+};
+
  static DEFINE_MUTEX(ctl_mutex);	  /* Serialize open/close/setup/teardown */
 
  static LIST_HEAD(rbd_dev_list);    /* devices */

@@ -756,7 +762,8 @@ static int rbd_dev_set_mapping(struct rbd_device
 *rbd_dev)
  			goto done;
  		rbd_dev->mapping.read_only = true;
  	}
-	atomic_set(&rbd_dev->exists, 1);
+	set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
+
  done:
  	return ret;

  }

drivers/block/rbd.c +2 -2

@@ -1654,7 +1661,7 @@ static void rbd_rq_fn(struct request_queue *q)
  			snapc = ceph_get_snap_context(rbd_dev->header.snapc);
  			up_read(&rbd_dev->header_rwsem);
  			rbd_assert(snapc != NULL);
-		} else if (!atomic_read(&rbd_dev->exists)) {
+		} else if (!test_bit(rbd_dev_flag_exists, &rbd_dev->flags)) {
  			rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
  			dout("request for non-existent snapshot");
  			result = -ENXIO;

@@ -2270,7 +2277,7 @@ struct rbd_device *rbd_dev_create(struct
 rbd_client *rbdc,
  		return NULL;
 
  	spin_lock_init(&rbd_dev->lock);
-	atomic_set(&rbd_dev->exists, 0);
+	rbd_dev->flags = 0;
  	INIT_LIST_HEAD(&rbd_dev->node);
  	INIT_LIST_HEAD(&rbd_dev->snaps);

  	init_rwsem(&rbd_dev->header_rwsem);

drivers/block/rbd.c +1 -1

@@ -2902,7 +2909,7 @@ static int rbd_dev_snaps_update(struct rbd_device
 *rbd_dev)
  			/* Existing snapshot not in the new snap context */
 
  			if (rbd_dev->spec->snap_id == snap->id)
-				atomic_set(&rbd_dev->exists, 0);
+				set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
  			rbd_remove_snap_dev(snap);
  			dout("%ssnap id %llu has been removed\n",

rbd_dev->spec->snap_id == snap->id ?

I see that set_bit is atomic, but I don't see that test_bit is. Am I missing a subtlety?

Quoted 77 lines

Define a new rbd device flags field, manipulated using atomic bit operations. Replace the use of the current "exists" flag with a bit in this new "flags" field.

Signed-off-by Alex Elder <[email protected]>

---
  drivers/block/rbd.c |   17 ++++++++++++-----
  1 file changed, 12 insertions(+), 5 deletions(-)

drivers/block/rbd.c +7 -1

index 02002b1..9eb1631 100644
@@ -232,7 +232,7 @@ struct rbd_device {
      spinlock_t        lock;        /* queue lock */
 
      struct rbd_image_header    header;
-    atomic_t        exists;
+    unsigned long        flags;
      struct rbd_spec        *spec;
 
      char            *header_name;

@@ -260,6 +260,12 @@ struct rbd_device {
      unsigned long        open_count;
  };
 
+/* Flag bits for rbd_dev->flags */
+
+enum rbd_dev_flags {
+    rbd_dev_flag_exists,    /* mapped snapshot has not been deleted */
+};
+
  static DEFINE_MUTEX(ctl_mutex);      /* Serialize

open/close/setup/teardown */

static LIST_HEAD(rbd_dev_list); /* devices */

drivers/block/rbd.c +2 -1

@@ -756,7 +762,8 @@ static int rbd_dev_set_mapping(struct rbd_device
 *rbd_dev)
              goto done;
          rbd_dev->mapping.read_only = true;
      }
-    atomic_set(&rbd_dev->exists, 1);
+    set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
+
  done:
      return ret;

  }

drivers/block/rbd.c +2 -2

@@ -1654,7 +1661,7 @@ static void rbd_rq_fn(struct request_queue *q)
              snapc = ceph_get_snap_context(rbd_dev->header.snapc);
              up_read(&rbd_dev->header_rwsem);
              rbd_assert(snapc != NULL);
-        } else if (!atomic_read(&rbd_dev->exists)) {
+        } else if (!test_bit(rbd_dev_flag_exists, &rbd_dev->flags)) {
              rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
              dout("request for non-existent snapshot");
              result = -ENXIO;

@@ -2270,7 +2277,7 @@ struct rbd_device *rbd_dev_create(struct
 rbd_client *rbdc,
          return NULL;
 
      spin_lock_init(&rbd_dev->lock);
-    atomic_set(&rbd_dev->exists, 0);
+    rbd_dev->flags = 0;
      INIT_LIST_HEAD(&rbd_dev->node);
      INIT_LIST_HEAD(&rbd_dev->snaps);

      init_rwsem(&rbd_dev->header_rwsem);

drivers/block/rbd.c +1 -1

@@ -2902,7 +2909,7 @@ static int rbd_dev_snaps_update(struct rbd_device
 *rbd_dev)
              /* Existing snapshot not in the new snap context */
 
              if (rbd_dev->spec->snap_id == snap->id)
-                atomic_set(&rbd_dev->exists, 0);
+                set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
              rbd_remove_snap_dev(snap);
              dout("%ssnap id %llu has been removed\n",

                  rbd_dev->spec->snap_id == snap->id ?

Quoted 2 lines · Dan Mick

I see that set_bit is atomic, but I don't see that test_bit is. Am I missing a subtlety?

That's an interesting observation. I'm certain it's safe, but I needed to research it a bit, and I still haven't verified it to my satisfaction.

I *think* (but please look over the following and see if you come to the same conclusion) that this operation doesn't need to be made atomic, because the implementation of the routines that implement the "set" operations guarantee their effects are visible once they are done.

But I'm not sure whether "visible" here means precisely that another CPU will be forced to go read the updated memory when it calls test_bit().

http://www.kernel.org/doc/Documentation/atomic_ops.txt
The section of interest can be found by looking for the sentence I'm talking about:
Likewise, the atomic bit operation must be visible globally before any subsequent memory operation is made visible.

It doesn't come right and explain it though. Please let me know what you think.

-Alex

Quoted 78 lines · Alex Elder

Quoted 77 lines

Define a new rbd device flags field, manipulated using atomic bit operations. Replace the use of the current "exists" flag with a bit in this new "flags" field.

Signed-off-by Alex Elder <[email protected]>

---
   drivers/block/rbd.c |   17 ++++++++++++-----
   1 file changed, 12 insertions(+), 5 deletions(-)

drivers/block/rbd.c +7 -1

index 02002b1..9eb1631 100644
@@ -232,7 +232,7 @@ struct rbd_device {
       spinlock_t        lock;        /* queue lock */
 
       struct rbd_image_header    header;
-    atomic_t        exists;
+    unsigned long        flags;
       struct rbd_spec        *spec;
 
       char            *header_name;

@@ -260,6 +260,12 @@ struct rbd_device {
       unsigned long        open_count;
   };
 
+/* Flag bits for rbd_dev->flags */
+
+enum rbd_dev_flags {
+    rbd_dev_flag_exists,    /* mapped snapshot has not been deleted */
+};
+
   static DEFINE_MUTEX(ctl_mutex);      /* Serialize

open/close/setup/teardown */

static LIST_HEAD(rbd_dev_list); /* devices */

drivers/block/rbd.c +2 -1

@@ -756,7 +762,8 @@ static int rbd_dev_set_mapping(struct rbd_device
 *rbd_dev)
               goto done;
           rbd_dev->mapping.read_only = true;
       }
-    atomic_set(&rbd_dev->exists, 1);
+    set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
+
   done:
       return ret;

   }

drivers/block/rbd.c +2 -2

@@ -1654,7 +1661,7 @@ static void rbd_rq_fn(struct request_queue *q)
               snapc = ceph_get_snap_context(rbd_dev->header.snapc);
               up_read(&rbd_dev->header_rwsem);
               rbd_assert(snapc != NULL);
-        } else if (!atomic_read(&rbd_dev->exists)) {
+        } else if (!test_bit(rbd_dev_flag_exists, &rbd_dev->flags)) {
               rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
               dout("request for non-existent snapshot");
               result = -ENXIO;

@@ -2270,7 +2277,7 @@ struct rbd_device *rbd_dev_create(struct
 rbd_client *rbdc,
           return NULL;
 
       spin_lock_init(&rbd_dev->lock);
-    atomic_set(&rbd_dev->exists, 0);
+    rbd_dev->flags = 0;
       INIT_LIST_HEAD(&rbd_dev->node);
       INIT_LIST_HEAD(&rbd_dev->snaps);

       init_rwsem(&rbd_dev->header_rwsem);

drivers/block/rbd.c +1 -1

@@ -2902,7 +2909,7 @@ static int rbd_dev_snaps_update(struct rbd_device
 *rbd_dev)
               /* Existing snapshot not in the new snap context */
 
               if (rbd_dev->spec->snap_id == snap->id)
-                atomic_set(&rbd_dev->exists, 0);
+                set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
               rbd_remove_snap_dev(snap);
               dout("%ssnap id %llu has been removed\n",

                   rbd_dev->spec->snap_id == snap->id ?

Quoted 2 lines · Dan Mick

I see that set_bit is atomic, but I don't see that test_bit is. Am I missing a subtlety?

That's an interesting observation. I'm certain it's safe, but I needed to research it a bit, and I still haven't verified it to my satisfaction.

I *think* (but please look over the following and see if you come to the same conclusion) that this operation doesn't need to be made atomic, because the implementation of the routines that implement the "set" operations guarantee their effects are visible once they are done.

But I'm not sure whether "visible" here means precisely that another CPU will be forced to go read the updated memory when it calls test_bit().

http://www.kernel.org/doc/Documentation/atomic_ops.txt
The section of interest can be found by looking for the sentence I'm talking about:
Likewise, the atomic bit operation must be visible globally before any subsequent memory operation is made visible.

It doesn't come right and explain it though. Please let me know what you think.

-Alex

Quoted 78 lines · Alex Elder

Quoted 77 lines

Define a new rbd device flags field, manipulated using atomic bit operations. Replace the use of the current "exists" flag with a bit in this new "flags" field.

Signed-off-by Alex Elder <[email protected]>

---
   drivers/block/rbd.c |   17 ++++++++++++-----
   1 file changed, 12 insertions(+), 5 deletions(-)

drivers/block/rbd.c +7 -1

index 02002b1..9eb1631 100644
@@ -232,7 +232,7 @@ struct rbd_device {
       spinlock_t        lock;        /* queue lock */
 
       struct rbd_image_header    header;
-    atomic_t        exists;
+    unsigned long        flags;
       struct rbd_spec        *spec;
 
       char            *header_name;

@@ -260,6 +260,12 @@ struct rbd_device {
       unsigned long        open_count;
   };
 
+/* Flag bits for rbd_dev->flags */
+
+enum rbd_dev_flags {
+    rbd_dev_flag_exists,    /* mapped snapshot has not been deleted */
+};
+
   static DEFINE_MUTEX(ctl_mutex);      /* Serialize

open/close/setup/teardown */

static LIST_HEAD(rbd_dev_list); /* devices */

drivers/block/rbd.c +2 -1

@@ -756,7 +762,8 @@ static int rbd_dev_set_mapping(struct rbd_device
 *rbd_dev)
               goto done;
           rbd_dev->mapping.read_only = true;
       }
-    atomic_set(&rbd_dev->exists, 1);
+    set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
+
   done:
       return ret;

   }

drivers/block/rbd.c +2 -2

@@ -1654,7 +1661,7 @@ static void rbd_rq_fn(struct request_queue *q)
               snapc = ceph_get_snap_context(rbd_dev->header.snapc);
               up_read(&rbd_dev->header_rwsem);
               rbd_assert(snapc != NULL);
-        } else if (!atomic_read(&rbd_dev->exists)) {
+        } else if (!test_bit(rbd_dev_flag_exists, &rbd_dev->flags)) {
               rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
               dout("request for non-existent snapshot");
               result = -ENXIO;

@@ -2270,7 +2277,7 @@ struct rbd_device *rbd_dev_create(struct
 rbd_client *rbdc,
           return NULL;
 
       spin_lock_init(&rbd_dev->lock);
-    atomic_set(&rbd_dev->exists, 0);
+    rbd_dev->flags = 0;
       INIT_LIST_HEAD(&rbd_dev->node);
       INIT_LIST_HEAD(&rbd_dev->snaps);

       init_rwsem(&rbd_dev->header_rwsem);

drivers/block/rbd.c +1 -1

@@ -2902,7 +2909,7 @@ static int rbd_dev_snaps_update(struct rbd_device
 *rbd_dev)
               /* Existing snapshot not in the new snap context */
 
               if (rbd_dev->spec->snap_id == snap->id)
-                atomic_set(&rbd_dev->exists, 0);
+                set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
               rbd_remove_snap_dev(snap);
               dout("%ssnap id %llu has been removed\n",

                   rbd_dev->spec->snap_id == snap->id ?

Quoted 18 lines · Alex Elder

Quoted 2 lines · Dan Mick

I see that set_bit is atomic, but I don't see that test_bit is. Am I missing a subtlety?

That's an interesting observation. I'm certain it's safe, but I needed to research it a bit, and I still haven't verified it to my satisfaction.

I *think* (but please look over the following and see if you come to the same conclusion) that this operation doesn't need to be made atomic, because the implementation of the routines that implement the "set" operations guarantee their effects are visible once they are done.

But I'm not sure whether "visible" here means precisely that another CPU will be forced to go read the updated memory when it calls test_bit().

Quoted 5 lines

http://www.kernel.org/doc/Documentation/atomic_ops.txt
The section of interest can be found by looking for the sentence I'm talking about:
Likewise, the atomic bit operation must be visible globally before any subsequent memory operation is made visible.

I read that differently. I think that only applies to the test_and_set style operations mentioned directly above, not set_bit.

Documentation/memory-barriers.txt confirms this interpretation:

The following operations are potential problems as they do _not_ imply memory barriers, but might be used for implementing such things as UNLOCK-class operations:

atomic_set();
set_bit();
clear_bit();
change_bit();

With these the appropriate explicit memory barrier should be used if necessary (smp_mb__before_clear_bit() for instance).

And:

Memory operations that occur after an UNLOCK operation may appear to happen before it completes.

So I think we need a memory barrier before and after set_bit for the removing flag, but we don't need barriers for the exists flag, since it's a best-effort value that can't stop already-in-flight requests.

Define a new rbd device flags field, manipulated using atomic bit operations. Replace the use of the current "exists" flag with a bit in this new "flags" field.

Signed-off-by Alex Elder <[email protected]>

---
  drivers/block/rbd.c |   17 ++++++++++++-----
  1 file changed, 12 insertions(+), 5 deletions(-)

drivers/block/rbd.c +9 -2

index 02002b1..9eb1631 100644
@@ -232,7 +232,7 @@ struct rbd_device {
  	spinlock_t		lock;		/* queue lock */
 
  	struct rbd_image_header	header;
-	atomic_t		exists;
+	unsigned long		flags;
  	struct rbd_spec		*spec;
 
  	char			*header_name;

@@ -260,6 +260,12 @@ struct rbd_device {
  	unsigned long		open_count;
  };
 
+/* Flag bits for rbd_dev->flags */
+
+enum rbd_dev_flags {
+	rbd_dev_flag_exists,	/* mapped snapshot has not been deleted */
+};
+
  static DEFINE_MUTEX(ctl_mutex);	  /* Serialize open/close/setup/teardown */
 
  static LIST_HEAD(rbd_dev_list);    /* devices */

@@ -756,7 +762,8 @@ static int rbd_dev_set_mapping(struct rbd_device
 *rbd_dev)
  			goto done;
  		rbd_dev->mapping.read_only = true;
  	}
-	atomic_set(&rbd_dev->exists, 1);
+	set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
+
  done:
  	return ret;

  }

drivers/block/rbd.c +2 -2

@@ -1654,7 +1661,7 @@ static void rbd_rq_fn(struct request_queue *q)
  			snapc = ceph_get_snap_context(rbd_dev->header.snapc);
  			up_read(&rbd_dev->header_rwsem);
  			rbd_assert(snapc != NULL);
-		} else if (!atomic_read(&rbd_dev->exists)) {
+		} else if (!test_bit(rbd_dev_flag_exists, &rbd_dev->flags)) {
  			rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
  			dout("request for non-existent snapshot");
  			result = -ENXIO;

@@ -2270,7 +2277,7 @@ struct rbd_device *rbd_dev_create(struct
 rbd_client *rbdc,
  		return NULL;
 
  	spin_lock_init(&rbd_dev->lock);
-	atomic_set(&rbd_dev->exists, 0);
+	rbd_dev->flags = 0;
  	INIT_LIST_HEAD(&rbd_dev->node);
  	INIT_LIST_HEAD(&rbd_dev->snaps);

  	init_rwsem(&rbd_dev->header_rwsem);

drivers/block/rbd.c +1 -1

@@ -2902,7 +2909,7 @@ static int rbd_dev_snaps_update(struct rbd_device
 *rbd_dev)
  			/* Existing snapshot not in the new snap context */
 
  			if (rbd_dev->spec->snap_id == snap->id)
-				atomic_set(&rbd_dev->exists, 0);
+				set_bit(rbd_dev_flag_exists, &rbd_dev->flags);
  			rbd_remove_snap_dev(snap);
  			dout("%ssnap id %llu has been removed\n",

rbd_dev->spec->snap_id == snap->id ?