On Mon, 2012-08-13 at 15:16 -0300, Herton Ronaldo Krzesinski wrote:
> After commit 3f9a5aa ("floppy: Cleanup disk->queue before caling
> put_disk() if add_disk() was never called"), if something fails in the
> add_disk loop, we unconditionally set disks[dr]->queue to NULL. But
> that's wrong, since we may have succesfully done an add_disk on some of
> the drives previously in the loop, and in this case we would end up with
> an extra reference to the disks[dr]->queue.
> 
> Add a new global array to mark "registered" disks, and use that to check
> if we did an add_disk on one of the disks already. Using an array to
> track added disks also will help to simplify/cleanup code later, as
> suggested by Vivek Goyal.
[...]

It's totally ridiculous that a driver should have to do this.  Any
registered disk should have the GENHD_FL_UP flag set... so why can't
genhd check it?  It doesn't look like floppy is the only driver affected
by this problem, either.  So I suggest the following general fix
(untested):

---
Subject: genhd: Make put_disk() safe for disks that have not been registered

Since commit 9f53d2f ('block: fix __blkdev_get and add_disk race
condition'), add_disk() adds a reference to disk->queue, which is then
dropped by disk_release().  But if a disk is destroyed without being
registered through add_disk() (or if add_disk() fails at the first
hurdle) then we have a reference imbalance.

Use the GENHD_FL_UP flag to tell whether this extra reference has been
added.  Remove the incomplete workaround from the floppy driver.

Signed-off-by: Ben Hutchings <b...@decadent.org.uk>
Cc: sta...@vger.kernel.org
---
 block/genhd.c          |    6 +++---
 drivers/block/floppy.c |   16 +---------------
 2 files changed, 4 insertions(+), 18 deletions(-)

diff --git a/block/genhd.c b/block/genhd.c
index cac7366..6201212 100644
--- a/block/genhd.c
+++ b/block/genhd.c
@@ -587,8 +587,6 @@ void add_disk(struct gendisk *disk)
        WARN_ON(disk->minors && !(disk->major || disk->first_minor));
        WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
 
-       disk->flags |= GENHD_FL_UP;
-
        retval = blk_alloc_devt(&disk->part0, &devt);
        if (retval) {
                WARN_ON(1);
@@ -596,6 +594,8 @@ void add_disk(struct gendisk *disk)
        }
        disk_to_dev(disk)->devt = devt;
 
+       disk->flags |= GENHD_FL_UP;
+
        /* ->major and ->first_minor aren't supposed to be
         * dereferenced from here on, but set them just in case.
         */
@@ -1105,7 +1105,7 @@ static void disk_release(struct device *dev)
        disk_replace_part_tbl(disk, NULL);
        free_part_stats(&disk->part0);
        free_part_info(&disk->part0);
-       if (disk->queue)
+       if (disk->queue && disk->flags & GENHD_FL_UP)
                blk_put_queue(disk->queue);
        kfree(disk);
 }
diff --git a/drivers/block/floppy.c b/drivers/block/floppy.c
index a7d6347..cf27922 100644
--- a/drivers/block/floppy.c
+++ b/drivers/block/floppy.c
@@ -4325,14 +4325,8 @@ out_unreg_blkdev:
 out_put_disk:
        while (dr--) {
                del_timer_sync(&motor_off_timer[dr]);
-               if (disks[dr]->queue) {
+               if (disks[dr]->queue)
                        blk_cleanup_queue(disks[dr]->queue);
-                       /*
-                        * put_disk() is not paired with add_disk() and
-                        * will put queue reference one extra time. fix it.
-                        */
-                       disks[dr]->queue = NULL;
-               }
                put_disk(disks[dr]);
        }
        return err;
@@ -4560,14 +4554,6 @@ static void __exit floppy_module_exit(void)
                }
                blk_cleanup_queue(disks[drive]->queue);
 
-               /*
-                * These disks have not called add_disk().  Don't put down
-                * queue reference in put_disk().
-                */
-               if (!(allowed_drive_mask & (1 << drive)) ||
-                   fdc_state[FDC(drive)].version == FDC_NONE)
-                       disks[drive]->queue = NULL;
-
                put_disk(disks[drive]);
        }
 


-- 
Ben Hutchings
I say we take off; nuke the site from orbit.  It's the only way to be sure.

Attachment: signature.asc
Description: This is a digitally signed message part

Reply via email to