block: Hold invalidate_lock in BLKRESETZONE ioctl

commit 86399ea071099ec8ee0a83ac9ad67f7df96a50ad upstream.

When BLKRESETZONE ioctl and data read race, the data read leaves stale
page cache. The commit e5113505904e ("block: Discard page cache of zone
reset target range") added page cache truncation to avoid stale page
cache after the ioctl. However, the stale page cache still can be read
during the reset zone operation for the ioctl. To avoid the stale page
cache completely, hold invalidate_lock of the block device file mapping.

Fixes: e5113505904e ("block: Discard page cache of zone reset target range")
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Cc: stable@vger.kernel.org # v5.15
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20211111085238.942492-1-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/block/blk-zoned.c b/block/blk-zoned.c
index 1d0c76c18fc523e95ed688eda4120cb7d6b350f1..774ecc598bee28a882c65138470735dad0cd3e29 100644 (file)
--- a/block/blk-zoned.c
+++ b/block/blk-zoned.c
@@ -429,9 +429,10 @@ int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
                op = REQ_OP_ZONE_RESET;
 
                /* Invalidate the page cache, including dirty pages. */
+               filemap_invalidate_lock(bdev->bd_inode->i_mapping);
                ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
                if (ret)
-                       return ret;
+                       goto fail;
                break;
        case BLKOPENZONE:
                op = REQ_OP_ZONE_OPEN;
@@ -449,15 +450,9 @@ int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
        ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
                               GFP_KERNEL);
 
-       /*
-        * Invalidate the page cache again for zone reset: writes can only be
-        * direct for zoned devices so concurrent writes would not add any page
-        * to the page cache after/during reset. The page cache may be filled
-        * again due to concurrent reads though and dropping the pages for
-        * these is fine.
-        */
-       if (!ret && cmd == BLKRESETZONE)
-               ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
+fail:
+       if (cmd == BLKRESETZONE)
+               filemap_invalidate_unlock(bdev->bd_inode->i_mapping);
 
        return ret;
 }