xen/blkfront: cleanup stale persistent grants

Add a periodic cleanup function to remove old persistent grants which
are no longer in use on the backend side. This avoids starvation in
case there are lots of persistent grants for a device which no longer
is involved in I/O business.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 8986ada..a2a395f 100644 (file)
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -46,6 +46,7 @@
 #include <linux/scatterlist.h>
 #include <linux/bitmap.h>
 #include <linux/list.h>
+#include <linux/workqueue.h>
 
 #include <xen/xen.h>
 #include <xen/xenbus.h>
@@ -121,6 +122,8 @@ static inline struct blkif_req *blkif_req(struct request *rq)
 
 static DEFINE_MUTEX(blkfront_mutex);
 static const struct block_device_operations xlvbd_block_fops;
+static struct delayed_work blkfront_work;
+static LIST_HEAD(info_list);
 
 /*
  * Maximum number of segments in indirect requests, the actual value used by
@@ -216,6 +219,7 @@ struct blkfront_info
        /* Save uncomplete reqs and bios for migration. */
        struct list_head requests;
        struct bio_list bio_list;
+       struct list_head info_list;
 };
 
 static unsigned int nr_minors;
@@ -1759,6 +1763,12 @@ abort_transaction:
        return err;
 }
 
+static void free_info(struct blkfront_info *info)
+{
+       list_del(&info->info_list);
+       kfree(info);
+}
+
 /* Common code used when first setting up, and when resuming. */
 static int talk_to_blkback(struct xenbus_device *dev,
                           struct blkfront_info *info)
@@ -1880,7 +1890,10 @@ again:
  destroy_blkring:
        blkif_free(info, 0);
 
-       kfree(info);
+       mutex_lock(&blkfront_mutex);
+       free_info(info);
+       mutex_unlock(&blkfront_mutex);
+
        dev_set_drvdata(&dev->dev, NULL);
 
        return err;
@@ -1991,6 +2004,10 @@ static int blkfront_probe(struct xenbus_device *dev,
        info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
        dev_set_drvdata(&dev->dev, info);
 
+       mutex_lock(&blkfront_mutex);
+       list_add(&info->info_list, &info_list);
+       mutex_unlock(&blkfront_mutex);
+
        return 0;
 }
 
@@ -2301,6 +2318,12 @@ static void blkfront_gather_backend_features(struct blkfront_info *info)
        if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
                indirect_segments = 0;
        info->max_indirect_segments = indirect_segments;
+
+       if (info->feature_persistent) {
+               mutex_lock(&blkfront_mutex);
+               schedule_delayed_work(&blkfront_work, HZ * 10);
+               mutex_unlock(&blkfront_mutex);
+       }
 }
 
 /*
@@ -2482,7 +2505,9 @@ static int blkfront_remove(struct xenbus_device *xbdev)
        mutex_unlock(&info->mutex);
 
        if (!bdev) {
-               kfree(info);
+               mutex_lock(&blkfront_mutex);
+               free_info(info);
+               mutex_unlock(&blkfront_mutex);
                return 0;
        }
 
@@ -2502,7 +2527,9 @@ static int blkfront_remove(struct xenbus_device *xbdev)
        if (info && !bdev->bd_openers) {
                xlvbd_release_gendisk(info);
                disk->private_data = NULL;
-               kfree(info);
+               mutex_lock(&blkfront_mutex);
+               free_info(info);
+               mutex_unlock(&blkfront_mutex);
        }
 
        mutex_unlock(&bdev->bd_mutex);
@@ -2585,7 +2612,7 @@ static void blkif_release(struct gendisk *disk, fmode_t mode)
                dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
                xlvbd_release_gendisk(info);
                disk->private_data = NULL;
-               kfree(info);
+               free_info(info);
        }
 
 out:
@@ -2618,6 +2645,61 @@ static struct xenbus_driver blkfront_driver = {
        .is_ready = blkfront_is_ready,
 };
 
+static void purge_persistent_grants(struct blkfront_info *info)
+{
+       unsigned int i;
+       unsigned long flags;
+
+       for (i = 0; i < info->nr_rings; i++) {
+               struct blkfront_ring_info *rinfo = &info->rinfo[i];
+               struct grant *gnt_list_entry, *tmp;
+
+               spin_lock_irqsave(&rinfo->ring_lock, flags);
+
+               if (rinfo->persistent_gnts_c == 0) {
+                       spin_unlock_irqrestore(&rinfo->ring_lock, flags);
+                       continue;
+               }
+
+               list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
+                                        node) {
+                       if (gnt_list_entry->gref == GRANT_INVALID_REF ||
+                           gnttab_query_foreign_access(gnt_list_entry->gref))
+                               continue;
+
+                       list_del(&gnt_list_entry->node);
+                       gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
+                       rinfo->persistent_gnts_c--;
+                       __free_page(gnt_list_entry->page);
+                       kfree(gnt_list_entry);
+               }
+
+               spin_unlock_irqrestore(&rinfo->ring_lock, flags);
+       }
+}
+
+static void blkfront_delay_work(struct work_struct *work)
+{
+       struct blkfront_info *info;
+       bool need_schedule_work = false;
+
+       mutex_lock(&blkfront_mutex);
+
+       list_for_each_entry(info, &info_list, info_list) {
+               if (info->feature_persistent) {
+                       need_schedule_work = true;
+                       mutex_lock(&info->mutex);
+                       purge_persistent_grants(info);
+                       mutex_unlock(&info->mutex);
+               }
+       }
+
+       if (need_schedule_work)
+               schedule_delayed_work(&blkfront_work, HZ * 10);
+
+       mutex_unlock(&blkfront_mutex);
+}
+
 static int __init xlblk_init(void)
 {
        int ret;
@@ -2650,6 +2732,8 @@ static int __init xlblk_init(void)
                return -ENODEV;
        }
 
+       INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
+
        ret = xenbus_register_frontend(&blkfront_driver);
        if (ret) {
                unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
@@ -2663,6 +2747,8 @@ module_init(xlblk_init);
 
 static void __exit xlblk_exit(void)
 {
+       cancel_delayed_work_sync(&blkfront_work);
+
        xenbus_unregister_driver(&blkfront_driver);
        unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
        kfree(minors);