btrfs: add framework to handle device flush error as a volume

This adds comments to the flush error handling part of the code, and
hopes to maintain the same logic with a framework which can be used to
handle the errors at the volume level.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 5f678dcb20e634d936c56f6bdcba0c86b624c21b..bafdd2fe8f88c015593fe1bd123390e62d34f34c 100644 (file)
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -3509,6 +3509,10 @@ static int write_dev_flush(struct btrfs_device *device, int wait)
        if (wait) {
                bio = device->flush_bio;
                if (!bio)
+                       /*
+                        * This means the alloc has failed with ENOMEM, however
+                        * here we return 0, as its not a device error.
+                        */
                        return 0;
 
                wait_for_completion(&device->flush_wait);
@@ -3548,6 +3552,32 @@ static int write_dev_flush(struct btrfs_device *device, int wait)
        return 0;
 }
 
+static int check_barrier_error(struct btrfs_fs_devices *fsdevs)
+{
+       int submit_flush_error = 0;
+       int dev_flush_error = 0;
+       struct btrfs_device *dev;
+       int tolerance;
+
+       list_for_each_entry_rcu(dev, &fsdevs->devices, dev_list) {
+               if (!dev->bdev) {
+                       submit_flush_error++;
+                       dev_flush_error++;
+                       continue;
+               }
+               if (dev->last_flush_error == -ENOMEM)
+                       submit_flush_error++;
+               if (dev->last_flush_error && dev->last_flush_error != -ENOMEM)
+                       dev_flush_error++;
+       }
+
+       tolerance = fsdevs->fs_info->num_tolerated_disk_barrier_failures;
+       if (submit_flush_error > tolerance || dev_flush_error > tolerance)
+               return -EIO;
+
+       return 0;
+}
+
 /*
  * send an empty flush down to each device in parallel,
  * then wait for them
@@ -3575,6 +3605,7 @@ static int barrier_all_devices(struct btrfs_fs_info *info)
                ret = write_dev_flush(dev, 0);
                if (ret)
                        errors_send++;
+               dev->last_flush_error = ret;
        }
 
        /* wait for all the barriers */
@@ -3589,12 +3620,30 @@ static int barrier_all_devices(struct btrfs_fs_info *info)
                        continue;
 
                ret = write_dev_flush(dev, 1);
-               if (ret)
+               if (ret) {
+                       dev->last_flush_error = ret;
                        errors_wait++;
+               }
+       }
+
+       /*
+        * Try hard in case of flush. Lets say, in RAID1 we have
+        * the following situation
+        *  dev1: EIO dev2: ENOMEM
+        * this is not a fatal error as we hope to recover from
+        * ENOMEM in the next attempt to flush.
+        * But the following is considered as fatal
+        *  dev1: ENOMEM dev2: ENOMEM
+        *  dev1: bdev == NULL dev2: ENOMEM
+        */
+       if (errors_send || errors_wait) {
+               /*
+                * At some point we need the status of all disks
+                * to arrive at the volume status. So error checking
+                * is being pushed to a separate loop.
+                */
+               return check_barrier_error(info->fs_devices);
        }
-       if (errors_send > info->num_tolerated_disk_barrier_failures ||
-           errors_wait > info->num_tolerated_disk_barrier_failures)
-               return -EIO;
        return 0;
 }
 

diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index c7d0fbc915cabdee7cfec437e18f795ace77abbb..094521729df12a6c99d1ee91e1f3aabc693da023 100644 (file)
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -74,6 +74,7 @@ struct btrfs_device {
        int missing;
        int can_discard;
        int is_tgtdev_for_dev_replace;
+       int last_flush_error;
 
 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED
        seqcount_t data_seqcount;