Fix zpl_mount() deadlock

Commit 93b43af10 inadvertently introduced the following scenario which
can result in a deadlock.  This issue was most easily reproduced by
LXD containers using a ZFS storage backend but should be reproducible
under any workload which is frequently mounting and unmounting.

-- THREAD A --
spa_sync()
  spa_sync_upgrades()
    rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); <- Waiting on B

-- THREAD B --
mount_fs()
  zpl_mount()
    zpl_mount_impl()
      dmu_objset_hold()
        dmu_objset_hold_flags()
          dsl_pool_hold()
            dsl_pool_config_enter()
              rrw_enter(&dp->dp_config_rwlock, RW_READER, tag);
    sget()
      sget_userns()
        grab_super()
          down_write(&s->s_umount); <- Waiting on C

-- THREAD C --
cleanup_mnt()
  deactivate_super()
    down_write(&s->s_umount);
    deactivate_locked_super()
      zpl_kill_sb()
        kill_anon_super()
          generic_shutdown_super()
            sync_filesystem()
              zpl_sync_fs()
                zfs_sync()
                  zil_commit()
                    txg_wait_synced() <- Waiting on A

Reviewed by: Alek Pinchuk <apinchuk@datto.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #7598
Closes #7659
Closes #7691
Closes #7693

diff --git a/include/sys/zfs_vfsops.h b/include/sys/zfs_vfsops.h
index 2326da42218315629b9e7caeca08a9c1af70515b..927153b25af96c44bd1007deb00c426111c78277 100644 (file)
--- a/include/sys/zfs_vfsops.h
+++ b/include/sys/zfs_vfsops.h
@@ -32,6 +32,7 @@
 #include <sys/zil.h>
 #include <sys/sa.h>
 #include <sys/rrwlock.h>
+#include <sys/dsl_dataset.h>
 #include <sys/zfs_ioctl.h>
 
 #ifdef __cplusplus

diff --git a/module/zfs/zpl_super.c b/module/zfs/zpl_super.c
index fc10271b787fbd3691314eaa840f683958e7bf2d..5c426b0a9fb2ec715104ba69f7681bcba9662277 100644 (file)
--- a/module/zfs/zpl_super.c
+++ b/module/zfs/zpl_super.c
@@ -271,8 +271,17 @@ zpl_mount_impl(struct file_system_type *fs_type, int flags, zfs_mnt_t *zm)
        if (err)
                return (ERR_PTR(-err));
 
+       /*
+        * The dsl pool lock must be released prior to calling sget().
+        * It is possible sget() may block on the lock in grab_super()
+        * while deactivate_super() holds that same lock and waits for
+        * a txg sync.  If the dsl_pool lock is held over over sget()
+        * this can prevent the pool sync and cause a deadlock.
+        */
+       dsl_pool_rele(dmu_objset_pool(os), FTAG);
        s = zpl_sget(fs_type, zpl_test_super, set_anon_super, flags, os);
-       dmu_objset_rele(os, FTAG);
+       dsl_dataset_rele(dmu_objset_ds(os), FTAG);
+
        if (IS_ERR(s))
                return (ERR_CAST(s));