zfs_inode_update should not call dmu_object_size_from_db under spinlock

We should never block when holding a spin lock, but zfs_inode_update can
block in the critical section of a spin lock in zfs_inode_update:

zfs_inode_update -> dmu_object_size_from_db -> zrl_add -> mutex_enter

Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #3858

diff --git a/module/zfs/zfs_znode.c b/module/zfs/zfs_znode.c
index d39743de943161d8366316c4909ca5a7eca2e8a3..8fd16408b021e4564929ecaf42f239f2c8058a93 100644 (file)
--- a/module/zfs/zfs_znode.c
+++ b/module/zfs/zfs_znode.c
@@ -485,6 +485,7 @@ zfs_inode_update(znode_t *zp)
        zfs_sb_t        *zsb;
        struct inode    *ip;
        uint32_t        blksize;
+       u_longlong_t    i_blocks;
        uint64_t        atime[2], mtime[2], ctime[2];
 
        ASSERT(zp != NULL);
@@ -499,6 +500,8 @@ zfs_inode_update(znode_t *zp)
        sa_lookup(zp->z_sa_hdl, SA_ZPL_MTIME(zsb), &mtime, 16);
        sa_lookup(zp->z_sa_hdl, SA_ZPL_CTIME(zsb), &ctime, 16);
 
+       dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize, &i_blocks);
+
        spin_lock(&ip->i_lock);
        ip->i_generation = zp->z_gen;
        ip->i_uid = SUID_TO_KUID(zp->z_uid);
@@ -507,8 +510,7 @@ zfs_inode_update(znode_t *zp)
        ip->i_mode = zp->z_mode;
        zfs_set_inode_flags(zp, ip);
        ip->i_blkbits = SPA_MINBLOCKSHIFT;
-       dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize,
-           (u_longlong_t *)&ip->i_blocks);
+       ip->i_blocks = i_blocks;
 
        ZFS_TIME_DECODE(&ip->i_atime, atime);
        ZFS_TIME_DECODE(&ip->i_mtime, mtime);