]> git.karo-electronics.de Git - linux-beck.git/commitdiff
ceph: skip invalid dentry during dcache readdir
authorYan, Zheng <zheng.z.yan@intel.com>
Sun, 6 Apr 2014 06:10:04 +0000 (14:10 +0800)
committerSage Weil <sage@inktank.com>
Sun, 6 Apr 2014 16:13:14 +0000 (09:13 -0700)
skip dentries that were added before MDS issued FILE_SHARED to
client.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
fs/ceph/dir.c

index 46cd092cb0138e2e8bf0debaf7562ed12f592855..766410a12c2cb209a224fcfd97f63a055fd20801 100644 (file)
@@ -119,7 +119,8 @@ static int fpos_cmp(loff_t l, loff_t r)
  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
  * the MDS if/when the directory is modified).
  */
-static int __dcache_readdir(struct file *file, struct dir_context *ctx)
+static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
+                           u32 shared_gen)
 {
        struct ceph_file_info *fi = file->private_data;
        struct dentry *parent = file->f_dentry;
@@ -133,8 +134,8 @@ static int __dcache_readdir(struct file *file, struct dir_context *ctx)
        last = fi->dentry;
        fi->dentry = NULL;
 
-       dout("__dcache_readdir %p at %llu (last %p)\n", dir, ctx->pos,
-            last);
+       dout("__dcache_readdir %p v%u at %llu (last %p)\n",
+            dir, shared_gen, ctx->pos, last);
 
        spin_lock(&parent->d_lock);
 
@@ -161,7 +162,8 @@ more:
                        goto out_unlock;
                }
                spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
-               if (!d_unhashed(dentry) && dentry->d_inode &&
+               if (di->lease_shared_gen == shared_gen &&
+                   !d_unhashed(dentry) && dentry->d_inode &&
                    ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
                    ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
                    fpos_cmp(ctx->pos, di->offset) <= 0)
@@ -289,8 +291,9 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx)
            ceph_snap(inode) != CEPH_SNAPDIR &&
            __ceph_dir_is_complete(ci) &&
            __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
+               u32 shared_gen = ci->i_shared_gen;
                spin_unlock(&ci->i_ceph_lock);
-               err = __dcache_readdir(file, ctx);
+               err = __dcache_readdir(file, ctx, shared_gen);
                if (err != -EAGAIN)
                        return err;
        } else {