GFS2: glock livelock
This patch fixes a couple gfs2 problems with the reclaiming of
unlinked dinodes. First, there were a couple of livelocks where
everything would come to a halt waiting for a glock that was
seemingly held by a process that no longer existed. In fact, the
process did exist, it just had the wrong pid number in the holder
information. Second, there was a lock ordering problem between
inode locking and glock locking. Third, glock/inode contention
could sometimes cause inodes to be improperly marked invalid by
iget_failed.
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
diff --git a/fs/gfs2/inode.c b/fs/gfs2/inode.c
index b1bf269..51d8061 100644
--- a/fs/gfs2/inode.c
+++ b/fs/gfs2/inode.c
@@ -158,7 +158,6 @@
* @sb: The super block
* @no_addr: The inode number
* @type: The type of the inode
- * @skip_freeing: set this not return an inode if it is currently being freed.
*
* Returns: A VFS inode, or an error
*/
@@ -166,17 +165,14 @@
struct inode *gfs2_inode_lookup(struct super_block *sb,
unsigned int type,
u64 no_addr,
- u64 no_formal_ino, int skip_freeing)
+ u64 no_formal_ino)
{
struct inode *inode;
struct gfs2_inode *ip;
struct gfs2_glock *io_gl;
int error;
- if (skip_freeing)
- inode = gfs2_iget_skip(sb, no_addr);
- else
- inode = gfs2_iget(sb, no_addr);
+ inode = gfs2_iget(sb, no_addr);
ip = GFS2_I(inode);
if (!inode)
@@ -234,11 +230,98 @@
fail_iopen:
gfs2_glock_put(io_gl);
fail_put:
+ if (inode->i_state & I_NEW)
+ ip->i_gl->gl_object = NULL;
+ gfs2_glock_put(ip->i_gl);
+fail:
+ if (inode->i_state & I_NEW)
+ iget_failed(inode);
+ else
+ iput(inode);
+ return ERR_PTR(error);
+}
+
+/**
+ * gfs2_unlinked_inode_lookup - Lookup an unlinked inode for reclamation
+ * @sb: The super block
+ * no_addr: The inode number
+ * @@inode: A pointer to the inode found, if any
+ *
+ * Returns: 0 and *inode if no errors occurred. If an error occurs,
+ * the resulting *inode may or may not be NULL.
+ */
+
+int gfs2_unlinked_inode_lookup(struct super_block *sb, u64 no_addr,
+ struct inode **inode)
+{
+ struct gfs2_sbd *sdp;
+ struct gfs2_inode *ip;
+ struct gfs2_glock *io_gl;
+ int error;
+ struct gfs2_holder gh;
+
+ *inode = gfs2_iget_skip(sb, no_addr);
+
+ if (!(*inode))
+ return -ENOBUFS;
+
+ if (!((*inode)->i_state & I_NEW))
+ return -ENOBUFS;
+
+ ip = GFS2_I(*inode);
+ sdp = GFS2_SB(*inode);
+ ip->i_no_formal_ino = -1;
+
+ error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
+ if (unlikely(error))
+ goto fail;
+ ip->i_gl->gl_object = ip;
+
+ error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
+ if (unlikely(error))
+ goto fail_put;
+
+ set_bit(GIF_INVALID, &ip->i_flags);
+ error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, LM_FLAG_TRY | GL_EXACT,
+ &ip->i_iopen_gh);
+ if (unlikely(error)) {
+ if (error == GLR_TRYFAILED)
+ error = 0;
+ goto fail_iopen;
+ }
+ ip->i_iopen_gh.gh_gl->gl_object = ip;
+ gfs2_glock_put(io_gl);
+
+ (*inode)->i_mode = DT2IF(DT_UNKNOWN);
+
+ /*
+ * We must read the inode in order to work out its type in
+ * this case. Note that this doesn't happen often as we normally
+ * know the type beforehand. This code path only occurs during
+ * unlinked inode recovery (where it is safe to do this glock,
+ * which is not true in the general case).
+ */
+ error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, LM_FLAG_TRY,
+ &gh);
+ if (unlikely(error)) {
+ if (error == GLR_TRYFAILED)
+ error = 0;
+ goto fail_glock;
+ }
+ /* Inode is now uptodate */
+ gfs2_glock_dq_uninit(&gh);
+ gfs2_set_iop(*inode);
+
+ return 0;
+fail_glock:
+ gfs2_glock_dq(&ip->i_iopen_gh);
+fail_iopen:
+ gfs2_glock_put(io_gl);
+fail_put:
ip->i_gl->gl_object = NULL;
gfs2_glock_put(ip->i_gl);
fail:
- iget_failed(inode);
- return ERR_PTR(error);
+ return error;
}
static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
@@ -862,7 +945,7 @@
goto fail_gunlock2;
inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), inum.no_addr,
- inum.no_formal_ino, 0);
+ inum.no_formal_ino);
if (IS_ERR(inode))
goto fail_gunlock2;