Merge tag 'gfs2-4.14.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2

Pull GFS2 updates from Bob Peterson:
 "We've got a whopping 29 GFS2 patches for this merge window, mainly
  because we held some back from the previous merge window until we
  could get them perfected and well tested. We have a couple patch sets,
  including my patch set for protecting glock gl_object and Andreas
  Gruenbacher's patch set to fix the long-standing shrink- slab hang,
  plus a bunch of assorted bugs and cleanups.

  Summary:

   - I fixed a bug whereby an IO error would lead to a double-brelse.

   - Andreas Gruenbacher made a minor cleanup to call his relatively new
     function, gfs2_holder_initialized, rather than doing it manually.
     This was just missed by a previous patch set.

   - Jan Kara fixed a bug whereby the SGID was being cleared when
     inheriting ACLs.

   - Andreas found a bug and fixed it in his previous patch, "Get rid of
     flush_delayed_work in gfs2_evict_inode". A call to
     flush_delayed_work was deleted from *gfs2_inode_lookup and added to
     gfs2_create_inode.

   - Wang Xibo found and fixed a list_add call in inode_go_lock that
     specified the parameters in the wrong order.

   - Coly Li submitted a patch to add the REQ_PRIO to some of GFS2's
     metadata reads that were accidentally missing them.

   - I submitted a 4-patch set to protect the glock gl_object field.
     GFS2 was setting and checking gl_object with no locking mechanism,
     so the value was occasionally stomped on, which caused file system
     corruption.

   - I submitted a small cleanup to function gfs2_clear_rgrpd. It was
     needlessly adding rgrp glocks to the lru list, then pulling them
     back off immediately. The rgrp glocks don't use the lru list
     anyway, so doing so was just a waste of time.

   - I submitted a patch that checks the GLOF_LRU flag on a glock before
     trying to remove it from the lru_list. This avoids a lot of
     unnecessary spin_lock contention.

   - I submitted a patch to delete GFS2's debugfs files only after we
     evict all the glocks. Before this patch, GFS2 would delete the
     debugfs files, and if unmount hung waiting for a glock, there was
     no way to debug the problem. Now, if a hang occurs during umount,
     we can examine the debugfs files to figure out why it's hung.

   - Andreas Gruenbacher submitted a patch to fix some trivial typos.

   - Andreas also submitted a five-part patch set to fix the
     longstanding hang involving the slab shrinker: dlm requires memory,
     calls the inode shrinker, which calls gfs2's evict, which calls
     back into DLM before it can evict an inode.

   - Abhi Das submitted a patch to forcibly flush the active items list
     to relieve memory pressure. This fixes a long-standing bug whereby
     GFS2 was getting hung permanently in balance_dirty_pages.

   - Thomas Tai submitted a patch to fix a slab corruption problem due
     to a residual pointer left in the lock_dlm lockstruct.

   - I submitted a patch to withdraw the file system if IO errors are
     encountered while writing to the journals or statfs system file
     which were previously not being sent back up. Before, some IO
     errors were sometimes not be detected for several hours, and at
     recovery time, the journal errors made journal replay impossible.

   - Andreas has a patch to fix an annoying format-truncation compiler
     warning so GFS2 compiles cleanly.

   - I have a patch that fixes a handful of sparse compiler warnings.

   - Andreas fixed up an useless gl_object warning caused by an earlier
     patch.

   - Arvind Yadav added a patch to properly constify our rhashtable
     params declare.

   - I added a patch to fix a regression caused by the non-recursive
     delete and truncate patch that caused file system blocks to not be
     properly freed.

   - Ernesto A. Fernández added a patch to fix a place where GFS2 would
     send back the wrong return code setting extended attributes.

   - Ernesto also added a patch to fix a case in which GFS2 was
     improperly setting an inode's i_mode, potentially granting access
     to the wrong users"

* tag 'gfs2-4.14.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2: (29 commits)
  gfs2: preserve i_mode if __gfs2_set_acl() fails
  gfs2: don't return ENODATA in __gfs2_xattr_set unless replacing
  GFS2: Fix non-recursive truncate bug
  gfs2: constify rhashtable_params
  GFS2: Fix gl_object warnings
  GFS2: Fix up some sparse warnings
  gfs2: Silence gcc format-truncation warning
  GFS2: Withdraw for IO errors writing to the journal or statfs
  gfs2: fix slab corruption during mounting and umounting gfs file system
  gfs2: forcibly flush ail to relieve memory pressure
  gfs2: Clean up waiting on glocks
  gfs2: Defer deleting inodes under memory pressure
  gfs2: gfs2_evict_inode: Put glocks asynchronously
  gfs2: Get rid of gfs2_set_nlink
  gfs2: gfs2_glock_get: Wait on freeing glocks
  gfs2: Fix trivial typos
  GFS2: Delete debugfs files only after we evict the glocks
  GFS2: Don't waste time locking lru_lock for non-lru glocks
  GFS2: Don't bother trying to add rgrps to the lru list
  GFS2: Clear gl_object when deleting an inode in gfs2_delete_inode
  ...

Author: torvalds

fs/gfs2/acl.c

@@ -86,19 +86,6 @@ int __gfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 	char *data;
 	const char *name = gfs2_acl_name(type);
 
-	if (acl && acl->a_count > GFS2_ACL_MAX_ENTRIES(GFS2_SB(inode)))
-		return -E2BIG;
-
-	if (type == ACL_TYPE_ACCESS) {
-		umode_t mode = inode->i_mode;
-
-		error = posix_acl_update_mode(inode, &inode->i_mode, &acl);
-		if (error)
-			return error;
-		if (mode != inode->i_mode)
-			mark_inode_dirty(inode);
-	}
-
 	if (acl) {
 		len = posix_acl_to_xattr(&init_user_ns, acl, NULL, 0);
 		if (len == 0)
@@ -129,6 +116,10 @@ int gfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 	struct gfs2_holder gh;
 	bool need_unlock = false;
 	int ret;
+	umode_t mode;
+
+	if (acl && acl->a_count > GFS2_ACL_MAX_ENTRIES(GFS2_SB(inode)))
+		return -E2BIG;
 
 	ret = gfs2_rsqa_alloc(ip);
 	if (ret)
@@ -140,7 +131,20 @@ int gfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 			return ret;
 		need_unlock = true;
 	}
+
+	mode = inode->i_mode;
+	if (type == ACL_TYPE_ACCESS && acl) {
+		ret = posix_acl_update_mode(inode, &mode, &acl);
+		if (ret)
+			goto unlock;
+	}
+
 	ret = __gfs2_set_acl(inode, acl, type);
+	if (!ret && mode != inode->i_mode) {
+		inode->i_mode = mode;
+		mark_inode_dirty(inode);
+	}
+unlock:
 	if (need_unlock)
 		gfs2_glock_dq_uninit(&gh);
 	return ret;

fs/gfs2/aops.c

@@ -234,7 +234,19 @@ static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc
 static int gfs2_writepages(struct address_space *mapping,
 			   struct writeback_control *wbc)
 {
-	return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
+	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
+	int ret = mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
+
+	/*
+	 * Even if we didn't write any pages here, we might still be holding
+	 * dirty pages in the ail. We forcibly flush the ail because we don't
+	 * want balance_dirty_pages() to loop indefinitely trying to write out
+	 * pages held in the ail that it can't find.
+	 */
+	if (ret == 0)
+		set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
+
+	return ret;
 }
 
 /**

fs/gfs2/bmap.c

@@ -291,8 +291,9 @@ static void gfs2_metapath_ra(struct gfs2_glock *gl,
 		if (trylock_buffer(rabh)) {
 			if (!buffer_uptodate(rabh)) {
 				rabh->b_end_io = end_buffer_read_sync;
-				submit_bh(REQ_OP_READ, REQ_RAHEAD | REQ_META,
-						rabh);
+				submit_bh(REQ_OP_READ,
+					  REQ_RAHEAD | REQ_META | REQ_PRIO,
+					  rabh);
 				continue;
 			}
 			unlock_buffer(rabh);
@@ -1103,8 +1104,15 @@ static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp,
 
 	while (true) {
 		ptr = metapointer(h, mp);
-		if (*ptr) /* if we have a non-null pointer */
+		if (*ptr) { /* if we have a non-null pointer */
+			/* Now zero the metapath after the current height. */
+			h++;
+			if (h < GFS2_MAX_META_HEIGHT)
+				memset(&mp->mp_list[h], 0,
+				       (GFS2_MAX_META_HEIGHT - h) *
+				       sizeof(mp->mp_list[0]));
 			return true;
+		}
 
 		if (mp->mp_list[h] < ptrs)
 			mp->mp_list[h]++;
@@ -1120,6 +1128,13 @@ enum dealloc_states {
 	DEALLOC_DONE = 3,       /* process complete */
 };
 
+static bool mp_eq_to_hgt(struct metapath *mp, __u16 *nbof, unsigned int h)
+{
+	if (memcmp(mp->mp_list, nbof, h * sizeof(mp->mp_list[0])))
+		return false;
+	return true;
+}
+
 /**
  * trunc_dealloc - truncate a file down to a desired size
  * @ip: inode to truncate
@@ -1197,8 +1212,7 @@ static int trunc_dealloc(struct gfs2_inode *ip, u64 newsize)
 			/* If we're truncating to a non-zero size and the mp is
 			   at the beginning of file for the strip height, we
 			   need to preserve the first metadata pointer. */
-			preserve1 = (newsize &&
-				     (mp.mp_list[mp_h] == nbof[mp_h]));
+			preserve1 = (newsize && mp_eq_to_hgt(&mp, nbof, mp_h));
 			bh = mp.mp_bh[mp_h];
 			gfs2_assert_withdraw(sdp, bh);
 			if (gfs2_assert_withdraw(sdp,

fs/gfs2/dir.c

@@ -1513,7 +1513,9 @@ static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
 				continue;
 			}
 			bh->b_end_io = end_buffer_read_sync;
-			submit_bh(REQ_OP_READ, REQ_RAHEAD | REQ_META, bh);
+			submit_bh(REQ_OP_READ,
+				  REQ_RAHEAD | REQ_META | REQ_PRIO,
+				  bh);
 			continue;
 		}
 		brelse(bh);

fs/gfs2/file.c

@@ -1030,8 +1030,7 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
 
 	mutex_lock(&fp->f_fl_mutex);
 
-	gl = fl_gh->gh_gl;
-	if (gl) {
+	if (gfs2_holder_initialized(fl_gh)) {
 		if (fl_gh->gh_state == state)
 			goto out;
 		locks_lock_file_wait(file,