Merge tag 'gfs2-merge-window' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2

Pull GFS2 updates from Bob Peterson:
 "Here is a list of patches we've accumulated for GFS2 for the current
  upstream merge window.  Last window's set was short, but I warned that
  this one would be bigger, and so it is.  We've got 19 patches:

   - A patch from Abhi Das to propagate the GFS2_DIF_SYSTEM bit so that
     newly added journals don't get flagged, deleted, and recreated by
     fsck.gfs2.

   - Two patches from Andreas Gruenbacher to improve GFS2 performance
     where extended attributes are involved.

   - A patch from Andy Price to fix a suspicious rcu dereference error.

   - Two patches from Ben Marzinski that rework how GFS2's NFS cookies
     are managed.  This fixes readdir problems with nfs-over-gfs2.

   - A patch from Ben Marzinski that fixes a race in unmounting GFS2.

   - A set of four patches from me to move the resource group
     reservations inside the gfs2 inode to improve performance and fix a
     bug whereby get_write_access improperly prevented some operations
     like chown.

   - A patch from me to spinlock-protect the setting of system statfs
     file data.  This was causing small discrepancies between df and du.

   - A patch from me to reintroduce a timeout while clearing glocks
     which was accidentally dropped some time ago.

   - A patch from me to wait for iopen glock dequeues in order to
     improve deleting of files that were unlinked from a different
     cluster node.

   - A patch from me to ensure metadata address spaces get truncated
     when an inode is evicted.

   - A patch from me to fix a bug in which a memory leak could occur in
     some error cases when inodes were trying to be created.

   - A patch to consistently use iopen glocks to transition from the
     unlinked state to the deleted state.

   - A patch to fix a glock reference count error when inode creation
     fails.

   - A patch from Junxiao Bi to fix an flock panic.

   - A patch from Markus Elfring that removes an unnecessary if"

* tag 'gfs2-merge-window' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2:
  gfs2: fix flock panic issue
  GFS2: Don't do glock put on when inode creation fails
  GFS2: Always use iopen glock for gl_deletes
  GFS2: Release iopen glock in gfs2_create_inode error cases
  GFS2: Truncate address space mapping when deleting an inode
  GFS2: Wait for iopen glock dequeues
  gfs2: clear journal live bit in	gfs2_log_flush
  gfs2: change gfs2 readdir cookie
  gfs2: keep offset when splitting dir leaf blocks
  GFS2: Reintroduce a timeout in function gfs2_gl_hash_clear
  GFS2: Update master statfs buffer with sd_statfs_spin locked
  GFS2: Reduce size of incore inode
  GFS2: Make rgrp reservations part of the gfs2_inode structure
  GFS2: Extract quota data from reservations structure (revert 5407e24)
  gfs2: Extended attribute readahead optimization
  gfs2: Extended attribute readahead
  GFS2: Use rht_for_each_entry_rcu in glock_hash_walk
  GFS2: Delete an unnecessary check before the function call "iput"
  gfs2: Automatically set GFS2_DIF_SYSTEM flag on system files

1 files changed, 135 insertions, 40 deletions

diff --git a/fs/gfs2/dir.c b/fs/gfs2/dir.c
index ad8a5b757cc7..6a92592304fb 100644
--- a/fs/gfs2/dir.c
+++ b/fs/gfs2/dir.c
@@ -82,6 +82,8 @@
 
 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
+#define GFS2_HASH_INDEX_MASK 0xffffc000
+#define GFS2_USE_HASH_FLAG 0x2000
 
 struct qstr gfs2_qdot __read_mostly;
 struct qstr gfs2_qdotdot __read_mostly;
@@ -108,7 +110,7 @@ static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
 	struct buffer_head *bh;
 	int error;
 
-	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
+	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
 	if (error)
 		return error;
 	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
@@ -305,7 +307,7 @@ static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
 			BUG_ON(extlen < 1);
 			bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
 		} else {
-			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
+			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
 			if (error)
 				goto fail;
 		}
@@ -443,6 +445,27 @@ static int gfs2_dirent_last(const struct gfs2_dirent *dent,
 	return 0;
 }
 
+/* Look for the dirent that contains the offset specified in data. Once we
+ * find that dirent, there must be space available there for the new dirent */
+static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
+				  const struct qstr *name,
+				  void *ptr)
+{
+	unsigned required = GFS2_DIRENT_SIZE(name->len);
+	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
+	unsigned totlen = be16_to_cpu(dent->de_rec_len);
+
+	if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
+		return 0;
+	if (gfs2_dirent_sentinel(dent))
+		actual = 0;
+	if (ptr < (void *)dent + actual)
+		return -1;
+	if ((void *)dent + totlen >= ptr + required)
+		return 1;
+	return -1;
+}
+
 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
 				  const struct qstr *name,
 				  void *opaque)
@@ -682,6 +705,27 @@ static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
 	prev->de_rec_len = cpu_to_be16(prev_rec_len);
 }
 
+
+static struct gfs2_dirent *do_init_dirent(struct inode *inode,
+					  struct gfs2_dirent *dent,
+					  const struct qstr *name,
+					  struct buffer_head *bh,
+					  unsigned offset)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_dirent *ndent;
+	unsigned totlen;
+
+	totlen = be16_to_cpu(dent->de_rec_len);
+	BUG_ON(offset + name->len > totlen);
+	gfs2_trans_add_meta(ip->i_gl, bh);
+	ndent = (struct gfs2_dirent *)((char *)dent + offset);
+	dent->de_rec_len = cpu_to_be16(offset);
+	gfs2_qstr2dirent(name, totlen - offset, ndent);
+	return ndent;
+}
+
+
 /*
  * Takes a dent from which to grab space as an argument. Returns the
  * newly created dent.
@@ -691,31 +735,25 @@ static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
 					    const struct qstr *name,
 					    struct buffer_head *bh)
 {
-	struct gfs2_inode *ip = GFS2_I(inode);
-	struct gfs2_dirent *ndent;
-	unsigned offset = 0, totlen;
+	unsigned offset = 0;
 
 	if (!gfs2_dirent_sentinel(dent))
 		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
-	totlen = be16_to_cpu(dent->de_rec_len);
-	BUG_ON(offset + name->len > totlen);
-	gfs2_trans_add_meta(ip->i_gl, bh);
-	ndent = (struct gfs2_dirent *)((char *)dent + offset);
-	dent->de_rec_len = cpu_to_be16(offset);
-	gfs2_qstr2dirent(name, totlen - offset, ndent);
-	return ndent;
+	return do_init_dirent(inode, dent, name, bh, offset);
 }
 
-static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
-					     struct buffer_head *bh,
-					     const struct qstr *name)
+static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
+						   struct buffer_head *bh,
+						   const struct qstr *name,
+						   void *ptr)
 {
 	struct gfs2_dirent *dent;
 	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
-				gfs2_dirent_find_space, name, NULL);
+				gfs2_dirent_find_offset, name, ptr);
 	if (!dent || IS_ERR(dent))
 		return dent;
-	return gfs2_init_dirent(inode, dent, name, bh);
+	return do_init_dirent(inode, dent, name, bh,
+			      (unsigned)(ptr - (void *)dent));
 }
 
 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
@@ -723,7 +761,7 @@ static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
 {
 	int error;
 
-	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
+	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
 	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
 		/* pr_info("block num=%llu\n", leaf_no); */
 		error = -EIO;
@@ -1051,10 +1089,11 @@ static int dir_split_leaf(struct inode *inode, const struct qstr *name)
 		if (!gfs2_dirent_sentinel(dent) &&
 		    be32_to_cpu(dent->de_hash) < divider) {
 			struct qstr str;
+			void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
 			str.name = (char*)(dent+1);
 			str.len = be16_to_cpu(dent->de_name_len);
 			str.hash = be32_to_cpu(dent->de_hash);
-			new = gfs2_dirent_alloc(inode, nbh, &str);
+			new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
 			if (IS_ERR(new)) {
 				error = PTR_ERR(new);
 				break;
@@ -1186,10 +1225,10 @@ static int compare_dents(const void *a, const void *b)
 	int ret = 0;
 
 	dent_a = *(const struct gfs2_dirent **)a;
-	hash_a = be32_to_cpu(dent_a->de_hash);
+	hash_a = dent_a->de_cookie;
 
 	dent_b = *(const struct gfs2_dirent **)b;
-	hash_b = be32_to_cpu(dent_b->de_hash);
+	hash_b = dent_b->de_cookie;
 
 	if (hash_a > hash_b)
 		ret = 1;
@@ -1227,19 +1266,20 @@ static int compare_dents(const void *a, const void *b)
  */
 
 static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
-			   const struct gfs2_dirent **darr, u32 entries,
-			   int *copied)
+			   struct gfs2_dirent **darr, u32 entries,
+			   u32 sort_start, int *copied)
 {
 	const struct gfs2_dirent *dent, *dent_next;
 	u64 off, off_next;
 	unsigned int x, y;
 	int run = 0;
 
-	sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
+	if (sort_start < entries)
+		sort(&darr[sort_start], entries - sort_start,
+		     sizeof(struct gfs2_dirent *), compare_dents, NULL);
 
 	dent_next = darr[0];
-	off_next = be32_to_cpu(dent_next->de_hash);
-	off_next = gfs2_disk_hash2offset(off_next);
+	off_next = dent_next->de_cookie;
 
 	for (x = 0, y = 1; x < entries; x++, y++) {
 		dent = dent_next;
@@ -1247,8 +1287,7 @@ static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
 
 		if (y < entries) {
 			dent_next = darr[y];
-			off_next = be32_to_cpu(dent_next->de_hash);
-			off_next = gfs2_disk_hash2offset(off_next);
+			off_next = dent_next->de_cookie;
 
 			if (off < ctx->pos)
 				continue;
@@ -1295,6 +1334,40 @@ static void *gfs2_alloc_sort_buffer(unsigned size)
 	return ptr;
 }
 
+
+static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
+			    unsigned leaf_nr, struct gfs2_dirent **darr,
+			    unsigned entries)
+{
+	int sort_id = -1;
+	int i;
+	
+	for (i = 0; i < entries; i++) {
+		unsigned offset;
+
+		darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
+		darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
+
+		if (!sdp->sd_args.ar_loccookie)
+			continue;
+		offset = (char *)(darr[i]) -
+			 (bh->b_data + gfs2_dirent_offset(bh->b_data));
+		offset /= GFS2_MIN_DIRENT_SIZE;
+		offset += leaf_nr * sdp->sd_max_dents_per_leaf;
+		if (offset >= GFS2_USE_HASH_FLAG ||
+		    leaf_nr >= GFS2_USE_HASH_FLAG) {
+			darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
+			if (sort_id < 0)
+				sort_id = i;
+			continue;
+		}
+		darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
+		darr[i]->de_cookie |= offset;
+	}
+	return sort_id;
+}	
+
+
 static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
 			      int *copied, unsigned *depth,
 			      u64 leaf_no)
@@ -1304,12 +1377,11 @@ static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
 	struct buffer_head *bh;
 	struct gfs2_leaf *lf;
 	unsigned entries = 0, entries2 = 0;
-	unsigned leaves = 0;
-	const struct gfs2_dirent **darr, *dent;
+	unsigned leaves = 0, leaf = 0, offset, sort_offset;
+	struct gfs2_dirent **darr, *dent;
 	struct dirent_gather g;
 	struct buffer_head **larr;
-	int leaf = 0;
-	int error, i;
+	int error, i, need_sort = 0, sort_id;
 	u64 lfn = leaf_no;
 
 	do {
@@ -1325,6 +1397,11 @@ static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
 		brelse(bh);
 	} while(lfn);
 
+	if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
+		need_sort = 1;
+		sort_offset = 0;
+	}
+
 	if (!entries)
 		return 0;
 
@@ -1338,8 +1415,8 @@ static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
 	larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
 	if (!larr)
 		goto out;
-	darr = (const struct gfs2_dirent **)(larr + leaves);
-	g.pdent = darr;
+	darr = (struct gfs2_dirent **)(larr + leaves);
+	g.pdent = (const struct gfs2_dirent **)darr;
 	g.offset = 0;
 	lfn = leaf_no;
 
@@ -1350,6 +1427,7 @@ static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
 		lf = (struct gfs2_leaf *)bh->b_data;
 		lfn = be64_to_cpu(lf->lf_next);
 		if (lf->lf_entries) {
+			offset = g.offset;
 			entries2 += be16_to_cpu(lf->lf_entries);
 			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 						gfs2_dirent_gather, NULL, &g);
@@ -1367,17 +1445,26 @@ static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
 				goto out_free;
 			}
 			error = 0;
+			sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
+						   be16_to_cpu(lf->lf_entries));
+			if (!need_sort && sort_id >= 0) {
+				need_sort = 1;
+				sort_offset = offset + sort_id;
+			}
 			larr[leaf++] = bh;
 		} else {
+			larr[leaf++] = NULL;
 			brelse(bh);
 		}
 	} while(lfn);
 
 	BUG_ON(entries2 != entries);
-	error = do_filldir_main(ip, ctx, darr, entries, copied);
+	error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
+				sort_offset : entries, copied);
 out_free:
 	for(i = 0; i < leaf; i++)
-		brelse(larr[i]);
+		if (larr[i])
+			brelse(larr[i]);
 	kvfree(larr);
 out:
 	return error;
@@ -1483,7 +1570,7 @@ int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
 	struct gfs2_inode *dip = GFS2_I(inode);
 	struct gfs2_sbd *sdp = GFS2_SB(inode);
 	struct dirent_gather g;
-	const struct gfs2_dirent **darr, *dent;
+	struct gfs2_dirent **darr, *dent;
 	struct buffer_head *dibh;
 	int copied = 0;
 	int error;
@@ -1507,7 +1594,7 @@ int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
 	/* 96 is max number of dirents which can be stuffed into an inode */
 	darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
 	if (darr) {
-		g.pdent = darr;
+		g.pdent = (const struct gfs2_dirent **)darr;
 		g.offset = 0;
 		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
 					gfs2_dirent_gather, NULL, &g);
@@ -1524,8 +1611,9 @@ int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
 			error = -EIO;
 			goto out;
 		}
+		gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
 		error = do_filldir_main(dip, ctx, darr,
-					dip->i_entries, &copied);
+					dip->i_entries, 0, &copied);
 out:
 		kfree(darr);
 	}
@@ -1560,15 +1648,22 @@ struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
 
 	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
 	if (dent) {
+		struct inode *inode;
+		u16 rahead;
+
 		if (IS_ERR(dent))
 			return ERR_CAST(dent);
 		dtype = be16_to_cpu(dent->de_type);
+		rahead = be16_to_cpu(dent->de_rahead);
 		addr = be64_to_cpu(dent->de_inum.no_addr);
 		formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
 		brelse(bh);
 		if (fail_on_exist)
 			return ERR_PTR(-EEXIST);
-		return gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, 0);
+		inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, 0);
+		if (!IS_ERR(inode))
+			GFS2_I(inode)->i_rahead = rahead;
+		return inode;
 	}
 	return ERR_PTR(-ENOENT);
 }