[Cluster-devel] [GFS2 Patch][TRY 3] GFS2: Extend the life of the reservations structure

Bob Peterson rpeterso at redhat.com
Mon May 14 16:15:01 UTC 2012


----- Original Message -----
| Hi,
| 
| On Mon, 2012-05-14 at 08:58 -0400, Bob Peterson wrote:
| > | >  fail:
| > | > +	gfs2_rs_delete(dip);
| > | 
| > | Should we really be dropping the ip->i_res here I wonder? I'm not
| > | sure
| > | that we want to forget this information just because (for
| > | example)
| > | someone has tried to create a new inode and it has been failed
| > | for
| > | some
| > | reason or other.
| > | 
| > | Otherwise I think this looks good,
| > | 
| > | Steve.
| > 
| > Hi,
| > 
| > In this implementation, function gfs2_inplace_release just unlocks
| > the
| > rgrp glock and sets rs_requested to zero. In fact, the reservation
| > still
| > hangs around, attached to the inode until function gfs2_rs_delete
| > is called
| > from gfs2_release or gfs2_evict_inode. So instead of having two
| > functions
| > for allocation and deallocation, we now have four:
| > 
| That is all ok I think, but I'm referring to the call to
| gfs2_rs_delete() in this case. We should probably call the combined
| structure something like struct gfs2_alloc, since it deals with
| allocation and not just reservations. However, the issue here was
| dropping the reservation from the directory in the case that the
| allocation has failed for some reason (maybe permisions, or something
| like that)
| 
| Steve.

Hi,

Ah, I misunderstood. I see your point now. Here's the revised version.

Regards,

Bob Peterson
Red Hat File Systems

Signed-off-by: Bob Peterson <rpeterso at redhat.com> 
---
Author: Bob Peterson <rpeterso at redhat.com>
Date:   Thu Apr 12 11:37:24 2012 -0500

GFS2: Extend the life of the reservations structure

This patch lengthens the lifespan of the reservations structure for
inodes. Before, they were allocated and deallocated for every write
operation. With this patch, they are allocated when the first write
occurs, and deallocated when the last process closes the file.
It's more efficient to do it this way because it saves GFS2 a lot of
unnecessary allocates and frees. It also gives us more flexibility
for the future: (1) we can now fold the qadata structure back into
the structure and save those alloc/frees, (2) we can use this for
multi-block reservations.

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index e80a464..aba77b5 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -878,7 +878,7 @@ static int gfs2_write_end(struct file *file, struct address_space *mapping,
 	brelse(dibh);
 failed:
 	gfs2_trans_end(sdp);
-	if (ip->i_res)
+	if (gfs2_mb_reserved(ip))
 		gfs2_inplace_release(ip);
 	if (qa) {
 		gfs2_quota_unlock(ip);
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 31b199f..3790617 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -376,6 +376,10 @@ static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 	 */
 	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 
+	ret = gfs2_rs_alloc(ip);
+	if (ret)
+		return ret;
+
 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 	ret = gfs2_glock_nq(&gh);
 	if (ret)
@@ -569,10 +573,15 @@ static int gfs2_release(struct inode *inode, struct file *file)
 {
 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 	struct gfs2_file *fp;
+	struct gfs2_inode *ip = GFS2_I(inode);
 
 	fp = file->private_data;
 	file->private_data = NULL;
 
+	if ((file->f_mode & FMODE_WRITE) && ip->i_res &&
+	    (atomic_read(&inode->i_writecount) == 1))
+		gfs2_rs_delete(ip);
+
 	if (gfs2_assert_warn(sdp, fp))
 		return -EIO;
 
@@ -653,12 +662,16 @@ static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
 				   unsigned long nr_segs, loff_t pos)
 {
 	struct file *file = iocb->ki_filp;
+	struct dentry *dentry = file->f_dentry;
+	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
+	int ret;
+
+	ret = gfs2_rs_alloc(ip);
+	if (ret)
+		return ret;
 
 	if (file->f_flags & O_APPEND) {
-		struct dentry *dentry = file->f_dentry;
-		struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
 		struct gfs2_holder gh;
-		int ret;
 
 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 		if (ret)
@@ -774,6 +787,10 @@ static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
 	if (bytes == 0)
 		bytes = sdp->sd_sb.sb_bsize;
 
+	error = gfs2_rs_alloc(ip);
+	if (error)
+		return error;
+
 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
 	error = gfs2_glock_nq(&ip->i_gh);
 	if (unlikely(error))
diff --git a/fs/gfs2/inode.c b/fs/gfs2/inode.c
index a9ba244..ec7d93b 100644
--- a/fs/gfs2/inode.c
+++ b/fs/gfs2/inode.c
@@ -667,6 +667,10 @@ static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
 	if (!name->len || name->len > GFS2_FNAMESIZE)
 		return -ENAMETOOLONG;
 
+	error = gfs2_rs_alloc(dip);
+	if (error)
+		return error;
+
 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
 	if (error)
 		goto fail;
@@ -704,6 +708,11 @@ static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
 	if (error)
 		goto fail_gunlock2;
 
+	/* the new inode needs a reservation so it can allocate xattrs. */
+	error = gfs2_rs_alloc(GFS2_I(inode));
+	if (error)
+		goto fail_gunlock2;
+
 	error = gfs2_acl_create(dip, inode);
 	if (error)
 		goto fail_gunlock2;
@@ -722,7 +731,7 @@ static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
 	gfs2_trans_end(sdp);
 	/* Check if we reserved space in the rgrp. Function link_dinode may
 	   not, depending on whether alloc is required. */
-	if (dip->i_res)
+	if (gfs2_mb_reserved(dip))
 		gfs2_inplace_release(dip);
 	gfs2_quota_unlock(dip);
 	gfs2_qadata_put(dip);
diff --git a/fs/gfs2/rgrp.c b/fs/gfs2/rgrp.c
index f74fb9b..e944fef 100644
--- a/fs/gfs2/rgrp.c
+++ b/fs/gfs2/rgrp.c
@@ -417,6 +417,39 @@ void gfs2_free_clones(struct gfs2_rgrpd *rgd)
 	}
 }
 
+/**
+ * gfs2_rs_alloc - make sure we have a reservation assigned to the inode
+ * @ip: the inode for this reservation
+ */
+int gfs2_rs_alloc(struct gfs2_inode *ip)
+{
+	int error = 0;
+
+	down_write(&ip->i_rw_mutex);
+	if (!ip->i_res) {
+		ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS);
+		if (!ip->i_res)
+			error = -ENOMEM;
+	}
+	up_write(&ip->i_rw_mutex);
+	return error;
+}
+
+/**
+ * gfs2_rs_delete - delete a reservation
+ * @ip: The inode for this reservation
+ *
+ */
+void gfs2_rs_delete(struct gfs2_inode *ip)
+{
+	down_write(&ip->i_rw_mutex);
+	if (ip->i_res) {
+		kmem_cache_free(gfs2_rsrv_cachep, ip->i_res);
+		ip->i_res = NULL;
+	}
+	up_write(&ip->i_rw_mutex);
+}
+
 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp)
 {
 	struct rb_node *n;
@@ -993,22 +1026,6 @@ struct gfs2_qadata *gfs2_qadata_get(struct gfs2_inode *ip)
 }
 
 /**
- * gfs2_blkrsv_get - get the struct gfs2_blkreserv structure for an inode
- * @ip: the incore GFS2 inode structure
- *
- * Returns: the struct gfs2_qadata
- */
-
-static int gfs2_blkrsv_get(struct gfs2_inode *ip)
-{
-	BUG_ON(ip->i_res != NULL);
-	ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS);
-	if (!ip->i_res)
-		return -ENOMEM;
-	return 0;
-}
-
-/**
  * try_rgrp_fit - See if a given reservation will fit in a given RG
  * @rgd: the RG data
  * @ip: the inode
@@ -1162,13 +1179,6 @@ static int get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)
 	return -ENOSPC;
 }
 
-static void gfs2_blkrsv_put(struct gfs2_inode *ip)
-{
-	BUG_ON(ip->i_res == NULL);
-	kmem_cache_free(gfs2_rsrv_cachep, ip->i_res);
-	ip->i_res = NULL;
-}
-
 /**
  * gfs2_inplace_reserve - Reserve space in the filesystem
  * @ip: the inode to reserve space for
@@ -1181,14 +1191,10 @@ int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested)
 {
 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 	struct gfs2_blkreserv *rs;
-	int error;
+	int error = 0;
 	u64 last_unlinked = NO_BLOCK;
 	int tries = 0;
 
-	error = gfs2_blkrsv_get(ip);
-	if (error)
-		return error;
-
 	rs = ip->i_res;
 	rs->rs_requested = requested;
 	if (gfs2_assert_warn(sdp, requested)) {
@@ -1213,7 +1219,7 @@ int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested)
 
 out:
 	if (error)
-		gfs2_blkrsv_put(ip);
+		rs->rs_requested = 0;
 	return error;
 }
 
@@ -1230,7 +1236,7 @@ void gfs2_inplace_release(struct gfs2_inode *ip)
 
 	if (rs->rs_rgd_gh.gh_gl)
 		gfs2_glock_dq_uninit(&rs->rs_rgd_gh);
-	gfs2_blkrsv_put(ip);
+	rs->rs_requested = 0;
 }
 
 /**
@@ -1496,7 +1502,7 @@ int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks,
 	/* Only happens if there is a bug in gfs2, return something distinctive
 	 * to ensure that it is noticed.
 	 */
-	if (ip->i_res == NULL)
+	if (ip->i_res->rs_requested == 0)
 		return -ECANCELED;
 
 	rgd = ip->i_rgd;
diff --git a/fs/gfs2/rgrp.h b/fs/gfs2/rgrp.h
index b4b10f4..d9eda5f 100644
--- a/fs/gfs2/rgrp.h
+++ b/fs/gfs2/rgrp.h
@@ -43,6 +43,8 @@ extern void gfs2_inplace_release(struct gfs2_inode *ip);
 extern int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *n,
 			     bool dinode, u64 *generation);
 
+extern int gfs2_rs_alloc(struct gfs2_inode *ip);
+extern void gfs2_rs_delete(struct gfs2_inode *ip);
 extern void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta);
 extern void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen);
 extern void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip);
@@ -68,4 +70,12 @@ extern int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset,
 				   const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed);
 extern int gfs2_fitrim(struct file *filp, void __user *argp);
 
+/* This is how to tell if a reservation is "inplace" reserved: */
+static inline int gfs2_mb_reserved(struct gfs2_inode *ip)
+{
+	if (ip->i_res && ip->i_res->rs_requested)
+		return 1;
+	return 0;
+}
+
 #endif /* __RGRP_DOT_H__ */
diff --git a/fs/gfs2/super.c b/fs/gfs2/super.c
index 6172fa7..a42df66 100644
--- a/fs/gfs2/super.c
+++ b/fs/gfs2/super.c
@@ -1554,6 +1554,7 @@ out_unlock:
 out:
 	/* Case 3 starts here */
 	truncate_inode_pages(&inode->i_data, 0);
+	gfs2_rs_delete(ip);
 	end_writeback(inode);
 	gfs2_dir_hash_inval(ip);
 	ip->i_gl->gl_object = NULL;
@@ -1576,6 +1577,7 @@ static struct inode *gfs2_alloc_inode(struct super_block *sb)
 		ip->i_flags = 0;
 		ip->i_gl = NULL;
 		ip->i_rgd = NULL;
+		ip->i_res = NULL;
 	}
 	return &ip->i_inode;
 }
diff --git a/fs/gfs2/trans.h b/fs/gfs2/trans.h
index 125d457..41f42cd 100644
--- a/fs/gfs2/trans.h
+++ b/fs/gfs2/trans.h
@@ -31,7 +31,7 @@ struct gfs2_glock;
 static inline unsigned int gfs2_rg_blocks(const struct gfs2_inode *ip)
 {
 	const struct gfs2_blkreserv *rs = ip->i_res;
-	if (rs->rs_requested < ip->i_rgd->rd_length)
+	if (rs && rs->rs_requested < ip->i_rgd->rd_length)
 		return rs->rs_requested + 1;
 	return ip->i_rgd->rd_length;
 }




More information about the Cluster-devel mailing list