2.6.5-quotafix.patch

非常经典的一个分布式系统

 	spin_unlock(&dq_list_lock);
 
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
-		if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt) && info_dirty(&dqopt->info[cnt])) {
-			down(&dqopt->dqio_sem);
-			dqopt->ops[cnt]->write_file_info(sb, cnt);
-			up(&dqopt->dqio_sem);
-		}
+		if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt) 
+				&& info_dirty(&dqopt->info[cnt]))
+			sb->dq_op->write_info(sb, cnt);
 	spin_lock(&dq_list_lock);
 	dqstats.syncs++;
 	spin_unlock(&dq_list_lock);
@@ -432,11 +513,19 @@ we_slept:
 		spin_unlock(&dq_list_lock);
 		return;
 	}
-	if (dquot_dirty(dquot)) {
+	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 		spin_unlock(&dq_list_lock);
 		dquot->dq_sb->dq_op->write_dquot(dquot);
 		goto we_slept;
 	}
+	/* Clear flag in case dquot was inactive (something bad happened) */
+	test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
+	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
+		spin_unlock(&dq_list_lock);
+		dquot->dq_sb->dq_op->release_dquot(dquot);
+		goto we_slept;
+	}
+			
 	atomic_dec(&dquot->dq_count);
 #ifdef __DQUOT_PARANOIA
 	/* sanity check */
@@ -495,7 +584,6 @@ we_slept:
 		insert_dquot_hash(dquot);
 		dqstats.lookups++;
 		spin_unlock(&dq_list_lock);
-		read_dqblk(dquot);
 	} else {
 		if (!atomic_read(&dquot->dq_count))
 			remove_free_dquot(dquot);
@@ -503,10 +591,15 @@ we_slept:
 		dqstats.cache_hits++;
 		dqstats.lookups++;
 		spin_unlock(&dq_list_lock);
-		wait_on_dquot(dquot);
 		if (empty)
 			kmem_cache_free(dquot_cachep, empty);
 	}
+	wait_on_dquot(dquot);
+	/* Read the dquot and instantiate it (everything done only if needed) */
+	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
+		dqput(dquot);
+		return NODQUOT;
+	}
 
 #ifdef __DQUOT_PARANOIA
 	if (!dquot->dq_sb)	/* Has somebody invalidated entry under us? */
@@ -819,19 +912,19 @@ static int check_bdq(struct dquot *dquot
  *
  * Note: this is a blocking operation.
  */
-void dquot_initialize(struct inode *inode, int type)
+int dquot_initialize(struct inode *inode, int type)
 {
 	unsigned int id = 0;
 	int cnt;
 
 	/* Solve deadlock when we recurse when holding dqptr_sem... */
 	if (IS_NOQUOTA(inode))
-		return;
+		return 0;
 	down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 	/* Having dqptr_sem we know NOQUOTA flags can't be altered... */
 	if (IS_NOQUOTA(inode)) {
 		up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
-		return;
+		return 0;
 	}
 	/* Build list of quotas to initialize... */
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
@@ -852,13 +945,14 @@ void dquot_initialize(struct inode *inod
 		}
 	}
 	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	return 0;
 }
 
 /*
  * 	Release all quotas referenced by inode
  *	Needs dqonoff_sem to guard dqput()
  */
-void dquot_drop(struct inode *inode)
+int dquot_drop(struct inode *inode)
 {
 	int cnt;
 
@@ -871,6 +965,7 @@ void dquot_drop(struct inode *inode)
 		}
 	}
 	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	return 0;
 }
 
 /*
@@ -958,14 +1053,14 @@ warn_put_all:
 /*
  * This is a non-blocking operation.
  */
-void dquot_free_space(struct inode *inode, qsize_t number)
+int dquot_free_space(struct inode *inode, qsize_t number)
 {
 	unsigned int cnt;
 
 	/* Solve deadlock when we recurse when holding dqptr_sem... */
 	if (IS_NOQUOTA(inode)) {
 		inode_add_bytes(inode, number);
-		return;
+		return QUOTA_OK;
 	}
 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
 	spin_lock(&dq_data_lock);
@@ -981,23 +1076,24 @@ sub_bytes:
 	inode_sub_bytes(inode, number);
 	spin_unlock(&dq_data_lock);
 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	return QUOTA_OK;
 }
 
 /*
  * This is a non-blocking operation.
  */
-void dquot_free_inode(const struct inode *inode, unsigned long number)
+int dquot_free_inode(const struct inode *inode, unsigned long number)
 {
 	unsigned int cnt;
 
 	/* Solve deadlock when we recurse when holding dqptr_sem... */
 	if (IS_NOQUOTA(inode))
-		return;
+		return QUOTA_OK;
 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
 	/* Now recheck reliably when holding dqptr_sem */
 	if (IS_NOQUOTA(inode)) {
 		up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
-		return;
+		return QUOTA_OK;
 	}
 	spin_lock(&dq_data_lock);
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
@@ -1007,6 +1103,7 @@ void dquot_free_inode(const struct inode
 	}
 	spin_unlock(&dq_data_lock);
 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	return QUOTA_OK;
 }
 
 /*
@@ -1104,6 +1201,20 @@ warn_put_all:
 }
 
 /*
+ * Write info of quota file to disk
+ */
+int dquot_commit_info(struct super_block *sb, int type)
+{
+	int ret;
+	struct quota_info *dqopt = sb_dqopt(sb);
+
+	down(&dqopt->dqio_sem);
+	ret = dqopt->ops[type]->write_file_info(sb, type);
+	up(&dqopt->dqio_sem);
+	return ret;
+}
+
+/*
  * Definitions of diskquota operations.
  */
 struct dquot_operations dquot_operations = {
@@ -1114,7 +1225,10 @@ struct dquot_operations dquot_operations
 	.free_space	= dquot_free_space,
 	.free_inode	= dquot_free_inode,
 	.transfer	= dquot_transfer,
-	.write_dquot	= commit_dqblk
+	.write_dquot	= dquot_commit,
+	.acquire_dquot  = dquot_acquire,
+	.release_dquot	= dquot_release,
+	.write_info     = dquot_commit_info				
 };
 
 /* Function used by filesystems for initializing the dquot_operations structure */
@@ -1154,13 +1268,14 @@ int vfs_quota_off(struct super_block *sb
 {
 	int cnt;
 	struct quota_info *dqopt = sb_dqopt(sb);
-
-	if (!sb)
-		goto out;
+	struct inode *toputinode[MAXQUOTAS];
+	struct vfsmount *toputmnt[MAXQUOTAS];
 
 	/* We need to serialize quota_off() for device */
 	down(&dqopt->dqonoff_sem);
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
+		toputinode[cnt] = NULL;
+		toputmnt[cnt] = NULL;
 		if (type != -1 && cnt != type)
 			continue;
 		if (!sb_has_quota_enabled(sb, cnt))
@@ -1172,94 +1287,115 @@ int vfs_quota_off(struct super_block *sb
 		invalidate_dquots(sb, cnt);
 		/*
 		 * Now all dquots should be invalidated, all writes done so we should be only
-		 * users of the info. No locks needed.
+		 * users of the info.
 		 */
-		if (info_dirty(&dqopt->info[cnt])) {
-			down(&dqopt->dqio_sem);
-			dqopt->ops[cnt]->write_file_info(sb, cnt);
-			up(&dqopt->dqio_sem);
-		}
+		if (info_dirty(&dqopt->info[cnt]))
+			sb->dq_op->write_info(sb, cnt);
 		if (dqopt->ops[cnt]->free_file_info)
 			dqopt->ops[cnt]->free_file_info(sb, cnt);
 		put_quota_format(dqopt->info[cnt].dqi_format);
 
-		fput(dqopt->files[cnt]);
-		dqopt->files[cnt] = (struct file *)NULL;
+		toputinode[cnt] = dqopt->files[cnt];
+		toputmnt[cnt] = dqopt->mnt[cnt];
+		dqopt->files[cnt] = NULL;
+		dqopt->mnt[cnt] = NULL;
 		dqopt->info[cnt].dqi_flags = 0;
 		dqopt->info[cnt].dqi_igrace = 0;
 		dqopt->info[cnt].dqi_bgrace = 0;
 		dqopt->ops[cnt] = NULL;
 	}
 	up(&dqopt->dqonoff_sem);
-out:
+	/* Sync the superblock so that buffers with quota data are written to
+	 * disk (and so userspace sees correct data afterwards).
+	 * The reference to vfsmnt we are still holding protects us from
+	 * umount (we don't have it only when quotas are turned on/off for
+	 * journal replay but in that case we are guarded by the fs anyway). */
+	if (sb->s_op->sync_fs)
+		sb->s_op->sync_fs(sb, 1);
+	sync_blockdev(sb->s_bdev);
+	/* Now the quota files are just ordinary files and we can set the
+	 * inode flags back. Moreover we discard the pagecache so that
+	 * userspace sees the writes we did bypassing the pagecache. We
+	 * must also discard the blockdev buffers so that we see the
+	 * changes done by userspace on the next quotaon() */
+	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
+		if (toputinode[cnt]) {
+			down(&dqopt->dqonoff_sem);
+			/* If quota was reenabled in the meantime, we have
+			 * nothing to do */
+			if (!sb_has_quota_enabled(sb, cnt)) {
+				down(&toputinode[cnt]->i_sem);
+				toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
+				  S_NOATIME | S_NOQUOTA);
+				truncate_inode_pages(&toputinode[cnt]->i_data, 0);
+				up(&toputinode[cnt]->i_sem);
+				mark_inode_dirty(toputinode[cnt]);
+				iput(toputinode[cnt]);
+			}
+			up(&dqopt->dqonoff_sem);
+			/* We don't hold the reference when we turned on quotas
+			 * just for the journal replay... */
+			if (toputmnt[cnt])
+				mntput(toputmnt[cnt]);
+		}
+	invalidate_bdev(sb->s_bdev, 0);
 	return 0;
 }
 
-int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
+/*
+ *	Turn quotas on on a device
+ */
+
+/* Helper function when we already have the inode */
+static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
 {
-	struct file *f;
-	struct inode *inode;
-	struct quota_info *dqopt = sb_dqopt(sb);
 	struct quota_format_type *fmt = find_quota_format(format_id);
-	int error, cnt;
-	struct dquot *to_drop[MAXQUOTAS];
-	unsigned int oldflags;
+	struct super_block *sb = inode->i_sb;
+	struct quota_info *dqopt = sb_dqopt(sb);
+	int error;
+	int oldflags = -1;
 
 	if (!fmt)
 		return -ESRCH;
-	f = filp_open(path, O_RDWR, 0600);
-	if (IS_ERR(f)) {
-		error = PTR_ERR(f);
+	if (!S_ISREG(inode->i_mode)) {
+		error = -EACCES;
+		goto out_fmt;
+	}
+	if (IS_RDONLY(inode)) {
+		error = -EROFS;
+		goto out_fmt;
+	}
+	if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
+		error = -EINVAL;
 		goto out_fmt;
 	}
-	error = -EIO;
-	if (!f->f_op || !f->f_op->read || !f->f_op->write)
-		goto out_f;
-	error = security_quota_on(f);
-	if (error)
-		goto out_f;
-	inode = f->f_dentry->d_inode;
-	error = -EACCES;
-	if (!S_ISREG(inode->i_mode))
-		goto out_f;
 
+	/* As we bypass the pagecache we must now flush the inode so that
+	 * we see all the changes from userspace... */
+	write_inode_now(inode, 1);
+	/* And now flush the block cache so that kernel sees the changes */
+	invalidate_bdev(sb->s_bdev, 0);
+	down(&inode->i_sem);
 	down(&dqopt->dqonoff_sem);
 	if (sb_has_quota_enabled(sb, type)) {
 		error = -EBUSY;
 		goto out_lock;
 	}
-	oldflags = inode->i_flags;
-	dqopt->files[type] = f;
-	error = -EINVAL;
-	if (!fmt->qf_ops->check_quota_file(sb, type))
-		goto out_file_init;
 	/* We don't want quota and atime on quota files (deadlocks possible)
-	 * We also need to set GFP mask differently because we cannot recurse
-	 * into filesystem when allocating page for quota inode */
+	 * Also nobody should write to the file - we use special IO operations
+	 * which ignore the immutable bit. */
 	down_write(&dqopt->dqptr_sem);
-	inode->i_flags |= S_NOQUOTA | S_NOATIME;
-
-	/*
-	 * We write to quota files deep within filesystem code.  We don't want
-	 * the VFS to reenter filesystem code when it tries to allocate a
-	 * pagecache page for the quota file write.  So clear __GFP_FS in
-	 * the quota file's allocation flags.
-	 */
-	mapping_set_gfp_mask(inode->i_mapping,
-		mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
-
-	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
-		to_drop[cnt] = inode->i_dquot[cnt];
-		inode->i_dquot[cnt] = NODQUOT;
-	}
-	inode->i_flags &= ~S_QUOTA;
+	oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
+	inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
 	up_write(&dqopt->dqptr_sem);
-	/* We must put dquots outside of dqptr_sem because we may need to
-	 * start transaction for write */
-	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
-		if (to_drop[cnt])
-			dqput(to_drop[cnt]);
-	}
+
+	error = -EIO;
+	dqopt->files[type] = igrab(inode);
+	if (!dqopt->files[type])
+		goto out_lock;
+	error = -EINVAL;
+	if (!fmt->qf_ops->check_quota_file(sb, type))
+		goto out_file_init;
 
 	dqopt->ops[type] = fmt->qf_ops;
 	dqopt->info[type].dqi_format = fmt;
@@ -1269,6 +1405,7 @@ int vfs_quota_on(struct super_block *sb,
 		goto out_file_init;
 	}
 	up(&dqopt->dqio_sem);
+	up(&inode->i_sem);
 	set_enable_flags(dqopt, type);
 
 	add_dquot_ref(sb, type);
@@ -1277,18 +1414,51 @@ int vfs_quota_on(struct super_block *sb,
 	return 0;
 
 out_file_init:
-	inode->i_flags = oldflags;