lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Wed, 04 Jun 2014 12:23:19 +0800
From:	Niu Yawei <yawei.niu@...il.com>
To:	Jan Kara <jack@...e.cz>
CC:	Christoph Hellwig <hch@...radead.org>,
	linux-fsdevel@...r.kernel.org, linux-ext4@...r.kernel.org,
	yawei.niu@...el.com, andreas.dilger@...el.com, lai.siyao@...el.com
Subject: [PATCH 5/5] quota: remove dqptr_sem

Remove dqptr_sem to make quota code scalable: Remove the dqptr_sem,
accessing inode->i_dquot now protected by dquot_srcu, and changing
inode->i_dquot is now serialized by dq_data_lock.

Signed-off-by: Lai Siyao <lai.siyao@...el.com>
Signed-off-by: Niu Yawei <yawei.niu@...el.com>
---
 fs/quota/dquot.c      |  114 +++++++++++++++++++++----------------------------
 fs/super.c            |    1 -
 include/linux/quota.h |    1 -
 3 files changed, 49 insertions(+), 67 deletions(-)

diff --git a/fs/quota/dquot.c b/fs/quota/dquot.c
index a94d464..5d2e02c 100644
--- a/fs/quota/dquot.c
+++ b/fs/quota/dquot.c
@@ -96,13 +96,16 @@
  * Note that some things (eg. sb pointer, type, id) doesn't change during
  * the life of the dquot structure and so needn't to be protected by a lock
  *
- * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
- * operation is just reading pointers from inode (or not using them at all) the
- * read lock is enough. If pointers are altered function must hold write lock.
+ * Operation accessing dquots via inode pointers are protected by dquot_srcu.
+ * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
+ * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
+ * inode and before dropping dquot references to avoid use of dquots after
+ * they are freed. dq_data_lock is used to serialize the pointer setting and
+ * clearing operations.
  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
  * inode is a quota file). Functions adding pointers from inode to dquots have
- * to check this flag under dqptr_sem and then (if S_NOQUOTA is not set) they
- * have to do all pointer modifications before dropping dqptr_sem. This makes
+ * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
+ * have to do all pointer modifications before dropping dq_data_lock. This makes
  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
  * then drops all pointers to dquots from an inode.
  *
@@ -116,21 +119,15 @@
  * spinlock to internal buffers before writing.
  *
  * Lock ordering (including related VFS locks) is the following:
- *   dqonoff_mutex > i_mutex > journal_lock > dqptr_sem > dquot->dq_lock >
- *   dqio_mutex
+ *   dqonoff_mutex > i_mutex > journal_lock > dquot->dq_lock > dqio_mutex
  * dqonoff_mutex > i_mutex comes from dquot_quota_sync, dquot_enable, etc.
- * The lock ordering of dqptr_sem imposed by quota code is only dqonoff_sem >
- * dqptr_sem. But filesystem has to count with the fact that functions such as
- * dquot_alloc_space() acquire dqptr_sem and they usually have to be called
- * from inside a transaction to keep filesystem consistency after a crash. Also
- * filesystems usually want to do some IO on dquot from ->mark_dirty which is
- * called with dqptr_sem held.
  */
 
 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
 EXPORT_SYMBOL(dq_data_lock);
+DEFINE_STATIC_SRCU(dquot_srcu);
 
 void __quota_error(struct super_block *sb, const char *func,
 		   const char *fmt, ...)
@@ -964,7 +961,6 @@ static void add_dquot_ref(struct super_block *sb, int type)
 /*
  * Remove references to dquots from inode and add dquot to list for freeing
  * if we have the last reference to dquot
- * We can't race with anybody because we hold dqptr_sem for writing...
  */
 static void remove_inode_dquot_ref(struct inode *inode, int type,
 				   struct list_head *tofree_head)
@@ -1024,13 +1020,15 @@ static void remove_dquot_ref(struct super_block *sb, int type,
 		 *  We have to scan also I_NEW inodes because they can already
 		 *  have quota pointer initialized. Luckily, we need to touch
 		 *  only quota pointers and these have separate locking
-		 *  (dqptr_sem).
+		 *  (dq_data_lock).
 		 */
+		spin_lock(&dq_data_lock);
 		if (!IS_NOQUOTA(inode)) {
 			if (unlikely(inode_get_rsv_space(inode) > 0))
 				reserved = 1;
 			remove_inode_dquot_ref(inode, type, tofree_head);
 		}
+		spin_unlock(&dq_data_lock);
 	}
 	spin_unlock(&inode_sb_list_lock);
 #ifdef CONFIG_QUOTA_DEBUG
@@ -1048,9 +1046,8 @@ static void drop_dquot_ref(struct super_block *sb, int type)
 	LIST_HEAD(tofree_head);
 
 	if (sb->dq_op) {
-		down_write(&sb_dqopt(sb)->dqptr_sem);
 		remove_dquot_ref(sb, type, &tofree_head);
-		up_write(&sb_dqopt(sb)->dqptr_sem);
+		synchronize_srcu(&dquot_srcu);
 		put_dquot_list(&tofree_head);
 	}
 }
@@ -1381,9 +1378,6 @@ static int dquot_active(const struct inode *inode)
 /*
  * Initialize quota pointers in inode
  *
- * We do things in a bit complicated way but by that we avoid calling
- * dqget() and thus filesystem callbacks under dqptr_sem.
- *
  * It is better to call this function outside of any transaction as it
  * might need a lot of space in journal for dquot structure allocation.
  */
@@ -1394,8 +1388,6 @@ static void __dquot_initialize(struct inode *inode, int type)
 	struct super_block *sb = inode->i_sb;
 	qsize_t rsv;
 
-	/* First test before acquiring mutex - solves deadlocks when we
-         * re-enter the quota code and are already holding the mutex */
 	if (!dquot_active(inode))
 		return;
 
@@ -1427,7 +1419,7 @@ static void __dquot_initialize(struct inode *inode, int type)
 	if (!init_needed)
 		return;
 
-	down_write(&sb_dqopt(sb)->dqptr_sem);
+	spin_lock(&dq_data_lock);
 	if (IS_NOQUOTA(inode))
 		goto out_err;
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
@@ -1447,15 +1439,12 @@ static void __dquot_initialize(struct inode *inode, int type)
 			 * did a write before quota was turned on
 			 */
 			rsv = inode_get_rsv_space(inode);
-			if (unlikely(rsv)) {
-				spin_lock(&dq_data_lock);
+			if (unlikely(rsv))
 				dquot_resv_space(inode->i_dquot[cnt], rsv);
-				spin_unlock(&dq_data_lock);
-			}
 		}
 	}
 out_err:
-	up_write(&sb_dqopt(sb)->dqptr_sem);
+	spin_unlock(&dq_data_lock);
 	/* Drop unused references */
 	dqput_all(got);
 }
@@ -1467,19 +1456,24 @@ void dquot_initialize(struct inode *inode)
 EXPORT_SYMBOL(dquot_initialize);
 
 /*
- * 	Release all quotas referenced by inode
+ * Release all quotas referenced by inode.
+ *
+ * This function only be called on inode free or converting
+ * a file to quota file, no other users for the i_dquot in
+ * both cases, so we needn't call synchronize_srcu() after
+ * clearing i_dquot.
  */
 static void __dquot_drop(struct inode *inode)
 {
 	int cnt;
 	struct dquot *put[MAXQUOTAS];
 
-	down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	spin_lock(&dq_data_lock);
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 		put[cnt] = inode->i_dquot[cnt];
 		inode->i_dquot[cnt] = NULL;
 	}
-	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	spin_unlock(&dq_data_lock);
 	dqput_all(put);
 }
 
@@ -1597,15 +1591,11 @@ static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
  */
 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
 {
-	int cnt, ret = 0;
+	int cnt, ret = 0, index;
 	struct dquot_warn warn[MAXQUOTAS];
 	struct dquot **dquots = inode->i_dquot;
 	int reserve = flags & DQUOT_SPACE_RESERVE;
 
-	/*
-	 * First test before acquiring mutex - solves deadlocks when we
-	 * re-enter the quota code and are already holding the mutex
-	 */
 	if (!dquot_active(inode)) {
 		inode_incr_space(inode, number, reserve);
 		goto out;
@@ -1614,7 +1604,7 @@ int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 		warn[cnt].w_type = QUOTA_NL_NOWARN;
 
-	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	index = srcu_read_lock(&dquot_srcu);
 	spin_lock(&dq_data_lock);
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 		if (!dquots[cnt])
@@ -1641,7 +1631,7 @@ int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
 		goto out_flush_warn;
 	mark_all_dquot_dirty(dquots);
 out_flush_warn:
-	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	srcu_read_unlock(&dquot_srcu, index);
 	flush_warnings(warn);
 out:
 	return ret;
@@ -1653,17 +1643,16 @@ EXPORT_SYMBOL(__dquot_alloc_space);
  */
 int dquot_alloc_inode(const struct inode *inode)
 {
-	int cnt, ret = 0;
+	int cnt, ret = 0, index;
 	struct dquot_warn warn[MAXQUOTAS];
 	struct dquot * const *dquots = inode->i_dquot;
 
-	/* First test before acquiring mutex - solves deadlocks when we
-         * re-enter the quota code and are already holding the mutex */
 	if (!dquot_active(inode))
 		return 0;
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 		warn[cnt].w_type = QUOTA_NL_NOWARN;
-	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+
+	index = srcu_read_lock(&dquot_srcu);
 	spin_lock(&dq_data_lock);
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 		if (!dquots[cnt])
@@ -1683,7 +1672,7 @@ warn_put_all:
 	spin_unlock(&dq_data_lock);
 	if (ret == 0)
 		mark_all_dquot_dirty(dquots);
-	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	srcu_read_unlock(&dquot_srcu, index);
 	flush_warnings(warn);
 	return ret;
 }
@@ -1694,14 +1683,14 @@ EXPORT_SYMBOL(dquot_alloc_inode);
  */
 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
 {
-	int cnt;
+	int cnt, index;
 
 	if (!dquot_active(inode)) {
 		inode_claim_rsv_space(inode, number);
 		return 0;
 	}
 
-	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	index = srcu_read_lock(&dquot_srcu);
 	spin_lock(&dq_data_lock);
 	/* Claim reserved quotas to allocated quotas */
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
@@ -1713,7 +1702,7 @@ int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
 	inode_claim_rsv_space(inode, number);
 	spin_unlock(&dq_data_lock);
 	mark_all_dquot_dirty(inode->i_dquot);
-	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	srcu_read_unlock(&dquot_srcu, index);
 	return 0;
 }
 EXPORT_SYMBOL(dquot_claim_space_nodirty);
@@ -1723,14 +1712,14 @@ EXPORT_SYMBOL(dquot_claim_space_nodirty);
  */
 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
 {
-	int cnt;
+	int cnt, index;
 
 	if (!dquot_active(inode)) {
 		inode_reclaim_rsv_space(inode, number);
 		return;
 	}
 
-	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	index = srcu_read_lock(&dquot_srcu);
 	spin_lock(&dq_data_lock);
 	/* Claim reserved quotas to allocated quotas */
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
@@ -1742,7 +1731,7 @@ void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
 	inode_reclaim_rsv_space(inode, number);
 	spin_unlock(&dq_data_lock);
 	mark_all_dquot_dirty(inode->i_dquot);
-	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	srcu_read_unlock(&dquot_srcu, index);
 	return;
 }
 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
@@ -1755,16 +1744,14 @@ void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
 	unsigned int cnt;
 	struct dquot_warn warn[MAXQUOTAS];
 	struct dquot **dquots = inode->i_dquot;
-	int reserve = flags & DQUOT_SPACE_RESERVE;
+	int reserve = flags & DQUOT_SPACE_RESERVE, index;
 
-	/* First test before acquiring mutex - solves deadlocks when we
-         * re-enter the quota code and are already holding the mutex */
 	if (!dquot_active(inode)) {
 		inode_decr_space(inode, number, reserve);
 		return;
 	}
 
-	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	index = srcu_read_lock(&dquot_srcu);
 	spin_lock(&dq_data_lock);
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 		int wtype;
@@ -1787,7 +1774,7 @@ void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
 		goto out_unlock;
 	mark_all_dquot_dirty(dquots);
 out_unlock:
-	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	srcu_read_unlock(&dquot_srcu, index);
 	flush_warnings(warn);
 }
 EXPORT_SYMBOL(__dquot_free_space);
@@ -1800,13 +1787,12 @@ void dquot_free_inode(const struct inode *inode)
 	unsigned int cnt;
 	struct dquot_warn warn[MAXQUOTAS];
 	struct dquot * const *dquots = inode->i_dquot;
+	int index;
 
-	/* First test before acquiring mutex - solves deadlocks when we
-         * re-enter the quota code and are already holding the mutex */
 	if (!dquot_active(inode))
 		return;
 
-	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	index = srcu_read_lock(&dquot_srcu);
 	spin_lock(&dq_data_lock);
 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 		int wtype;
@@ -1821,7 +1807,7 @@ void dquot_free_inode(const struct inode *inode)
 	}
 	spin_unlock(&dq_data_lock);
 	mark_all_dquot_dirty(dquots);
-	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
+	srcu_read_unlock(&dquot_srcu, index);
 	flush_warnings(warn);
 }
 EXPORT_SYMBOL(dquot_free_inode);
@@ -1835,6 +1821,8 @@ EXPORT_SYMBOL(dquot_free_inode);
  * This operation can block, but only after everything is updated
  * A transaction must be started when entering this function.
  *
+ * We are holding reference on transfer_from & transfer_to, no need to
+ * protect them by srcu_read_lock().
  */
 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
 {
@@ -1847,8 +1835,6 @@ int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
 	struct dquot_warn warn_from_inodes[MAXQUOTAS];
 	struct dquot_warn warn_from_space[MAXQUOTAS];
 
-	/* First test before acquiring mutex - solves deadlocks when we
-         * re-enter the quota code and are already holding the mutex */
 	if (IS_NOQUOTA(inode))
 		return 0;
 	/* Initialize the arrays */
@@ -1857,12 +1843,12 @@ int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
 		warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
 		warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
 	}
-	down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
+
+	spin_lock(&dq_data_lock);
 	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
-		up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
+		spin_unlock(&dq_data_lock);
 		return 0;
 	}
-	spin_lock(&dq_data_lock);
 	cur_space = inode_get_bytes(inode);
 	rsv_space = inode_get_rsv_space(inode);
 	space = cur_space + rsv_space;
@@ -1916,7 +1902,6 @@ int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
 		inode->i_dquot[cnt] = transfer_to[cnt];
 	}
 	spin_unlock(&dq_data_lock);
-	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 
 	mark_all_dquot_dirty(transfer_from);
 	mark_all_dquot_dirty(transfer_to);
@@ -1930,7 +1915,6 @@ int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
 	return 0;
 over_quota:
 	spin_unlock(&dq_data_lock);
-	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 	flush_warnings(warn_to);
 	return ret;
 }
diff --git a/fs/super.c b/fs/super.c
index 48377f7..a97aecf 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -214,7 +214,6 @@ static struct super_block *alloc_super(struct file_system_type *type, int flags)
 	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
 	mutex_init(&s->s_dquot.dqio_mutex);
 	mutex_init(&s->s_dquot.dqonoff_mutex);
-	init_rwsem(&s->s_dquot.dqptr_sem);
 	s->s_maxbytes = MAX_NON_LFS;
 	s->s_op = &default_op;
 	s->s_time_gran = 1000000000;
diff --git a/include/linux/quota.h b/include/linux/quota.h
index cc7494a..a1358ce 100644
--- a/include/linux/quota.h
+++ b/include/linux/quota.h
@@ -389,7 +389,6 @@ struct quota_info {
 	unsigned int flags;			/* Flags for diskquotas on this device */
 	struct mutex dqio_mutex;		/* lock device while I/O in progress */
 	struct mutex dqonoff_mutex;		/* Serialize quotaon & quotaoff */
-	struct rw_semaphore dqptr_sem;		/* serialize ops using quota_info struct, pointers from inode to dquots */
 	struct inode *files[MAXQUOTAS];		/* inodes of quotafiles */
 	struct mem_dqinfo info[MAXQUOTAS];	/* Information for each quota type */
 	const struct quota_format_ops *ops[MAXQUOTAS];	/* Operations for each type */
-- 
1.7.1


--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists