reiserfs locking patchset v4

PATCH In review Jeff Mahoney 2 participants 5 messages Aug 8, 2013 – Aug 9, 2013

Share & tools

Jeff Mahoney

author

2013-08-08 21:36 UTC #0 lore raw

v4 cover reiserfs locking patchset v4

This patchset untangles some of the locking in reiserfs. It has seen more testing as part of the maintenance cycle in SLE 11 SP2.

- We push the write lock out of the xattr code. It doesn't need it and we can simplify locking by releasing and reacquiring the locks around the xattr calls.

- Handle nested locks properly. There's confusion on when a lock is nested, when it's not, and how to drop it across schedules like the BKL it is modeled after. We make the distinction between taking/releasing the lock and when to drop it for schedules and simplify the logic. This fixes a number of deadlocks that happen because the intention was to drop the write lock but it really only decremented the use count.

- Fix the deadlocks with the quota code. This involves dropping the write lock before quota calls and reacquring it afterwards. Without this patch reiserfs quotas are essentially unusable.

Previous postings of this patchset were mismerged.

This revision integrates Jan's advice about search_by_key_reada returning the depth instead of a bool.

Jan - This series is in the for-3.12 branch of git.kernel.org:/pub/scm/linux/kernel/git/jeffm/linux-reiserfs.git

Thanks.

-Jeff

Jeff Mahoney

author

2013-08-08 21:36 UTC #1 lore raw

PATCH v4 1/3 reiserfs: locking, push write lock out of xattr code

The reiserfs xattr code doesn't need the write lock and sleeps all over the place. We can simplify the locking by releasing it and reacquiring after the xattr call.

Signed-off-by Jeff Mahoney <[email protected]>

---
 fs/reiserfs/inode.c     |   38 ++++++++++++++++++++------------------
 fs/reiserfs/super.c     |   48 +++++++++++++++++++++++-------------------------
 fs/reiserfs/xattr.c     |   46 +++++++++++++++++-----------------------------
 fs/reiserfs/xattr_acl.c |   16 ++++++++++------
 4 files changed, 70 insertions(+), 78 deletions(-)

fs/reiserfs/inode.c +20 ~~-18~~

@@ -30,7 +30,6 @@ void reiserfs_evict_inode(struct inode *
 	    JOURNAL_PER_BALANCE_CNT * 2 +
 	    2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
 	struct reiserfs_transaction_handle th;
-	int depth;
 	int err;
 
 	if (!inode->i_nlink && !is_bad_inode(inode))

@@ -40,12 +39,14 @@ void reiserfs_evict_inode(struct inode *
 	if (inode->i_nlink)
 		goto no_delete;
 
-	depth = reiserfs_write_lock_once(inode->i_sb);
-
 	/* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
 	if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {	/* also handles bad_inode case */
+		int depth;
+
 		reiserfs_delete_xattrs(inode);
 
+		depth = reiserfs_write_lock_once(inode->i_sb);
+
 		if (journal_begin(&th, inode->i_sb, jbegin_count))
 			goto out;
 		reiserfs_update_inode_transaction(inode);

@@ -72,12 +73,12 @@ void reiserfs_evict_inode(struct inode *
 		/* all items of file are deleted, so we can remove "save" link */
 		remove_save_link(inode, 0 /* not truncate */ );	/* we can't do anything
 								 * about an error here */
+out:
+		reiserfs_write_unlock_once(inode->i_sb, depth);
 	} else {
 		/* no object items are in the tree */
 		;
 	}
-      out:
-	reiserfs_write_unlock_once(inode->i_sb, depth);
 	clear_inode(inode);	/* note this must go after the journal_end to prevent deadlock */
 	dquot_drop(inode);
 	inode->i_blocks = 0;

@@ -1941,7 +1942,9 @@ int reiserfs_new_inode(struct reiserfs_t
 	}
 
 	if (reiserfs_posixacl(inode->i_sb)) {
+		reiserfs_write_unlock(inode->i_sb);
 		retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
+		reiserfs_write_lock(inode->i_sb);
 		if (retval) {
 			err = retval;
 			reiserfs_check_path(&path_to_key);

@@ -1956,7 +1959,9 @@ int reiserfs_new_inode(struct reiserfs_t
 		inode->i_flags |= S_PRIVATE;
 
 	if (security->name) {
+		reiserfs_write_unlock(inode->i_sb);
 		retval = reiserfs_security_write(th, inode, security);
+		reiserfs_write_lock(inode->i_sb);
 		if (retval) {
 			err = retval;
 			reiserfs_check_path(&path_to_key);

@@ -3129,6 +3134,7 @@ int reiserfs_setattr(struct dentry *dent
 		 */
 		if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
 		    attr->ia_size > MAX_NON_LFS) {
+			reiserfs_write_unlock_once(inode->i_sb, depth);
 			error = -EFBIG;
 			goto out;
 		}

@@ -3150,8 +3156,10 @@ int reiserfs_setattr(struct dentry *dent
 				if (err)
 					error = err;
 			}
-			if (error)
+			if (error) {
+				reiserfs_write_unlock_once(inode->i_sb, depth);
 				goto out;
+			}
 			/*
 			 * file size is changed, ctime and mtime are
 			 * to be updated

@@ -3159,6 +3167,7 @@ int reiserfs_setattr(struct dentry *dent
 			attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
 		}
 	}
+	reiserfs_write_unlock_once(inode->i_sb, depth);
 
 	if ((((attr->ia_valid & ATTR_UID) && (from_kuid(&init_user_ns, attr->ia_uid) & ~0xffff)) ||
 	     ((attr->ia_valid & ATTR_GID) && (from_kgid(&init_user_ns, attr->ia_gid) & ~0xffff))) &&

@@ -3183,14 +3192,16 @@ int reiserfs_setattr(struct dentry *dent
 			return error;
 
 		/* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
+		depth = reiserfs_write_lock_once(inode->i_sb);
 		error = journal_begin(&th, inode->i_sb, jbegin_count);
+		reiserfs_write_unlock_once(inode->i_sb, depth);
 		if (error)
 			goto out;
-		reiserfs_write_unlock_once(inode->i_sb, depth);
 		error = dquot_transfer(inode, attr);
 		depth = reiserfs_write_lock_once(inode->i_sb);
 		if (error) {
 			journal_end(&th, inode->i_sb, jbegin_count);
+			reiserfs_write_unlock_once(inode->i_sb, depth);
 			goto out;
 		}

@@ -3202,17 +3213,11 @@ int reiserfs_setattr(struct dentry *dent
 			inode->i_gid = attr->ia_gid;
 		mark_inode_dirty(inode);
 		error = journal_end(&th, inode->i_sb, jbegin_count);
+		reiserfs_write_unlock_once(inode->i_sb, depth);
 		if (error)
 			goto out;
 	}
 
-	/*
-	 * Relax the lock here, as it might truncate the
-	 * inode pages and wait for inode pages locks.
-	 * To release such page lock, the owner needs the
-	 * reiserfs lock
-	 */
-	reiserfs_write_unlock_once(inode->i_sb, depth);
 	if ((attr->ia_valid & ATTR_SIZE) &&
 	    attr->ia_size != i_size_read(inode)) {
 		error = inode_newsize_ok(inode, attr->ia_size);

@@ -3226,16 +3231,13 @@ int reiserfs_setattr(struct dentry *dent
 		setattr_copy(inode, attr);
 		mark_inode_dirty(inode);
 	}
-	depth = reiserfs_write_lock_once(inode->i_sb);
 
 	if (!error && reiserfs_posixacl(inode->i_sb)) {
 		if (attr->ia_valid & ATTR_MODE)
 			error = reiserfs_acl_chmod(inode);
 	}
 
-      out:
-	reiserfs_write_unlock_once(inode->i_sb, depth);
-
+out:
 	return error;
 }

fs/reiserfs/super.c +23 ~~-25~~

135 lines not on this page · raw message

fs/reiserfs/xattr.c +17 ~~-29~~

163 lines not on this page · raw message

fs/reiserfs/xattr_acl.c +10 -6

@@ -49,13 +49,15 @@ posix_acl_set(struct dentry *dentry, con
 
 	reiserfs_write_lock(inode->i_sb);
 	error = journal_begin(&th, inode->i_sb, jcreate_blocks);
+	reiserfs_write_unlock(inode->i_sb);
 	if (error == 0) {
 		error = reiserfs_set_acl(&th, inode, type, acl);
+		reiserfs_write_lock(inode->i_sb);
 		error2 = journal_end(&th, inode->i_sb, jcreate_blocks);
+		reiserfs_write_unlock(inode->i_sb);
 		if (error2)
 			error = error2;
 	}
-	reiserfs_write_unlock(inode->i_sb);
 
       release_and_out:
 	posix_acl_release(acl);

@@ -435,12 +437,14 @@ int reiserfs_cache_default_acl(struct in
 	return nblocks;
 }
 
+/*
+ * Called under i_mutex
+ */
 int reiserfs_acl_chmod(struct inode *inode)
 {
 	struct reiserfs_transaction_handle th;
 	struct posix_acl *acl;
 	size_t size;
-	int depth;
 	int error;
 
 	if (IS_PRIVATE(inode))

@@ -454,9 +458,7 @@ int reiserfs_acl_chmod(struct inode *ino
 		return 0;
 	}
 
-	reiserfs_write_unlock(inode->i_sb);
 	acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS);
-	reiserfs_write_lock(inode->i_sb);
 	if (!acl)
 		return 0;
 	if (IS_ERR(acl))

@@ -466,16 +468,18 @@ int reiserfs_acl_chmod(struct inode *ino
 		return error;
 
 	size = reiserfs_xattr_nblocks(inode, reiserfs_acl_size(acl->a_count));
-	depth = reiserfs_write_lock_once(inode->i_sb);
+	reiserfs_write_lock(inode->i_sb);
 	error = journal_begin(&th, inode->i_sb, size * 2);
+	reiserfs_write_unlock(inode->i_sb);
 	if (!error) {
 		int error2;
 		error = reiserfs_set_acl(&th, inode, ACL_TYPE_ACCESS, acl);
+		reiserfs_write_lock(inode->i_sb);
 		error2 = journal_end(&th, inode->i_sb, size * 2);
+		reiserfs_write_unlock(inode->i_sb);
 		if (error2)
 			error = error2;
 	}
-	reiserfs_write_unlock_once(inode->i_sb, depth);
 	posix_acl_release(acl);
 	return error;
 }

Jeff Mahoney

author

2013-08-08 21:36 UTC #2 lore raw

PATCH v4 2/3 reiserfs: locking, handle nested locks properly

In reply to Jeff Mahoney

The reiserfs write lock replaced the BKL and uses similar semantics.

Frederic's locking code makes a distinction between when the lock is nested and when it's being acquired/released, but I don't think that's the right distinction to make.

The right distinction is between the lock being released at end-of-use and the lock being released for a schedule. The unlock should return the depth and the lock should restore it, rather than the other way around as it is now.

This patch implements that and adds a number of places where the lock should be dropped.

Signed-off-by Jeff Mahoney <[email protected]>

---

 fs/reiserfs/bitmap.c   |    5 +-
 fs/reiserfs/dir.c      |    7 +--
 fs/reiserfs/fix_node.c |   26 ++++++------
 fs/reiserfs/inode.c    |   77 ++++++++++++++++--------------------
 fs/reiserfs/ioctl.c    |    7 +--
 fs/reiserfs/journal.c  |  104 ++++++++++++++++++++++++++-----------------------
 fs/reiserfs/lock.c     |   43 ++++++++++----------
 fs/reiserfs/namei.c    |   24 +++--------
 fs/reiserfs/prints.c   |    5 +-
 fs/reiserfs/reiserfs.h |   36 +++++++++-------
 fs/reiserfs/resize.c   |   10 +++-
 fs/reiserfs/stree.c    |   46 +++++++++------------
 fs/reiserfs/super.c    |    5 --
 13 files changed, 201 insertions(+), 194 deletions(-)

fs/reiserfs/bitmap.c +3 -2

@@ -1340,10 +1340,11 @@ struct buffer_head *reiserfs_read_bitmap
 		                 "reading failed", __func__, block);
 	else {
 		if (buffer_locked(bh)) {
+			int depth;
 			PROC_INFO_INC(sb, scan_bitmap.wait);
-			reiserfs_write_unlock(sb);
+			depth = reiserfs_write_unlock_nested(sb);
 			__wait_on_buffer(bh);
-			reiserfs_write_lock(sb);
+			reiserfs_write_lock_nested(sb, depth);
 		}
 		BUG_ON(!buffer_uptodate(bh));
 		BUG_ON(atomic_read(&bh->b_count) == 0);

fs/reiserfs/dir.c +4 -3

@@ -71,6 +71,7 @@ int reiserfs_readdir_inode(struct inode
 	char small_buf[32];	/* avoid kmalloc if we can */
 	struct reiserfs_dir_entry de;
 	int ret = 0;
+	int depth;
 
 	reiserfs_write_lock(inode->i_sb);

@@ -181,17 +182,17 @@ int reiserfs_readdir_inode(struct inode
 				 * Since filldir might sleep, we can release
 				 * the write lock here for other waiters
 				 */
-				reiserfs_write_unlock(inode->i_sb);
+				depth = reiserfs_write_unlock_nested(inode->i_sb);
 				if (!dir_emit
 				    (ctx, local_buf, d_reclen, d_ino,
 				     DT_UNKNOWN)) {
-					reiserfs_write_lock(inode->i_sb);
+					reiserfs_write_lock_nested(inode->i_sb, depth);
 					if (local_buf != small_buf) {
 						kfree(local_buf);
 					}
 					goto end;
 				}
-				reiserfs_write_lock(inode->i_sb);
+				reiserfs_write_lock_nested(inode->i_sb, depth);
 				if (local_buf != small_buf) {
 					kfree(local_buf);
 				}

fs/reiserfs/fix_node.c +14 ~~-12~~

@@ -1022,9 +1022,9 @@ static int get_far_parent(struct tree_ba
 	if (buffer_locked(*pcom_father)) {
 
 		/* Release the write lock while the buffer is busy */
-		reiserfs_write_unlock(tb->tb_sb);
+		int depth = reiserfs_write_unlock_nested(tb->tb_sb);
 		__wait_on_buffer(*pcom_father);
-		reiserfs_write_lock(tb->tb_sb);
+		reiserfs_write_lock_nested(tb->tb_sb, depth);
 		if (FILESYSTEM_CHANGED_TB(tb)) {
 			brelse(*pcom_father);
 			return REPEAT_SEARCH;

@@ -1929,9 +1929,9 @@ static int get_direct_parent(struct tree
 		return REPEAT_SEARCH;
 
 	if (buffer_locked(bh)) {
-		reiserfs_write_unlock(tb->tb_sb);
+		int depth = reiserfs_write_unlock_nested(tb->tb_sb);
 		__wait_on_buffer(bh);
-		reiserfs_write_lock(tb->tb_sb);
+		reiserfs_write_lock_nested(tb->tb_sb, depth);
 		if (FILESYSTEM_CHANGED_TB(tb))
 			return REPEAT_SEARCH;
 	}

@@ -1952,6 +1952,7 @@ static int get_neighbors(struct tree_bal
 	unsigned long son_number;
 	struct super_block *sb = tb->tb_sb;
 	struct buffer_head *bh;
+	int depth;
 
 	PROC_INFO_INC(sb, get_neighbors[h]);

@@ -1969,9 +1970,9 @@ static int get_neighbors(struct tree_bal
 		     tb->FL[h]) ? tb->lkey[h] : B_NR_ITEMS(tb->
 								       FL[h]);
 		son_number = B_N_CHILD_NUM(tb->FL[h], child_position);
-		reiserfs_write_unlock(sb);
+		depth = reiserfs_write_unlock_nested(tb->tb_sb);
 		bh = sb_bread(sb, son_number);
-		reiserfs_write_lock(sb);
+		reiserfs_write_lock_nested(tb->tb_sb, depth);
 		if (!bh)
 			return IO_ERROR;
 		if (FILESYSTEM_CHANGED_TB(tb)) {

@@ -2009,9 +2010,9 @@ static int get_neighbors(struct tree_bal
 		child_position =
 		    (bh == tb->FR[h]) ? tb->rkey[h] + 1 : 0;
 		son_number = B_N_CHILD_NUM(tb->FR[h], child_position);
-		reiserfs_write_unlock(sb);
+		depth = reiserfs_write_unlock_nested(tb->tb_sb);
 		bh = sb_bread(sb, son_number);
-		reiserfs_write_lock(sb);
+		reiserfs_write_lock_nested(tb->tb_sb, depth);
 		if (!bh)
 			return IO_ERROR;
 		if (FILESYSTEM_CHANGED_TB(tb)) {

@@ -2272,6 +2273,7 @@ static int wait_tb_buffers_until_unlocke
 		}
 
 		if (locked) {
+			int depth;
 #ifdef CONFIG_REISERFS_CHECK
 			repeat_counter++;
 			if ((repeat_counter % 10000) == 0) {

@@ -2286,9 +2288,9 @@ static int wait_tb_buffers_until_unlocke
 				    REPEAT_SEARCH : CARRY_ON;
 			}
 #endif
-			reiserfs_write_unlock(tb->tb_sb);
+			depth = reiserfs_write_unlock_nested(tb->tb_sb);
 			__wait_on_buffer(locked);
-			reiserfs_write_lock(tb->tb_sb);
+			reiserfs_write_lock_nested(tb->tb_sb, depth);
 			if (FILESYSTEM_CHANGED_TB(tb))
 				return REPEAT_SEARCH;
 		}

@@ -2359,9 +2361,9 @@ int fix_nodes(int op_mode, struct tree_b
 
 	/* if it possible in indirect_to_direct conversion */
 	if (buffer_locked(tbS0)) {
-		reiserfs_write_unlock(tb->tb_sb);
+		int depth = reiserfs_write_unlock_nested(tb->tb_sb);
 		__wait_on_buffer(tbS0);
-		reiserfs_write_lock(tb->tb_sb);
+		reiserfs_write_lock_nested(tb->tb_sb, depth);
 		if (FILESYSTEM_CHANGED_TB(tb))
 			return REPEAT_SEARCH;
 	}

fs/reiserfs/inode.c +36 ~~-41~~

270 lines not on this page · raw message

fs/reiserfs/ioctl.c +3 -4

@@ -167,7 +167,6 @@ int reiserfs_commit_write(struct file *f
 int reiserfs_unpack(struct inode *inode, struct file *filp)
 {
 	int retval = 0;
-	int depth;
 	int index;
 	struct page *page;
 	struct address_space *mapping;

@@ -183,11 +182,11 @@ int reiserfs_unpack(struct inode *inode,
 		return 0;
 	}
 
-	depth = reiserfs_write_lock_once(inode->i_sb);
-
 	/* we need to make sure nobody is changing the file size beneath us */
 	reiserfs_mutex_lock_safe(&inode->i_mutex, inode->i_sb);
 
+	reiserfs_write_lock(inode->i_sb);
+
 	write_from = inode->i_size & (blocksize - 1);
 	/* if we are on a block boundary, we are already unpacked.  */
 	if (write_from == 0) {

@@ -221,6 +220,6 @@ int reiserfs_unpack(struct inode *inode,
 
       out:
 	mutex_unlock(&inode->i_mutex);
-	reiserfs_write_unlock_once(inode->i_sb, depth);
+	reiserfs_write_unlock(inode->i_sb);
 	return retval;
 }

fs/reiserfs/journal.c +56 ~~-48~~

298 lines not on this page · raw message

fs/reiserfs/lock.c +23 ~~-20~~

62 lines not on this page · raw message

fs/reiserfs/namei.c +8 ~~-16~~

@@ -325,7 +325,6 @@ static struct dentry *reiserfs_lookup(st
 				      unsigned int flags)
 {
 	int retval;
-	int lock_depth;
 	struct inode *inode = NULL;
 	struct reiserfs_dir_entry de;
 	INITIALIZE_PATH(path_to_entry);

@@ -333,12 +332,7 @@ static struct dentry *reiserfs_lookup(st
 	if (REISERFS_MAX_NAME(dir->i_sb->s_blocksize) < dentry->d_name.len)
 		return ERR_PTR(-ENAMETOOLONG);
 
-	/*
-	 * Might be called with or without the write lock, must be careful
-	 * to not recursively hold it in case we want to release the lock
-	 * before rescheduling.
-	 */
-	lock_depth = reiserfs_write_lock_once(dir->i_sb);
+	reiserfs_write_lock(dir->i_sb);
 
 	de.de_gen_number_bit_string = NULL;
 	retval =

@@ -349,7 +343,7 @@ static struct dentry *reiserfs_lookup(st
 		inode = reiserfs_iget(dir->i_sb,
 				      (struct cpu_key *)&(de.de_dir_id));
 		if (!inode || IS_ERR(inode)) {
-			reiserfs_write_unlock_once(dir->i_sb, lock_depth);
+			reiserfs_write_unlock(dir->i_sb);
 			return ERR_PTR(-EACCES);
 		}

@@ -358,7 +352,7 @@ static struct dentry *reiserfs_lookup(st
 		if (IS_PRIVATE(dir))
 			inode->i_flags |= S_PRIVATE;
 	}
-	reiserfs_write_unlock_once(dir->i_sb, lock_depth);
+	reiserfs_write_unlock(dir->i_sb);
 	if (retval == IO_ERROR) {
 		return ERR_PTR(-EIO);
 	}

@@ -727,7 +721,6 @@ static int reiserfs_mkdir(struct inode *
 	struct inode *inode;
 	struct reiserfs_transaction_handle th;
 	struct reiserfs_security_handle security;
-	int lock_depth;
 	/* We need blocks for transaction + (user+group)*(quotas for new inode + update of quota for directory owner) */
 	int jbegin_count =
 	    JOURNAL_PER_BALANCE_CNT * 3 +

@@ -753,7 +746,7 @@ static int reiserfs_mkdir(struct inode *
 		return retval;
 	}
 	jbegin_count += retval;
-	lock_depth = reiserfs_write_lock_once(dir->i_sb);
+	reiserfs_write_lock(dir->i_sb);
 
 	retval = journal_begin(&th, dir->i_sb, jbegin_count);
 	if (retval) {

@@ -804,7 +797,7 @@ static int reiserfs_mkdir(struct inode *
 	d_instantiate(dentry, inode);
 	retval = journal_end(&th, dir->i_sb, jbegin_count);
 out_failed:
-	reiserfs_write_unlock_once(dir->i_sb, lock_depth);
+	reiserfs_write_unlock(dir->i_sb);
 	return retval;
 }

@@ -920,7 +913,6 @@ static int reiserfs_unlink(struct inode
 	struct reiserfs_transaction_handle th;
 	int jbegin_count;
 	unsigned long savelink;
-	int depth;
 
 	dquot_initialize(dir);

@@ -934,7 +926,7 @@ static int reiserfs_unlink(struct inode
 	    JOURNAL_PER_BALANCE_CNT * 2 + 2 +
 	    4 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
 
-	depth = reiserfs_write_lock_once(dir->i_sb);
+	reiserfs_write_lock(dir->i_sb);
 	retval = journal_begin(&th, dir->i_sb, jbegin_count);
 	if (retval)
 		goto out_unlink;

@@ -995,7 +987,7 @@ static int reiserfs_unlink(struct inode
 
 	retval = journal_end(&th, dir->i_sb, jbegin_count);
 	reiserfs_check_path(&path);
-	reiserfs_write_unlock_once(dir->i_sb, depth);
+	reiserfs_write_unlock(dir->i_sb);
 	return retval;
 
       end_unlink:

@@ -1005,7 +997,7 @@ static int reiserfs_unlink(struct inode
 	if (err)
 		retval = err;
       out_unlink:
-	reiserfs_write_unlock_once(dir->i_sb, depth);
+	reiserfs_write_unlock(dir->i_sb);
 	return retval;
 }

fs/reiserfs/prints.c +3 -2

@@ -358,12 +358,13 @@ void __reiserfs_panic(struct super_block
 	dump_stack();
 #endif
 	if (sb)
-		panic(KERN_WARNING "REISERFS panic (device %s): %s%s%s: %s\n",
+		printk(KERN_WARNING "REISERFS panic (device %s): %s%s%s: %s\n",
 		      sb->s_id, id ? id : "", id ? " " : "",
 		      function, error_buf);
 	else
-		panic(KERN_WARNING "REISERFS panic: %s%s%s: %s\n",
+		printk(KERN_WARNING "REISERFS panic: %s%s%s: %s\n",
 		      id ? id : "", id ? " " : "", function, error_buf);
+	BUG();
 }
 
 void __reiserfs_error(struct super_block *sb, const char *id,

fs/reiserfs/reiserfs.h +20 ~~-16~~

@@ -630,8 +630,8 @@ static inline int __reiserfs_is_journal_
  */
 void reiserfs_write_lock(struct super_block *s);
 void reiserfs_write_unlock(struct super_block *s);
-int reiserfs_write_lock_once(struct super_block *s);
-void reiserfs_write_unlock_once(struct super_block *s, int lock_depth);
+int __must_check reiserfs_write_unlock_nested(struct super_block *s);
+void reiserfs_write_lock_nested(struct super_block *s, int depth);
 
 #ifdef CONFIG_REISERFS_CHECK
 void reiserfs_lock_check_recursive(struct super_block *s);

@@ -667,31 +667,33 @@ static inline void reiserfs_lock_check_r
  * - The inode mutex
  */
 static inline void reiserfs_mutex_lock_safe(struct mutex *m,
-			       struct super_block *s)
+					    struct super_block *s)
 {
-	reiserfs_lock_check_recursive(s);
-	reiserfs_write_unlock(s);
+	int depth;
+
+	depth = reiserfs_write_unlock_nested(s);
 	mutex_lock(m);
-	reiserfs_write_lock(s);
+	reiserfs_write_lock_nested(s, depth);
 }
 
 static inline void
 reiserfs_mutex_lock_nested_safe(struct mutex *m, unsigned int subclass,
-			       struct super_block *s)
+				struct super_block *s)
 {
-	reiserfs_lock_check_recursive(s);
-	reiserfs_write_unlock(s);
+	int depth;
+
+	depth = reiserfs_write_unlock_nested(s);
 	mutex_lock_nested(m, subclass);
-	reiserfs_write_lock(s);
+	reiserfs_write_lock_nested(s, depth);
 }
 
 static inline void
 reiserfs_down_read_safe(struct rw_semaphore *sem, struct super_block *s)
 {
-	reiserfs_lock_check_recursive(s);
-	reiserfs_write_unlock(s);
-	down_read(sem);
-	reiserfs_write_lock(s);
+       int depth;
+       depth = reiserfs_write_unlock_nested(s);
+       down_read(sem);
+       reiserfs_write_lock_nested(s, depth);
 }
 
 /*

@@ -701,9 +703,11 @@ reiserfs_down_read_safe(struct rw_semaph
 static inline void reiserfs_cond_resched(struct super_block *s)
 {
 	if (need_resched()) {
-		reiserfs_write_unlock(s);
+		int depth;
+
+		depth = reiserfs_write_unlock_nested(s);
 		schedule();
-		reiserfs_write_lock(s);
+		reiserfs_write_lock_nested(s, depth);
 	}
 }

fs/reiserfs/resize.c +8 -2

@@ -34,6 +34,7 @@ int reiserfs_resize(struct super_block *
 	unsigned long int block_count, free_blocks;
 	int i;
 	int copy_size;
+	int depth;
 
 	sb = SB_DISK_SUPER_BLOCK(s);

@@ -43,7 +44,9 @@ int reiserfs_resize(struct super_block *
 	}
 
 	/* check the device size */
+	depth = reiserfs_write_unlock_nested(s);
 	bh = sb_bread(s, block_count_new - 1);
+	reiserfs_write_lock_nested(s, depth);
 	if (!bh) {
 		printk("reiserfs_resize: can\'t read last block\n");
 		return -EINVAL;

@@ -125,9 +128,12 @@ int reiserfs_resize(struct super_block *
 		 * transaction begins, and the new bitmaps don't matter if the
 		 * transaction fails. */
 		for (i = bmap_nr; i < bmap_nr_new; i++) {
+			int depth;
 			/* don't use read_bitmap_block since it will cache
 			 * the uninitialized bitmap */
+			depth = reiserfs_write_unlock_nested(s);
 			bh = sb_bread(s, i * s->s_blocksize * 8);
+			reiserfs_write_lock_nested(s, depth);
 			if (!bh) {
 				vfree(bitmap);
 				return -EIO;

@@ -138,9 +144,9 @@ int reiserfs_resize(struct super_block *
 
 			set_buffer_uptodate(bh);
 			mark_buffer_dirty(bh);
-			reiserfs_write_unlock(s);
+			depth = reiserfs_write_unlock_nested(s);
 			sync_dirty_buffer(bh);
-			reiserfs_write_lock(s);
+			reiserfs_write_lock_nested(s, depth);
 			// update bitmap_info stuff
 			bitmap[i].free_count = sb_blocksize(sb) * 8 - 1;
 			brelse(bh);

fs/reiserfs/stree.c +21 ~~-25~~

85 lines not on this page · raw message

fs/reiserfs/super.c +2 -3

@@ -624,7 +624,6 @@ static void reiserfs_dirty_inode(struct
 	struct reiserfs_transaction_handle th;
 
 	int err = 0;
-	int lock_depth;
 
 	if (inode->i_sb->s_flags & MS_RDONLY) {
 		reiserfs_warning(inode->i_sb, "clm-6006",

@@ -632,7 +631,7 @@ static void reiserfs_dirty_inode(struct
 				 inode->i_ino);
 		return;
 	}
-	lock_depth = reiserfs_write_lock_once(inode->i_sb);
+	reiserfs_write_lock(inode->i_sb);
 
 	/* this is really only used for atime updates, so they don't have
 	 ** to be included in O_SYNC or fsync

@@ -645,7 +644,7 @@ static void reiserfs_dirty_inode(struct
 	journal_end(&th, inode->i_sb, 1);
 
 out:
-	reiserfs_write_unlock_once(inode->i_sb, lock_depth);
+	reiserfs_write_unlock(inode->i_sb);
 }
 
 static int reiserfs_show_options(struct seq_file *seq, struct dentry *root)

Jeff Mahoney

author

2013-08-08 21:36 UTC #3 lore raw

PATCH v4 3/3 reiserfs: locking, release lock around quota operations

In reply to Jeff Mahoney

Previous commits released the write lock across quota operations but missed several places. In particular, the free operations can also call into the file system code and take the write lock, causing deadlocks.

This patch introduces some more helpers and uses them for quota call sites. Without this patch applied, reiserfs + quotas runs into deadlocks under anything more than trivial load.

Signed-off-by Jeff Mahoney <[email protected]>

---

 fs/reiserfs/bitmap.c |   17 +++++++++++++++--
 fs/reiserfs/inode.c  |   19 +++++++++++--------
 fs/reiserfs/stree.c  |   28 +++++++++++++++++++++++-----
 fs/reiserfs/super.c  |   22 ++++++++++++++--------
 4 files changed, 63 insertions(+), 23 deletions(-)

fs/reiserfs/bitmap.c +15 -2

@@ -423,8 +423,11 @@ static void _reiserfs_free_block(struct
 	set_sb_free_blocks(rs, sb_free_blocks(rs) + 1);
 
 	journal_mark_dirty(th, s, sbh);
-	if (for_unformatted)
+	if (for_unformatted) {
+		int depth = reiserfs_write_unlock_nested(s);
 		dquot_free_block_nodirty(inode, 1);
+		reiserfs_write_lock_nested(s, depth);
+	}
 }
 
 void reiserfs_free_block(struct reiserfs_transaction_handle *th,

@@ -1128,6 +1131,7 @@ static inline int blocknrs_and_prealloc_
 	b_blocknr_t finish = SB_BLOCK_COUNT(s) - 1;
 	int passno = 0;
 	int nr_allocated = 0;
+	int depth;
 
 	determine_prealloc_size(hint);
 	if (!hint->formatted_node) {

@@ -1137,10 +1141,13 @@ static inline int blocknrs_and_prealloc_
 			       "reiserquota: allocating %d blocks id=%u",
 			       amount_needed, hint->inode->i_uid);
 #endif
+		depth = reiserfs_write_unlock_nested(s);
 		quota_ret =
 		    dquot_alloc_block_nodirty(hint->inode, amount_needed);
-		if (quota_ret)	/* Quota exceeded? */
+		if (quota_ret) {	/* Quota exceeded? */
+			reiserfs_write_lock_nested(s, depth);
 			return QUOTA_EXCEEDED;
+		}
 		if (hint->preallocate && hint->prealloc_size) {
 #ifdef REISERQUOTA_DEBUG
 			reiserfs_debug(s, REISERFS_DEBUG_CODE,

@@ -1153,6 +1160,7 @@ static inline int blocknrs_and_prealloc_
 				hint->preallocate = hint->prealloc_size = 0;
 		}
 		/* for unformatted nodes, force large allocations */
+		reiserfs_write_lock_nested(s, depth);
 	}
 
 	do {

@@ -1181,9 +1189,11 @@ static inline int blocknrs_and_prealloc_
 					       hint->inode->i_uid);
 #endif
 				/* Free not allocated blocks */
+				depth = reiserfs_write_unlock_nested(s);
 				dquot_free_block_nodirty(hint->inode,
 					amount_needed + hint->prealloc_size -
 					nr_allocated);
+				reiserfs_write_lock_nested(s, depth);
 			}
 			while (nr_allocated--)
 				reiserfs_free_block(hint->th, hint->inode,

@@ -1214,10 +1224,13 @@ static inline int blocknrs_and_prealloc_
 			       REISERFS_I(hint->inode)->i_prealloc_count,
 			       hint->inode->i_uid);
 #endif
+
+		depth = reiserfs_write_unlock_nested(s);
 		dquot_free_block_nodirty(hint->inode, amount_needed +
 					 hint->prealloc_size - nr_allocated -
 					 REISERFS_I(hint->inode)->
 					 i_prealloc_count);
+		reiserfs_write_lock_nested(s, depth);
 	}
 
 	return CARRY_ON;

fs/reiserfs/inode.c +11 -8

@@ -57,8 +57,11 @@ void reiserfs_evict_inode(struct inode *
 		/* Do quota update inside a transaction for journaled quotas. We must do that
 		 * after delete_object so that quota updates go into the same transaction as
 		 * stat data deletion */
-		if (!err) 
+		if (!err) {
+			int depth = reiserfs_write_unlock_nested(inode->i_sb);
 			dquot_free_inode(inode);
+			reiserfs_write_lock_nested(inode->i_sb, depth);
+		}
 
 		if (journal_end(&th, inode->i_sb, jbegin_count))
 			goto out;

@@ -1768,7 +1771,7 @@ int reiserfs_new_inode(struct reiserfs_t
 		       struct inode *inode,
 		       struct reiserfs_security_handle *security)
 {
-	struct super_block *sb;
+	struct super_block *sb = dir->i_sb;
 	struct reiserfs_iget_args args;
 	INITIALIZE_PATH(path_to_key);
 	struct cpu_key key;

@@ -1780,9 +1783,9 @@ int reiserfs_new_inode(struct reiserfs_t
 
 	BUG_ON(!th->t_trans_id);
 
-	reiserfs_write_unlock(inode->i_sb);
+	depth = reiserfs_write_unlock_nested(sb);
 	err = dquot_alloc_inode(inode);
-	reiserfs_write_lock(inode->i_sb);
+	reiserfs_write_lock_nested(sb, depth);
 	if (err)
 		goto out_end_trans;
 	if (!dir->i_nlink) {

@@ -1790,8 +1793,6 @@ int reiserfs_new_inode(struct reiserfs_t
 		goto out_bad_inode;
 	}
 
-	sb = dir->i_sb;
-
 	/* item head of new item */
 	ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
 	ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));

@@ -1983,14 +1984,16 @@ int reiserfs_new_inode(struct reiserfs_t
 	INODE_PKEY(inode)->k_objectid = 0;
 
 	/* Quota change must be inside a transaction for journaling */
+	depth = reiserfs_write_unlock_nested(inode->i_sb);
 	dquot_free_inode(inode);
+	reiserfs_write_lock_nested(inode->i_sb, depth);
 
       out_end_trans:
 	journal_end(th, th->t_super, th->t_blocks_allocated);
-	reiserfs_write_unlock(inode->i_sb);
 	/* Drop can be outside and it needs more credits so it's better to have it outside */
+	depth = reiserfs_write_unlock_nested(inode->i_sb);
 	dquot_drop(inode);
-	reiserfs_write_lock(inode->i_sb);
+	reiserfs_write_lock_nested(inode->i_sb, depth);
 	inode->i_flags |= S_NOQUOTA;
 	make_bad_inode(inode);

fs/reiserfs/stree.c +23 -5

@@ -1186,6 +1186,7 @@ int reiserfs_delete_item(struct reiserfs
 	struct item_head *q_ih;
 	int quota_cut_bytes;
 	int ret_value, del_size, removed;
+	int depth;
 
 #ifdef CONFIG_REISERFS_CHECK
 	char mode;

@@ -1295,7 +1296,9 @@ int reiserfs_delete_item(struct reiserfs
 		       "reiserquota delete_item(): freeing %u, id=%u type=%c",
 		       quota_cut_bytes, inode->i_uid, head2type(&s_ih));
 #endif
+	depth = reiserfs_write_unlock_nested(inode->i_sb);
 	dquot_free_space_nodirty(inode, quota_cut_bytes);
+	reiserfs_write_lock_nested(inode->i_sb, depth);
 
 	/* Return deleted body length */
 	return ret_value;

@@ -1321,6 +1324,7 @@ int reiserfs_delete_item(struct reiserfs
 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
 				struct inode *inode, struct reiserfs_key *key)
 {
+	struct super_block *sb = th->t_super;
 	struct tree_balance tb;
 	INITIALIZE_PATH(path);
 	int item_len = 0;

@@ -1373,14 +1377,17 @@ void reiserfs_delete_solid_item(struct r
 		if (retval == CARRY_ON) {
 			do_balance(&tb, NULL, NULL, M_DELETE);
 			if (inode) {	/* Should we count quota for item? (we don't count quotas for save-links) */
+				int depth;
 #ifdef REISERQUOTA_DEBUG
 				reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
 					       "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
 					       quota_cut_bytes, inode->i_uid,
 					       key2type(key));
 #endif
+				depth = reiserfs_write_unlock_nested(sb);
 				dquot_free_space_nodirty(inode,
 							 quota_cut_bytes);
+				reiserfs_write_lock_nested(sb, depth);
 			}
 			break;
 		}

@@ -1557,6 +1564,7 @@ int reiserfs_cut_from_item(struct reiser
 	int retval2 = -1;
 	int quota_cut_bytes;
 	loff_t tail_pos = 0;
+	int depth;
 
 	BUG_ON(!th->t_trans_id);

@@ -1729,7 +1737,9 @@ int reiserfs_cut_from_item(struct reiser
 		       "reiserquota cut_from_item(): freeing %u id=%u type=%c",
 		       quota_cut_bytes, inode->i_uid, '?');
 #endif
+	depth = reiserfs_write_unlock_nested(sb);
 	dquot_free_space_nodirty(inode, quota_cut_bytes);
+	reiserfs_write_lock_nested(sb, depth);
 	return ret_value;
 }

@@ -1949,9 +1959,11 @@ int reiserfs_paste_into_item(struct reis
 			     const char *body,	/* Pointer to the bytes to paste.    */
 			     int pasted_size)
 {				/* Size of pasted bytes.             */
+	struct super_block *sb = inode->i_sb;
 	struct tree_balance s_paste_balance;
 	int retval;
 	int fs_gen;
+	int depth;
 
 	BUG_ON(!th->t_trans_id);

@@ -1964,9 +1976,9 @@ int reiserfs_paste_into_item(struct reis
 		       key2type(&(key->on_disk_key)));
 #endif
 
-	reiserfs_write_unlock(inode->i_sb);
+	depth = reiserfs_write_unlock_nested(sb);
 	retval = dquot_alloc_space_nodirty(inode, pasted_size);
-	reiserfs_write_lock(inode->i_sb);
+	reiserfs_write_lock_nested(sb, depth);
 	if (retval) {
 		pathrelse(search_path);
 		return retval;

@@ -2023,7 +2035,9 @@ int reiserfs_paste_into_item(struct reis
 		       pasted_size, inode->i_uid,
 		       key2type(&(key->on_disk_key)));
 #endif
+	depth = reiserfs_write_unlock_nested(sb);
 	dquot_free_space_nodirty(inode, pasted_size);
+	reiserfs_write_lock_nested(sb, depth);
 	return retval;
 }

@@ -2046,6 +2060,7 @@ int reiserfs_insert_item(struct reiserfs
 	BUG_ON(!th->t_trans_id);
 
 	if (inode) {		/* Do we count quotas for item? */
+		int depth;
 		fs_gen = get_generation(inode->i_sb);
 		quota_bytes = ih_item_len(ih);

@@ -2059,11 +2074,11 @@ int reiserfs_insert_item(struct reiserfs
 			       "reiserquota insert_item(): allocating %u id=%u type=%c",
 			       quota_bytes, inode->i_uid, head2type(ih));
 #endif
-		reiserfs_write_unlock(inode->i_sb);
 		/* We can't dirty inode here. It would be immediately written but
 		 * appropriate stat item isn't inserted yet... */
+		depth = reiserfs_write_unlock_nested(inode->i_sb);
 		retval = dquot_alloc_space_nodirty(inode, quota_bytes);
-		reiserfs_write_lock(inode->i_sb);
+		reiserfs_write_lock_nested(inode->i_sb, depth);
 		if (retval) {
 			pathrelse(path);
 			return retval;

@@ -2114,7 +2129,10 @@ int reiserfs_insert_item(struct reiserfs
 		       "reiserquota insert_item(): freeing %u id=%u type=%c",
 		       quota_bytes, inode->i_uid, head2type(ih));
 #endif
-	if (inode)
+	if (inode) {
+		int depth = reiserfs_write_unlock_nested(inode->i_sb);
 		dquot_free_space_nodirty(inode, quota_bytes);
+		reiserfs_write_lock_nested(inode->i_sb, depth);
+	}
 	return retval;
 }

fs/reiserfs/super.c +14 -8

@@ -243,6 +243,7 @@ static int finish_unfinished(struct supe
 	done = 0;
 	REISERFS_SB(s)->s_is_unlinked_ok = 1;
 	while (!retval) {
+		int depth;
 		retval = search_item(s, &max_cpu_key, &path);
 		if (retval != ITEM_NOT_FOUND) {
 			reiserfs_error(s, "vs-2140",

@@ -298,9 +299,9 @@ static int finish_unfinished(struct supe
 			retval = remove_save_link_only(s, &save_link_key, 0);
 			continue;
 		}
-		reiserfs_write_unlock(s);
+		depth = reiserfs_write_unlock_nested(inode->i_sb);
 		dquot_initialize(inode);
-		reiserfs_write_lock(s);
+		reiserfs_write_lock_nested(inode->i_sb, depth);
 
 		if (truncate && S_ISDIR(inode->i_mode)) {
 			/* We got a truncate request for a dir which is impossible.

@@ -356,10 +357,12 @@ static int finish_unfinished(struct supe
 
 #ifdef CONFIG_QUOTA
 	/* Turn quotas off */
+	reiserfs_write_unlock(s);
 	for (i = 0; i < MAXQUOTAS; i++) {
 		if (sb_dqopt(s)->files[i] && quota_enabled[i])
 			dquot_quota_off(s, i);
 	}
+	reiserfs_write_lock(s);
 	if (ms_active_set)
 		/* Restore the flag back */
 		s->s_flags &= ~MS_ACTIVE;

@@ -2098,6 +2101,7 @@ static int reiserfs_write_dquot(struct d
 {
 	struct reiserfs_transaction_handle th;
 	int ret, err;
+	int depth;
 
 	reiserfs_write_lock(dquot->dq_sb);
 	ret =

@@ -2105,9 +2109,9 @@ static int reiserfs_write_dquot(struct d
 			  REISERFS_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
 	if (ret)
 		goto out;
-	reiserfs_write_unlock(dquot->dq_sb);
+	depth = reiserfs_write_unlock_nested(dquot->dq_sb);
 	ret = dquot_commit(dquot);
-	reiserfs_write_lock(dquot->dq_sb);
+	reiserfs_write_lock_nested(dquot->dq_sb, depth);
 	err =
 	    journal_end(&th, dquot->dq_sb,
 			REISERFS_QUOTA_TRANS_BLOCKS(dquot->dq_sb));

@@ -2122,6 +2126,7 @@ static int reiserfs_acquire_dquot(struct
 {
 	struct reiserfs_transaction_handle th;
 	int ret, err;
+	int depth;
 
 	reiserfs_write_lock(dquot->dq_sb);
 	ret =

@@ -2129,9 +2134,9 @@ static int reiserfs_acquire_dquot(struct
 			  REISERFS_QUOTA_INIT_BLOCKS(dquot->dq_sb));
 	if (ret)
 		goto out;
-	reiserfs_write_unlock(dquot->dq_sb);
+	depth = reiserfs_write_unlock_nested(dquot->dq_sb);
 	ret = dquot_acquire(dquot);
-	reiserfs_write_lock(dquot->dq_sb);
+	reiserfs_write_lock_nested(dquot->dq_sb, depth);
 	err =
 	    journal_end(&th, dquot->dq_sb,
 			REISERFS_QUOTA_INIT_BLOCKS(dquot->dq_sb));

@@ -2184,15 +2189,16 @@ static int reiserfs_write_info(struct su
 {
 	struct reiserfs_transaction_handle th;
 	int ret, err;
+	int depth;
 
 	/* Data block + inode block */
 	reiserfs_write_lock(sb);
 	ret = journal_begin(&th, sb, 2);
 	if (ret)
 		goto out;
-	reiserfs_write_unlock(sb);
+	depth = reiserfs_write_unlock_nested(sb);
 	ret = dquot_commit_info(sb, type);
-	reiserfs_write_lock(sb);
+	reiserfs_write_lock_nested(sb, depth);
 	err = journal_end(&th, sb, 2);
 	if (!ret && err)
 		ret = err;

Jan Kara

2013-08-09 21:31 UTC #4 lore raw

In reply to Jeff Mahoney

Quoted 18 lines · Jeff Mahoney

This patchset untangles some of the locking in reiserfs. It has seen more testing as part of the maintenance cycle in SLE 11 SP2.
- We push the write lock out of the xattr code. It doesn't need it and we can simplify locking by releasing and reacquiring the locks around the xattr calls.
- Handle nested locks properly. There's confusion on when a lock is nested, when it's not, and how to drop it across schedules like the BKL it is modeled after. We make the distinction between taking/releasing the lock and when to drop it for schedules and simplify the logic. This fixes a number of deadlocks that happen because the intention was to drop the write lock but it really only decremented the use count.
- Fix the deadlocks with the quota code. This involves dropping the write lock before quota calls and reacquring it afterwards. Without this patch reiserfs quotas are essentially unusable.

Quoted 7 lines

Previous postings of this patchset were mismerged.
This revision integrates Jan's advice about search_by_key_reada returning the depth instead of a bool.
Jan - This series is in the for-3.12 branch of git.kernel.org:/pub/scm/linux/kernel/git/jeffm/linux-reiserfs.git

Thanks! I've pulled this series to my tree.

Honza

Signature · 2 lines

Jan Kara <[email protected]>
SUSE Labs, CR