linux-fsdevel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: NeilBrown <neilb@suse.de>
To: Al Viro <viro@ZenIV.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [PATCH 19/20] XFS: allow follow_link to often succeed in RCU-walk.
Date: Mon, 23 Mar 2015 13:37:40 +1100	[thread overview]
Message-ID: <20150323023740.8161.44210.stgit@notabene.brown> (raw)
In-Reply-To: <20150323023258.8161.32467.stgit@notabene.brown>

If LOOKUP_RCU is set, use GFP_ATOMIC rather than GFP_KERNEL,
and try to get the ilock without blocking.

When these succeed, follow_link() can succeed without dropping
out of RCU-walk.

As xfs_readlink can now races with xfs_fs_evict_inode:
- xfs_readlink must check if the inode is being evicted after
  getting the lock
- xfs_fs_evict_inode cannot assert that the lock is not held.

Signed-off-by: NeilBrown <neilb@suse.de>
---
 fs/xfs/xfs_ioctl.c   |    2 +-
 fs/xfs/xfs_iops.c    |   15 ++++++++++-----
 fs/xfs/xfs_super.c   |    2 --
 fs/xfs/xfs_symlink.c |   15 +++++++++++++--
 fs/xfs/xfs_symlink.h |    2 +-
 5 files changed, 25 insertions(+), 11 deletions(-)

diff --git a/fs/xfs/xfs_ioctl.c b/fs/xfs/xfs_ioctl.c
index ac4feae45eb3..29d95a1b76c0 100644
--- a/fs/xfs/xfs_ioctl.c
+++ b/fs/xfs/xfs_ioctl.c
@@ -303,7 +303,7 @@ xfs_readlink_by_handle(
 		goto out_dput;
 	}
 
-	error = xfs_readlink(XFS_I(dentry->d_inode), link);
+	error = xfs_readlink(XFS_I(dentry->d_inode), link, 0);
 	if (error)
 		goto out_kfree;
 	error = readlink_copy(hreq->ohandle, olen, link);
diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c
index c2c136ef3a50..631b1ec2e650 100644
--- a/fs/xfs/xfs_iops.c
+++ b/fs/xfs/xfs_iops.c
@@ -416,15 +416,20 @@ xfs_vn_follow_link(
 	int			flags)
 {
 	char			*link;
-	int			error = -ENOMEM;
+	int			error;
 
-	if (flags & LOOKUP_RCU)
-		return ERR_PTR(-ECHILD);
-	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
+	if (flags & LOOKUP_RCU) {
+		error = -ECHILD;
+		link = kmalloc(MAXPATHLEN+1, GFP_ATOMIC);
+	} else {
+		error = -ENOMEM;
+		link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
+	}
 	if (!link)
 		goto out_err;
 
-	error = xfs_readlink(XFS_I(inode), link);
+	error = xfs_readlink(XFS_I(inode), link,
+			     flags & LOOKUP_RCU);
 	if (unlikely(error))
 		goto out_kfree;
 
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index 3827be14383c..e041fa55912b 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -996,8 +996,6 @@ xfs_fs_evict_inode(
 {
 	xfs_inode_t		*ip = XFS_I(inode);
 
-	ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
-
 	trace_xfs_evict_inode(ip);
 
 	truncate_inode_pages_final(&inode->i_data);
diff --git a/fs/xfs/xfs_symlink.c b/fs/xfs/xfs_symlink.c
index 25791df6f638..228987b8e758 100644
--- a/fs/xfs/xfs_symlink.c
+++ b/fs/xfs/xfs_symlink.c
@@ -123,7 +123,8 @@ xfs_readlink_bmap(
 int
 xfs_readlink(
 	struct xfs_inode *ip,
-	char		*link)
+	char		*link,
+	int		rcu)
 {
 	struct xfs_mount *mp = ip->i_mount;
 	xfs_fsize_t	pathlen;
@@ -134,7 +135,15 @@ xfs_readlink(
 	if (XFS_FORCED_SHUTDOWN(mp))
 		return -EIO;
 
-	xfs_ilock(ip, XFS_ILOCK_SHARED);
+	if (rcu) {
+		if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED) == 0)
+			return -ECHILD;
+		if (ip->i_vnode.i_state & (I_FREEING | I_CLEAR)) {
+			xfs_iunlock(ip, XFS_ILOCK_EXCL);
+			return -ECHILD;
+		}
+	} else
+		xfs_ilock(ip, XFS_ILOCK_SHARED);
 
 	pathlen = ip->i_d.di_size;
 	if (!pathlen)
@@ -153,6 +162,8 @@ xfs_readlink(
 	if (ip->i_df.if_flags & XFS_IFINLINE) {
 		memcpy(link, ip->i_df.if_u1.if_data, pathlen);
 		link[pathlen] = '\0';
+	} else if (rcu) {
+		error = -ECHILD;
 	} else {
 		error = xfs_readlink_bmap(ip, link);
 	}
diff --git a/fs/xfs/xfs_symlink.h b/fs/xfs/xfs_symlink.h
index e75245d09116..a71d26643e20 100644
--- a/fs/xfs/xfs_symlink.h
+++ b/fs/xfs/xfs_symlink.h
@@ -21,7 +21,7 @@
 
 int xfs_symlink(struct xfs_inode *dp, struct xfs_name *link_name,
 		const char *target_path, umode_t mode, struct xfs_inode **ipp);
-int xfs_readlink(struct xfs_inode *ip, char *link);
+int xfs_readlink(struct xfs_inode *ip, char *link, int rcu);
 int xfs_inactive_symlink(struct xfs_inode *ip);
 
 #endif /* __XFS_SYMLINK_H */

  parent reply	other threads:[~2015-03-23  2:37 UTC|newest]

Thread overview: 29+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-03-23  2:37 [PATCH 00/20] Support follow_link in RCU-walk - V3 NeilBrown
2015-03-23  2:37 ` [PATCH 02/20] STAGING/lustre: limit follow_link recursion using stack space NeilBrown
2015-04-18  3:01   ` Al Viro
2015-04-19 20:57     ` Andreas Dilger
2015-04-19 21:33       ` Al Viro
2015-04-20  2:29         ` Al Viro
2015-03-23  2:37 ` [PATCH 03/20] VFS: replace {, total_}link_count in task_struct with pointer to nameidata NeilBrown
2015-03-23  2:37 ` [PATCH 01/20] Documentation: remove outdated information from automount-support.txt NeilBrown
2015-03-23  2:37 ` [PATCH 10/20] security: make inode_follow_link RCU-walk aware NeilBrown
2015-03-23  2:37 ` [PATCH 04/20] ovl: rearrange ovl_follow_link to it doesn't need to call ->put_link NeilBrown
2015-03-23  2:37 ` [PATCH 07/20] VFS: remove nameidata args from ->follow_link NeilBrown
2015-03-23  2:37 ` [PATCH 06/20] SECURITY: remove nameidata arg from inode_follow_link NeilBrown
2015-03-23  2:37 ` [PATCH 05/20] VFS: replace nameidata arg to ->put_link with a char* NeilBrown
2015-03-23  2:37 ` [PATCH 09/20] security/selinux: pass 'flags' arg to avc_audit() and avc_has_perm_flags() NeilBrown
2015-03-23  2:37 ` [PATCH 11/20] VFS/namei: use terminate_walk when symlink lookup fails NeilBrown
2015-03-23  2:37 ` [PATCH 08/20] VFS: make all ->follow_link handlers aware for LOOKUP_RCU NeilBrown
2015-03-23  2:37 ` [PATCH 13/20] VFS/namei: abort RCU-walk on symlink if atime needs updating NeilBrown
2015-03-23  2:37 ` [PATCH 14/20] VFS/namei: add 'inode' arg to put_link() NeilBrown
2015-04-17 16:25   ` Al Viro
2015-04-17 19:09     ` Al Viro
2015-04-18  8:09       ` Al Viro
2015-03-23  2:37 ` [PATCH 16/20] VFS/namei: enable RCU-walk when following symlinks NeilBrown
2015-03-23  2:37 ` NeilBrown [this message]
2015-03-23  2:37 ` [PATCH 20/20] NFS: support LOOKUP_RCU in nfs_follow_link NeilBrown
2015-03-23  2:37 ` [PATCH 18/20] xfs: use RCU to free 'struct xfs_mount' NeilBrown
2015-03-23  2:37 ` [PATCH 17/20] VFS/namei: handle LOOKUP_RCU in page_follow_link_light NeilBrown
2015-03-23  2:37 ` [PATCH 12/20] VFS/namei: new flag to support RCU symlinks: LOOKUP_LINK_RCU NeilBrown
2015-03-23  2:37 ` [PATCH 15/20] VFS/namei: enhance follow_link to support RCU-walk NeilBrown
2015-03-25 23:23 ` [PATCH 00/20] Support follow_link in RCU-walk - V3 NeilBrown

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20150323023740.8161.44210.stgit@notabene.brown \
    --to=neilb@suse.de \
    --cc=linux-fsdevel@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=viro@ZenIV.linux.org.uk \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).