do d_instantiate/unlock_new_inode combinations safely
authorAl Viro <viro@zeniv.linux.org.uk>
Fri, 4 May 2018 12:23:01 +0000 (08:23 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 30 May 2018 05:51:47 +0000 (07:51 +0200)
commit 1e2e547a93a00ebc21582c06ca3c6cfea2a309ee upstream.

For anything NFS-exported we do _not_ want to unlock new inode
before it has grown an alias; original set of fixes got the
ordering right, but missed the nasty complication in case of
lockdep being enabled - unlock_new_inode() does
lockdep_annotate_inode_mutex_key(inode)
which can only be done before anyone gets a chance to touch
->i_mutex.  Unfortunately, flipping the order and doing
unlock_new_inode() before d_instantiate() opens a window when
mkdir can race with open-by-fhandle on a guessed fhandle, leading
to multiple aliases for a directory inode and all the breakage
that follows from that.

Correct solution: a new primitive (d_instantiate_new())
combining these two in the right order - lockdep annotate, then
d_instantiate(), then the rest of unlock_new_inode().  All
combinations of d_instantiate() with unlock_new_inode() should
be converted to that.

Cc: stable@kernel.org # 2.6.29 and later
Tested-by: Mike Marshall <hubcap@omnibond.com>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
14 files changed:
fs/btrfs/inode.c
fs/dcache.c
fs/ecryptfs/inode.c
fs/ext2/namei.c
fs/ext4/namei.c
fs/f2fs/namei.c
fs/jffs2/dir.c
fs/jfs/namei.c
fs/nilfs2/namei.c
fs/orangefs/namei.c
fs/reiserfs/namei.c
fs/udf/namei.c
fs/ufs/namei.c
include/linux/dcache.h

index b0fa3a0..8ecbac3 100644 (file)
@@ -6664,8 +6664,7 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
                goto out_unlock_inode;
        } else {
                btrfs_update_inode(trans, root, inode);
-               unlock_new_inode(inode);
-               d_instantiate(dentry, inode);
+               d_instantiate_new(dentry, inode);
        }
 
 out_unlock:
@@ -6742,8 +6741,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
                goto out_unlock_inode;
 
        BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
 
 out_unlock:
        btrfs_end_transaction(trans);
@@ -6890,12 +6888,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
        if (err)
                goto out_fail_inode;
 
-       d_instantiate(dentry, inode);
-       /*
-        * mkdir is special.  We're unlocking after we call d_instantiate
-        * to avoid a race with nfsd calling d_instantiate.
-        */
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
        drop_on_err = 0;
 
 out_fail:
@@ -10573,8 +10566,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
                goto out_unlock_inode;
        }
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
 
 out_unlock:
        btrfs_end_transaction(trans);
index c28b9c9..d21e811 100644 (file)
@@ -1867,6 +1867,28 @@ void d_instantiate(struct dentry *entry, struct inode * inode)
 }
 EXPORT_SYMBOL(d_instantiate);
 
+/*
+ * This should be equivalent to d_instantiate() + unlock_new_inode(),
+ * with lockdep-related part of unlock_new_inode() done before
+ * anything else.  Use that instead of open-coding d_instantiate()/
+ * unlock_new_inode() combinations.
+ */
+void d_instantiate_new(struct dentry *entry, struct inode *inode)
+{
+       BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
+       BUG_ON(!inode);
+       lockdep_annotate_inode_mutex_key(inode);
+       security_d_instantiate(entry, inode);
+       spin_lock(&inode->i_lock);
+       __d_instantiate(entry, inode);
+       WARN_ON(!(inode->i_state & I_NEW));
+       inode->i_state &= ~I_NEW;
+       smp_mb();
+       wake_up_bit(&inode->i_state, __I_NEW);
+       spin_unlock(&inode->i_lock);
+}
+EXPORT_SYMBOL(d_instantiate_new);
+
 /**
  * d_instantiate_no_diralias - instantiate a non-aliased dentry
  * @entry: dentry to complete
index efc2db4..bda65a7 100644 (file)
@@ -284,8 +284,7 @@ ecryptfs_create(struct inode *directory_inode, struct dentry *ecryptfs_dentry,
                iget_failed(ecryptfs_inode);
                goto out;
        }
-       unlock_new_inode(ecryptfs_inode);
-       d_instantiate(ecryptfs_dentry, ecryptfs_inode);
+       d_instantiate_new(ecryptfs_dentry, ecryptfs_inode);
 out:
        return rc;
 }
index e078075..aa6ec19 100644 (file)
@@ -41,8 +41,7 @@ static inline int ext2_add_nondir(struct dentry *dentry, struct inode *inode)
 {
        int err = ext2_add_link(dentry, inode);
        if (!err) {
-               unlock_new_inode(inode);
-               d_instantiate(dentry, inode);
+               d_instantiate_new(dentry, inode);
                return 0;
        }
        inode_dec_link_count(inode);
@@ -269,8 +268,7 @@ static int ext2_mkdir(struct inode * dir, struct dentry * dentry, umode_t mode)
        if (err)
                goto out_fail;
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
 out:
        return err;
 
index fccf295..6747861 100644 (file)
@@ -2420,8 +2420,7 @@ static int ext4_add_nondir(handle_t *handle,
        int err = ext4_add_entry(handle, dentry, inode);
        if (!err) {
                ext4_mark_inode_dirty(handle, inode);
-               unlock_new_inode(inode);
-               d_instantiate(dentry, inode);
+               d_instantiate_new(dentry, inode);
                return 0;
        }
        drop_nlink(inode);
@@ -2660,8 +2659,7 @@ out_clear_inode:
        err = ext4_mark_inode_dirty(handle, dir);
        if (err)
                goto out_clear_inode;
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        if (IS_DIRSYNC(dir))
                ext4_handle_sync(handle);
 
index a4dab98..b80e7db 100644 (file)
@@ -201,8 +201,7 @@ static int f2fs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 
        alloc_nid_done(sbi, ino);
 
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
 
        if (IS_DIRSYNC(dir))
                f2fs_sync_fs(sbi->sb, 1);
@@ -529,8 +528,7 @@ static int f2fs_symlink(struct inode *dir, struct dentry *dentry,
        err = page_symlink(inode, disk_link.name, disk_link.len);
 
 err_out:
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
 
        /*
         * Let's flush symlink data in order to avoid broken symlink as much as
@@ -588,8 +586,7 @@ static int f2fs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 
        alloc_nid_done(sbi, inode->i_ino);
 
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
 
        if (IS_DIRSYNC(dir))
                f2fs_sync_fs(sbi->sb, 1);
@@ -637,8 +634,7 @@ static int f2fs_mknod(struct inode *dir, struct dentry *dentry,
 
        alloc_nid_done(sbi, inode->i_ino);
 
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
 
        if (IS_DIRSYNC(dir))
                f2fs_sync_fs(sbi->sb, 1);
index 0a754f3..e5a6deb 100644 (file)
@@ -209,8 +209,7 @@ static int jffs2_create(struct inode *dir_i, struct dentry *dentry,
                  __func__, inode->i_ino, inode->i_mode, inode->i_nlink,
                  f->inocache->pino_nlink, inode->i_mapping->nrpages);
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        return 0;
 
  fail:
@@ -430,8 +429,7 @@ static int jffs2_symlink (struct inode *dir_i, struct dentry *dentry, const char
        mutex_unlock(&dir_f->sem);
        jffs2_complete_reservation(c);
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        return 0;
 
  fail:
@@ -575,8 +573,7 @@ static int jffs2_mkdir (struct inode *dir_i, struct dentry *dentry, umode_t mode
        mutex_unlock(&dir_f->sem);
        jffs2_complete_reservation(c);
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        return 0;
 
  fail:
@@ -747,8 +744,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, umode_t mode
        mutex_unlock(&dir_f->sem);
        jffs2_complete_reservation(c);
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        return 0;
 
  fail:
index b41596d..56c3fcb 100644 (file)
@@ -178,8 +178,7 @@ static int jfs_create(struct inode *dip, struct dentry *dentry, umode_t mode,
                unlock_new_inode(ip);
                iput(ip);
        } else {
-               unlock_new_inode(ip);
-               d_instantiate(dentry, ip);
+               d_instantiate_new(dentry, ip);
        }
 
       out2:
@@ -313,8 +312,7 @@ static int jfs_mkdir(struct inode *dip, struct dentry *dentry, umode_t mode)
                unlock_new_inode(ip);
                iput(ip);
        } else {
-               unlock_new_inode(ip);
-               d_instantiate(dentry, ip);
+               d_instantiate_new(dentry, ip);
        }
 
       out2:
@@ -1059,8 +1057,7 @@ static int jfs_symlink(struct inode *dip, struct dentry *dentry,
                unlock_new_inode(ip);
                iput(ip);
        } else {
-               unlock_new_inode(ip);
-               d_instantiate(dentry, ip);
+               d_instantiate_new(dentry, ip);
        }
 
       out2:
@@ -1447,8 +1444,7 @@ static int jfs_mknod(struct inode *dir, struct dentry *dentry,
                unlock_new_inode(ip);
                iput(ip);
        } else {
-               unlock_new_inode(ip);
-               d_instantiate(dentry, ip);
+               d_instantiate_new(dentry, ip);
        }
 
       out1:
index 515d13c..1ba4719 100644 (file)
@@ -46,8 +46,7 @@ static inline int nilfs_add_nondir(struct dentry *dentry, struct inode *inode)
        int err = nilfs_add_link(dentry, inode);
 
        if (!err) {
-               d_instantiate(dentry, inode);
-               unlock_new_inode(inode);
+               d_instantiate_new(dentry, inode);
                return 0;
        }
        inode_dec_link_count(inode);
@@ -243,8 +242,7 @@ static int nilfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
                goto out_fail;
 
        nilfs_mark_inode_dirty(inode);
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
 out:
        if (!err)
                err = nilfs_transaction_commit(dir->i_sb);
index 7e9e5d0..f8f3c73 100644 (file)
@@ -71,8 +71,7 @@ static int orangefs_create(struct inode *dir,
                     get_khandle_from_ino(inode),
                     dentry);
 
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
        orangefs_set_timeout(dentry);
        ORANGEFS_I(inode)->getattr_time = jiffies - 1;
        ORANGEFS_I(inode)->getattr_mask = STATX_BASIC_STATS;
@@ -320,8 +319,7 @@ static int orangefs_symlink(struct inode *dir,
                     "Assigned symlink inode new number of %pU\n",
                     get_khandle_from_ino(inode));
 
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
        orangefs_set_timeout(dentry);
        ORANGEFS_I(inode)->getattr_time = jiffies - 1;
        ORANGEFS_I(inode)->getattr_mask = STATX_BASIC_STATS;
@@ -385,8 +383,7 @@ static int orangefs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode
                     "Assigned dir inode new number of %pU\n",
                     get_khandle_from_ino(inode));
 
-       d_instantiate(dentry, inode);
-       unlock_new_inode(inode);
+       d_instantiate_new(dentry, inode);
        orangefs_set_timeout(dentry);
        ORANGEFS_I(inode)->getattr_time = jiffies - 1;
        ORANGEFS_I(inode)->getattr_mask = STATX_BASIC_STATS;
index bd39a99..5089dac 100644 (file)
@@ -687,8 +687,7 @@ static int reiserfs_create(struct inode *dir, struct dentry *dentry, umode_t mod
        reiserfs_update_inode_transaction(inode);
        reiserfs_update_inode_transaction(dir);
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        retval = journal_end(&th);
 
 out_failed:
@@ -771,8 +770,7 @@ static int reiserfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode
                goto out_failed;
        }
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        retval = journal_end(&th);
 
 out_failed:
@@ -871,8 +869,7 @@ static int reiserfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode
        /* the above add_entry did not update dir's stat data */
        reiserfs_update_sd(&th, dir);
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        retval = journal_end(&th);
 out_failed:
        reiserfs_write_unlock(dir->i_sb);
@@ -1187,8 +1184,7 @@ static int reiserfs_symlink(struct inode *parent_dir,
                goto out_failed;
        }
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        retval = journal_end(&th);
 out_failed:
        reiserfs_write_unlock(parent_dir->i_sb);
index 885198d..041bf34 100644 (file)
@@ -621,8 +621,7 @@ static int udf_add_nondir(struct dentry *dentry, struct inode *inode)
        if (fibh.sbh != fibh.ebh)
                brelse(fibh.ebh);
        brelse(fibh.sbh);
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
 
        return 0;
 }
@@ -732,8 +731,7 @@ static int udf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
        inc_nlink(dir);
        dir->i_ctime = dir->i_mtime = current_time(dir);
        mark_inode_dirty(dir);
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        if (fibh.sbh != fibh.ebh)
                brelse(fibh.ebh);
        brelse(fibh.sbh);
index 32545cd..d5f43ba 100644 (file)
@@ -39,8 +39,7 @@ static inline int ufs_add_nondir(struct dentry *dentry, struct inode *inode)
 {
        int err = ufs_add_link(dentry, inode);
        if (!err) {
-               unlock_new_inode(inode);
-               d_instantiate(dentry, inode);
+               d_instantiate_new(dentry, inode);
                return 0;
        }
        inode_dec_link_count(inode);
@@ -193,8 +192,7 @@ static int ufs_mkdir(struct inode * dir, struct dentry * dentry, umode_t mode)
        if (err)
                goto out_fail;
 
-       unlock_new_inode(inode);
-       d_instantiate(dentry, inode);
+       d_instantiate_new(dentry, inode);
        return 0;
 
 out_fail:
index f05a659..006f4cc 100644 (file)
@@ -226,6 +226,7 @@ extern seqlock_t rename_lock;
  * These are the low-level FS interfaces to the dcache..
  */
 extern void d_instantiate(struct dentry *, struct inode *);
+extern void d_instantiate_new(struct dentry *, struct inode *);
 extern struct dentry * d_instantiate_unique(struct dentry *, struct inode *);
 extern int d_instantiate_no_diralias(struct dentry *, struct inode *);
 extern void __d_drop(struct dentry *dentry);