]> git.baikalelectronics.ru Git - kernel.git/commitdiff
fsnotify: invalidate dcache before IN_DELETE event
authorAmir Goldstein <amir73il@gmail.com>
Thu, 20 Jan 2022 21:53:04 +0000 (23:53 +0200)
committerJan Kara <jack@suse.cz>
Mon, 24 Jan 2022 13:16:46 +0000 (14:16 +0100)
Apparently, there are some applications that use IN_DELETE event as an
invalidation mechanism and expect that if they try to open a file with
the name reported with the delete event, that it should not contain the
content of the deleted file.

Commit c2262d9f1121 ("fsnotify: move fsnotify_nameremove() hook out of
d_delete()") moved the fsnotify delete hook before d_delete() so fsnotify
will have access to a positive dentry.

This allowed a race where opening the deleted file via cached dentry
is now possible after receiving the IN_DELETE event.

To fix the regression, create a new hook fsnotify_delete() that takes
the unlinked inode as an argument and use a helper d_delete_notify() to
pin the inode, so we can pass it to fsnotify_delete() after d_delete().

Backporting hint: this regression is from v5.3. Although patch will
apply with only trivial conflicts to v5.4 and v5.10, it won't build,
because fsnotify_delete() implementation is different in each of those
versions (see fsnotify_link()).

A follow up patch will fix the fsnotify_unlink/rmdir() calls in pseudo
filesystem that do not need to call d_delete().

Link: https://lore.kernel.org/r/20220120215305.282577-1-amir73il@gmail.com
Reported-by: Ivan Delalande <colona@arista.com>
Link: https://lore.kernel.org/linux-fsdevel/YeNyzoDM5hP5LtGW@visor/
Fixes: c2262d9f1121 ("fsnotify: move fsnotify_nameremove() hook out of d_delete()")
Cc: stable@vger.kernel.org # v5.3+
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Jan Kara <jack@suse.cz>
fs/btrfs/ioctl.c
fs/namei.c
include/linux/fsnotify.h

index a5bd6926f7ffe9741eeddcbd5ce246019728423e..7807b28b78923755e3b8af371f16da6aee4d8cdd 100644 (file)
@@ -3086,10 +3086,8 @@ static noinline int btrfs_ioctl_snap_destroy(struct file *file,
        btrfs_inode_lock(inode, 0);
        err = btrfs_delete_subvolume(dir, dentry);
        btrfs_inode_unlock(inode, 0);
-       if (!err) {
-               fsnotify_rmdir(dir, dentry);
-               d_delete(dentry);
-       }
+       if (!err)
+               d_delete_notify(dir, dentry);
 
 out_dput:
        dput(dentry);
index d81f04f8d81880b84d34598948126a63e33319f2..4ed0e41feab7c1285dd8807a5049f5e1d9f41c36 100644 (file)
@@ -3974,13 +3974,12 @@ int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
        dentry->d_inode->i_flags |= S_DEAD;
        dont_mount(dentry);
        detach_mounts(dentry);
-       fsnotify_rmdir(dir, dentry);
 
 out:
        inode_unlock(dentry->d_inode);
        dput(dentry);
        if (!error)
-               d_delete(dentry);
+               d_delete_notify(dir, dentry);
        return error;
 }
 EXPORT_SYMBOL(vfs_rmdir);
@@ -4102,7 +4101,6 @@ int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
                        if (!error) {
                                dont_mount(dentry);
                                detach_mounts(dentry);
-                               fsnotify_unlink(dir, dentry);
                        }
                }
        }
@@ -4110,9 +4108,11 @@ out:
        inode_unlock(target);
 
        /* We don't d_delete() NFS sillyrenamed files--they still exist. */
-       if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
+       if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
+               fsnotify_unlink(dir, dentry);
+       } else if (!error) {
                fsnotify_link_count(target);
-               d_delete(dentry);
+               d_delete_notify(dir, dentry);
        }
 
        return error;
index 3a2d7dc3c6071a424d81305486a146f155c875b1..bb8467cd11ae28eb795db5eb17d4b78774da61ca 100644 (file)
@@ -224,6 +224,43 @@ static inline void fsnotify_link(struct inode *dir, struct inode *inode,
                      dir, &new_dentry->d_name, 0);
 }
 
+/*
+ * fsnotify_delete - @dentry was unlinked and unhashed
+ *
+ * Caller must make sure that dentry->d_name is stable.
+ *
+ * Note: unlike fsnotify_unlink(), we have to pass also the unlinked inode
+ * as this may be called after d_delete() and old_dentry may be negative.
+ */
+static inline void fsnotify_delete(struct inode *dir, struct inode *inode,
+                                  struct dentry *dentry)
+{
+       __u32 mask = FS_DELETE;
+
+       if (S_ISDIR(inode->i_mode))
+               mask |= FS_ISDIR;
+
+       fsnotify_name(mask, inode, FSNOTIFY_EVENT_INODE, dir, &dentry->d_name,
+                     0);
+}
+
+/**
+ * d_delete_notify - delete a dentry and call fsnotify_delete()
+ * @dentry: The dentry to delete
+ *
+ * This helper is used to guaranty that the unlinked inode cannot be found
+ * by lookup of this name after fsnotify_delete() event has been delivered.
+ */
+static inline void d_delete_notify(struct inode *dir, struct dentry *dentry)
+{
+       struct inode *inode = d_inode(dentry);
+
+       ihold(inode);
+       d_delete(dentry);
+       fsnotify_delete(dir, inode, dentry);
+       iput(inode);
+}
+
 /*
  * fsnotify_unlink - 'name' was unlinked
  *
@@ -231,10 +268,10 @@ static inline void fsnotify_link(struct inode *dir, struct inode *inode,
  */
 static inline void fsnotify_unlink(struct inode *dir, struct dentry *dentry)
 {
-       /* Expected to be called before d_delete() */
-       WARN_ON_ONCE(d_is_negative(dentry));
+       if (WARN_ON_ONCE(d_is_negative(dentry)))
+               return;
 
-       fsnotify_dirent(dir, dentry, FS_DELETE);
+       fsnotify_delete(dir, d_inode(dentry), dentry);
 }
 
 /*
@@ -258,10 +295,10 @@ static inline void fsnotify_mkdir(struct inode *dir, struct dentry *dentry)
  */
 static inline void fsnotify_rmdir(struct inode *dir, struct dentry *dentry)
 {
-       /* Expected to be called before d_delete() */
-       WARN_ON_ONCE(d_is_negative(dentry));
+       if (WARN_ON_ONCE(d_is_negative(dentry)))
+               return;
 
-       fsnotify_dirent(dir, dentry, FS_DELETE | FS_ISDIR);
+       fsnotify_delete(dir, d_inode(dentry), dentry);
 }
 
 /*