]> Git Repo - linux.git/commitdiff
Revert "fs: Do not check if there is a fsnotify watcher on pseudo inodes"
authorMel Gorman <[email protected]>
Mon, 29 Jun 2020 14:41:45 +0000 (15:41 +0100)
committerLinus Torvalds <[email protected]>
Mon, 29 Jun 2020 16:40:55 +0000 (09:40 -0700)
This reverts commit e9c15badbb7b ("fs: Do not check if there is a
fsnotify watcher on pseudo inodes"). The commit intended to eliminate
fsnotify-related overhead for pseudo inodes but it is broken in
concept. inotify can receive events of pipe files under /proc/X/fd and
chromium relies on close and open events for sandboxing. Maxim Levitsky
reported the following

  Chromium starts as a white rectangle, shows few white rectangles that
  resemble its notifications and then crashes.

  The stdout output from chromium:

  [mlevitsk@starship ~]$chromium-freeworld
  mesa: for the   --simplifycfg-sink-common option: may only occur zero or one times!
  mesa: for the   --global-isel-abort option: may only occur zero or one times!
  [3379:3379:0628/135151.440930:ERROR:browser_switcher_service.cc(238)] XXX Init()
  ../../sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.cc:**CRASHING**:seccomp-bpf failure in syscall 0072
  Received signal 11 SEGV_MAPERR 0000004a9048

Crashes are not universal but even if chromium does not crash, it certainly
does not work properly. While filtering just modify and access might be
safe, the benefit is not worth the risk hence the revert.

Reported-by: Maxim Levitsky <[email protected]>
Fixes: e9c15badbb7b ("fs: Do not check if there is a fsnotify watcher on pseudo inodes")
Signed-off-by: Mel Gorman <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
fs/file_table.c

index 65603502fed6f4dc6106b02bbbc7735d3fd2678d..656647f9575a7cef916beda0081a03f9d0ecdc86 100644 (file)
@@ -230,7 +230,7 @@ struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
                d_set_d_op(path.dentry, &anon_ops);
        path.mnt = mntget(mnt);
        d_instantiate(path.dentry, inode);
-       file = alloc_file(&path, flags | FMODE_NONOTIFY, fops);
+       file = alloc_file(&path, flags, fops);
        if (IS_ERR(file)) {
                ihold(inode);
                path_put(&path);
This page took 0.042405 seconds and 4 git commands to generate.