queue-5.1/ovl-relax-warn_on-for-overlapping-layers-use-case.patch

   1 From acf3062a7e1ccf67c6f7e7c28671a6708fde63b0 Mon Sep 17 00:00:00 2001
   2 From: Amir Goldstein <amir73il@gmail.com>
   3 Date: Thu, 28 Mar 2019 17:38:29 +0200
   4 Subject: ovl: relax WARN_ON() for overlapping layers use case
   5
   6 From: Amir Goldstein <amir73il@gmail.com>
   7
   8 commit acf3062a7e1ccf67c6f7e7c28671a6708fde63b0 upstream.
   9
  10 This nasty little syzbot repro:
  11 https://syzkaller.appspot.com/x/repro.syz?x=12c7a94f400000
  12
  13 Creates overlay mounts where the same directory is both in upper and lower
  14 layers. Simplified example:
  15
  16   mkdir foo work
  17   mount -t overlay none foo -o"lowerdir=.,upperdir=foo,workdir=work"
  18
  19 The repro runs several threads in parallel that attempt to chdir into foo
  20 and attempt to symlink/rename/exec/mkdir the file bar.
  21
  22 The repro hits a WARN_ON() I placed in ovl_instantiate(), which suggests
  23 that an overlay inode already exists in cache and is hashed by the pointer
  24 of the real upper dentry that ovl_create_real() has just created. At the
  25 point of the WARN_ON(), for overlay dir inode lock is held and upper dir
  26 inode lock, so at first, I did not see how this was possible.
  27
  28 On a closer look, I see that after ovl_create_real(), because of the
  29 overlapping upper and lower layers, a lookup by another thread can find the
  30 file foo/bar that was just created in upper layer, at overlay path
  31 foo/foo/bar and hash the an overlay inode with the new real dentry as lower
  32 dentry. This is possible because the overlay directory foo/foo is not
  33 locked and the upper dentry foo/bar is in dcache, so ovl_lookup() can find
  34 it without taking upper dir inode shared lock.
  35
  36 Overlapping layers is considered a wrong setup which would result in
  37 unexpected behavior, but it shouldn't crash the kernel and it shouldn't
  38 trigger WARN_ON() either, so relax this WARN_ON() and leave a pr_warn()
  39 instead to cover all cases of failure to get an overlay inode.
  40
  41 The error returned from failure to insert new inode to cache with
  42 inode_insert5() was changed to -EEXIST, to distinguish from the error
  43 -ENOMEM returned on failure to get/allocate inode with iget5_locked().
  44
  45 Reported-by: syzbot+9c69c282adc4edd2b540@syzkaller.appspotmail.com
  46 Fixes: 01b39dcc9568 ("ovl: use inode_insert5() to hash a newly...")
  47 Signed-off-by: Amir Goldstein <amir73il@gmail.com>
  48 Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
  49 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  50
  51 ---
  52  fs/overlayfs/dir.c   |    2 +-
  53  fs/overlayfs/inode.c |    3 ++-
  54  2 files changed, 3 insertions(+), 2 deletions(-)
  55
  56 --- a/fs/overlayfs/dir.c
  57 +++ b/fs/overlayfs/dir.c
  58 @@ -260,7 +260,7 @@ static int ovl_instantiate(struct dentry
  59                  * hashed directory inode aliases.
  60                  */
  61                 inode = ovl_get_inode(dentry->d_sb, &oip);
  62 -               if (WARN_ON(IS_ERR(inode)))
  63 +               if (IS_ERR(inode))
  64                         return PTR_ERR(inode);
  65         } else {
  66                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
  67 --- a/fs/overlayfs/inode.c
  68 +++ b/fs/overlayfs/inode.c
  69 @@ -832,7 +832,7 @@ struct inode *ovl_get_inode(struct super
  70         int fsid = bylower ? oip->lowerpath->layer->fsid : 0;
  71         bool is_dir, metacopy = false;
  72         unsigned long ino = 0;
  73 -       int err = -ENOMEM;
  74 +       int err = oip->newinode ? -EEXIST : -ENOMEM;
  75
  76         if (!realinode)
  77                 realinode = d_inode(lowerdentry);
  78 @@ -917,6 +917,7 @@ out:
  79         return inode;
  80
  81  out_err:
  82 +       pr_warn_ratelimited("overlayfs: failed to get inode (%i)\n", err);
  83         inode = ERR_PTR(err);
  84         goto out;
  85  }