overlayfs: Warn on copy up if a process has a R/O fd open to the lower file

Commit Message

David Howells July 7, 2015, 2:04 p.m. UTC

Print a warning when overlayfs copies up a file if the process that triggered
the copy up has a R/O fd open to the lower file being copied up.

This can help catch applications that do things like the following:

	fd1 = open("foo", O_RDONLY);
	fd2 = open("foo", O_RDWR);

where they expect fd1 and fd2 to refer to the same file - which will no longer
be the case post-copy up.

With this patch, the following commands:

	bash 5</mnt/a/foo128
	6<>/mnt/a/foo128

assuming /mnt/a/foo128 to be an un-copied up file on an overlay will produce
the following warning in the kernel log:

	overlayfs: Copying up foo129, but open R/O on fd 5 which will cease to
	be coherent [pid=3818 bash]

This is enabled by setting:

	/sys/module/overlay/parameters/check_copy_up

to 1.

The warnings are ratelimited and are also limited to one warning per file -
assuming the copy up completes in each case.

Signed-off-by: David Howells <[e-mail]>
---
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Miklos Szeredi March 7, 2016, 10:06 a.m. UTC | #1

On Tue, Jul 7, 2015 at 4:04 PM, David Howells <[e-mail]> wrote:
>
> Print a warning when overlayfs copies up a file if the process that triggered
> the copy up has a R/O fd open to the lower file being copied up.
>
> This can help catch applications that do things like the following:
>
>         fd1 = open("foo", O_RDONLY);
>         fd2 = open("foo", O_RDWR);
>
> where they expect fd1 and fd2 to refer to the same file - which will no longer
> be the case post-copy up.
>
> With this patch, the following commands:
>
>         bash 5</mnt/a/foo128
>         6<>/mnt/a/foo128
>
> assuming /mnt/a/foo128 to be an un-copied up file on an overlay will produce
> the following warning in the kernel log:
>
>         overlayfs: Copying up foo129, but open R/O on fd 5 which will cease to
>         be coherent [pid=3818 bash]
>
> This is enabled by setting:
>
>         /sys/module/overlay/parameters/check_copy_up
>
> to 1.
>
> The warnings are ratelimited and are also limited to one warning per file -
> assuming the copy up completes in each case.
>

Queued.

Thanks,
Miklos
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/overlayfs/copy_up.c b/fs/overlayfs/copy_up.c
index 84d693d37428..e1aef2ba2f08 100644
--- a/fs/overlayfs/copy_up.c
+++ b/fs/overlayfs/copy_up.c
@@ -7,6 +7,7 @@ 
  * the Free Software Foundation.
  */
 
+#include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/slab.h>
 #include <linux/file.h>
@@ -16,10 +17,41 @@ 
 #include <linux/uaccess.h>
 #include <linux/sched.h>
 #include <linux/namei.h>
+#include <linux/fdtable.h>
+#include <linux/ratelimit.h>
 #include "overlayfs.h"
 
 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
 
+static bool __read_mostly ovl_check_copy_up;
+module_param_named(check_copy_up, ovl_check_copy_up, bool,
+		   S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(ovl_check_copy_up,
+		 "Warn on copy-up when causing process also has a R/O fd open");
+
+static int ovl_check_fd(const void *data, struct file *f, unsigned fd)
+{
+	const struct dentry *dentry = data;
+
+	if (f->f_inode == d_inode(dentry))
+		pr_warn_ratelimited("overlayfs: Warning: Copying up %pD, but open R/O on fd %u which will cease to be coherent [pid=%d %s]\n",
+				    f, fd, current->pid, current->comm);
+	return 0;
+}
+
+/*
+ * Check the fds open by this process and warn if something like the following
+ * scenario is about to occur:
+ *
+ *	fd1 = open("foo", O_RDONLY);
+ *	fd2 = open("foo", O_RDWR);
+ */
+static void ovl_do_check_copy_up(struct dentry *dentry)
+{
+	if (ovl_check_copy_up)
+		iterate_fd(current->files, 0, ovl_check_fd, dentry);
+}
+
 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
 {
 	ssize_t list_size, size;
@@ -302,6 +334,8 @@  int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 	if (WARN_ON(!workdir))
 		return -EROFS;
 
+	ovl_do_check_copy_up(lowerpath->dentry);
+
 	ovl_path_upper(parent, &parentpath);
 	upperdir = parentpath.dentry;