(I thought I sent this already?)
Marcelo,
On some architectures (ie. i386) the internal kernel data-structure
'struct inode' has a nlink resolution of only 16-bits. This can
obviously overflow is you put 2^16-2 (65534) directories in one place.
Older (compatible) syscalls don't have necessary resolution to return
correct values anyhow so checks are made and if there is a low of
value EOVERFLOW is returned (as suggest by Al Viro). This seems
pretty reasonable and it's unlikely anything much uses the old
syscalls and if they do probably not likely they will stat and
directories with more than 2^16-2 directory children.
This patch prevents the internal overflow and adds the checks to the
older.
Signed-of-by: Chris Wedgwood <[email protected]>
fs/stat.c | 4 ++++
include/linux/fs.h | 2 +-
2 files changed, 5 insertions(+), 1 deletion(-)
===== fs/stat.c 1.6 vs edited =====
--- 1.6/fs/stat.c 2003-11-28 19:26:56 -08:00
+++ edited/fs/stat.c 2004-06-14 22:30:15 -07:00
@@ -52,6 +52,8 @@
tmp.st_ino = inode->i_ino;
tmp.st_mode = inode->i_mode;
tmp.st_nlink = inode->i_nlink;
+ if (tmp.st_nlink != inode->i_nlink)
+ return -EOVERFLOW;
SET_OLDSTAT_UID(tmp, inode->i_uid);
SET_OLDSTAT_GID(tmp, inode->i_gid);
tmp.st_rdev = kdev_t_to_nr(inode->i_rdev);
@@ -78,6 +80,8 @@
tmp.st_ino = inode->i_ino;
tmp.st_mode = inode->i_mode;
tmp.st_nlink = inode->i_nlink;
+ if (tmp.st_nlink != inode->i_nlink)
+ return -EOVERFLOW;
SET_STAT_UID(tmp, inode->i_uid);
SET_STAT_GID(tmp, inode->i_gid);
tmp.st_rdev = kdev_t_to_nr(inode->i_rdev);
===== include/linux/fs.h 1.96 vs edited =====
--- 1.96/include/linux/fs.h 2004-01-08 07:03:26 -08:00
+++ edited/include/linux/fs.h 2004-06-14 22:29:26 -07:00
@@ -449,7 +449,7 @@
atomic_t i_count;
kdev_t i_dev;
umode_t i_mode;
- nlink_t i_nlink;
+ unsigned int i_nlink;
uid_t i_uid;
gid_t i_gid;
kdev_t i_rdev;