As Al Viro pointed out: https://lkml.org/lkml/2015/4/11/243 There are bugs in ll_getname() because of wrong assumptions of returning values from strncpy_from_user(). Moreover, what ll_getname want to do is just to try copy the file name from userland. Since we already have getname() for the same purpose, it's better to replace ll_getname() with getname(). To do that, we need to: 1) export the symbols of getname() and putname() to be used by modules. 2) actually replace ll_getname()/ll_putname() with getname()/putname(). One more thing is that as ll_getname() and getname() both treat a zero-length file name as an error(-ENOENT), and if ll_getname() or getname() has an error, ll_dir_ioctl() will return the error immediately, so checking whether these names are zero-length is unnecessary and -ENIVAL shall not return from that code path, no matter using ll_getname() or getname(). So remove the checking code. This patchset is based on v4.0, and I only did build tests, because I found a little difficult to set up a lustre environment. I'll try to do the testing once I'm able to set up an environment, in the meanwhile, comments, inputs from lustre's point of view and voluntary tests are welcome. Thank you. ;-) Regards, Boqun Feng Boqun Feng (2): vfs: Export symbol 'getname' and 'putname' staging: lustre: replace ll_{get,put}name() with {get,put}name() drivers/staging/lustre/lustre/llite/dir.c | 60 ++++++---------------- .../staging/lustre/lustre/llite/llite_internal.h | 2 +- drivers/staging/lustre/lustre/llite/namei.c | 2 +- fs/namei.c | 18 +++++++ 4 files changed, 36 insertions(+), 46 deletions(-) -- 2.3.5 _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
