Messages in this thread | | | Date | Wed, 18 Jun 2008 13:40:43 +0200 | From | Louis Rilling <> | Subject | Re: [BUGFIX][PATCH 3/3] configfs: Fix failing symlink() making rmdir() fail |
| |
On Tue, Jun 17, 2008 at 03:15:28PM -0700, Joel Becker wrote: > On Tue, Jun 17, 2008 at 07:37:23PM +0200, Louis Rilling wrote: > > For the parent's rmdir() case, we can use the same solution as with mkdir() vs > > rmdir(). For the target's rmdir() case, we cannot, since we do not and cannot > > lock the target's inode while in symlink(). Fortunately, once create_link() > > terminates, no further operation can fail in symlink(). So we first reorder the > > operations in create_link() to attach the new symlink to its target in last > > place, and second handle symlink creation failure the same way as a new item > > creation failure. > > Oh, no, ugh. We don't want to create vfs objects first and ask > questions later. Otherwise we wouldn't need ATTACHING - we'd just > create the symlink, then check dropping. > If you have ATTACHING set, the rmdir cannot continue - you can > check dropping at that time. That is, you keep the DROPPING check where > it is - if it is already set, you know that rmdir() is going to complete > successfully. You can bail before even calling configfs_create_link(). > If, however, it isn't set, your ATTACHING protects you from rmdir > throughout.
The problem is rmdir() of the target item (see below). ATTACHING only protects us from rmdir() of the parent. This is the exact reason why I attach the link to the target in last place, where we know that we won't have to rollback. And AFAICS, creating a VFS object can not hurt as long as we hold the parent i_mutex, right? Otherwise there already is a problem in configfs_attach_item() where a failure in populate_attrs() leads to rollback the creation of the VFS object already created for the item.
> > sl = kmalloc(sizeof(struct configfs_symlink), GFP_KERNEL); > if (sl) { > sl->sl_target = config_item_get(item); > spin_lock(&configfs_dirent_lock); > if (target_sd->s_type & CONFIGFS_USET_DROPPING) { > spin_unlock(&configfs_dirent_lock); > config_item_put(item); > kfree(sl); > return -ENOENT; > /* > * Force rmdir() of parent_item to wait until we know > * if we succeed. > */ > parent_sd->s_type |= CONFIGFS_USET_ATTACHING; > } > list_add(&sl->sl_list, &target_sd->s_links); > spin_unlock(&configfs_dirent_lock); > ret = configfs_create_link(sl, parent_item->ci_dentry, > dentry); > spin_lock(&configfs_dirent_lock); > parent_sd->s_type &= ~CONFIGFS_USET_ATTACHING; > if (ret) {
Here, if detach_prep() of the target failed because of the link attached above, it had no means to retry. rmdir() of the target fails because of this temporary link, which results in a failing symlink() making rmdir() of the target fail.
> list_del_init(&sl->sl_list); > spin_unlock(&configfs_dirent_lock); > config_item_put(item); > kfree(sl); > } else > spin_unlock(&configfs_dirent_lock); > } > > return ret; >
-- Dr Louis Rilling Kerlabs Skype: louis.rilling Batiment Germanium Phone: (+33|0) 6 80 89 08 23 80 avenue des Buttes de Coesmes http://www.kerlabs.com/ 35700 Rennes [unhandled content-type:application/pgp-signature] | |