49 lines · plain
1.. SPDX-License-Identifier: GPL-2.02 3=============4Sysfs tagging5=============6 7(Taken almost verbatim from Eric Biederman's netns tagging patch8commit msg)9 10The problem. Network devices show up in sysfs and with the network11namespace active multiple devices with the same name can show up in12the same directory, ouch!13 14To avoid that problem and allow existing applications in network15namespaces to see the same interface that is currently presented in16sysfs, sysfs now has tagging directory support.17 18By using the network namespace pointers as tags to separate out19the sysfs directory entries we ensure that we don't have conflicts20in the directories and applications only see a limited set of21the network devices.22 23Each sysfs directory entry may be tagged with a namespace via the24``void *ns member`` of its ``kernfs_node``. If a directory entry is tagged,25then ``kernfs_node->flags`` will have a flag between KOBJ_NS_TYPE_NONE26and KOBJ_NS_TYPES, and ns will point to the namespace to which it27belongs.28 29Each sysfs superblock's kernfs_super_info contains an array30``void *ns[KOBJ_NS_TYPES]``. When a task in a tagging namespace31kobj_nstype first mounts sysfs, a new superblock is created. It32will be differentiated from other sysfs mounts by having its33``s_fs_info->ns[kobj_nstype]`` set to the new namespace. Note that34through bind mounting and mounts propagation, a task can easily view35the contents of other namespaces' sysfs mounts. Therefore, when a36namespace exits, it will call kobj_ns_exit() to invalidate any37kernfs_node->ns pointers pointing to it.38 39Users of this interface:40 41- define a type in the ``kobj_ns_type`` enumeration.42- call kobj_ns_type_register() with its ``kobj_ns_type_operations`` which has43 44 - current_ns() which returns current's namespace45 - netlink_ns() which returns a socket's namespace46 - initial_ns() which returns the initial namespace47 48- call kobj_ns_exit() when an individual tag is no longer valid49