5 Shadow variables are a simple way for livepatch modules to associate
6 additional "shadow" data with existing data structures. Shadow data is
7 allocated separately from parent data structures, which are left
8 unmodified. The shadow variable API described in this document is used
9 to allocate/add and remove/free shadow variables to/from their parents.
11 The implementation introduces a global, in-kernel hashtable that
12 associates pointers to parent objects and a numeric identifier of the
13 shadow data. The numeric identifier is a simple enumeration that may be
14 used to describe shadow variable version, class or type, etc. More
15 specifically, the parent pointer serves as the hashtable key while the
16 numeric id subsequently filters hashtable queries. Multiple shadow
17 variables may attach to the same parent object, but their numeric
18 identifier distinguishes between them.
24 (See the full API usage docbook notes in livepatch/shadow.c.)
26 A hashtable references all shadow variables. These references are
27 stored and retrieved through a <obj, id> pair.
29 * The klp_shadow variable data structure encapsulates both tracking
30 meta-data and shadow-data:
34 - obj - pointer to parent object
35 - id - data identifier
37 - data[] - storage for shadow data
39 It is important to note that the klp_shadow_alloc() and
40 klp_shadow_get_or_alloc() are zeroing the variable by default.
41 They also allow to call a custom constructor function when a non-zero
42 value is needed. Callers should provide whatever mutual exclusion
45 Note that the constructor is called under klp_shadow_lock spinlock. It allows
46 to do actions that can be done only once when a new variable is allocated.
48 * klp_shadow_get() - retrieve a shadow variable data pointer
49 - search hashtable for <obj, id> pair
51 * klp_shadow_alloc() - allocate and add a new shadow variable
52 - search hashtable for <obj, id> pair
56 - WARN and return NULL
58 - if <obj, id> doesn't already exist
60 - allocate a new shadow variable
61 - initialize the variable using a custom constructor and data when provided
62 - add <obj, id> to the global hashtable
64 * klp_shadow_get_or_alloc() - get existing or alloc a new shadow variable
65 - search hashtable for <obj, id> pair
69 - return existing shadow variable
71 - if <obj, id> doesn't already exist
73 - allocate a new shadow variable
74 - initialize the variable using a custom constructor and data when provided
75 - add <obj, id> pair to the global hashtable
77 * klp_shadow_free() - detach and free a <obj, id> shadow variable
78 - find and remove a <obj, id> reference from global hashtable
82 - call destructor function if defined
83 - free shadow variable
85 * klp_shadow_free_all() - detach and free all <_, id> shadow variables
86 - find and remove any <_, id> references from global hashtable
90 - call destructor function if defined
91 - free shadow variable
97 (See the example shadow variable livepatch modules in samples/livepatch/
98 for full working demonstrations.)
100 For the following use-case examples, consider commit 1d147bfa6429
101 ("mac80211: fix AP powersave TX vs. wakeup race"), which added a
102 spinlock to net/mac80211/sta_info.h :: struct sta_info. Each use-case
103 example can be considered a stand-alone livepatch implementation of this
107 Matching parent's lifecycle
108 ---------------------------
110 If parent data structures are frequently created and destroyed, it may
111 be easiest to align their shadow variables lifetimes to the same
112 allocation and release functions. In this case, the parent data
113 structure is typically allocated, initialized, then registered in some
114 manner. Shadow variable allocation and setup can then be considered
115 part of the parent's initialization and should be completed before the
116 parent "goes live" (ie, any shadow variable get-API requests are made
117 for this <obj, id> pair.)
119 For commit 1d147bfa6429, when a parent sta_info structure is allocated,
120 allocate a shadow copy of the ps_lock pointer, then initialize it::
123 struct sta_info *sta_info_alloc(struct ieee80211_sub_if_data *sdata,
124 const u8 *addr, gfp_t gfp)
126 struct sta_info *sta;
129 /* Parent structure is created */
130 sta = kzalloc(sizeof(*sta) + hw->sta_data_size, gfp);
132 /* Attach a corresponding shadow variable, then initialize it */
133 ps_lock = klp_shadow_alloc(sta, PS_LOCK, sizeof(*ps_lock), gfp,
137 spin_lock_init(ps_lock);
140 When requiring a ps_lock, query the shadow variable API to retrieve one
141 for a specific struct sta_info:::
143 void ieee80211_sta_ps_deliver_wakeup(struct sta_info *sta)
147 /* sync with ieee80211_tx_h_unicast_ps_buf */
148 ps_lock = klp_shadow_get(sta, PS_LOCK);
153 When the parent sta_info structure is freed, first free the shadow
156 void sta_info_free(struct ieee80211_local *local, struct sta_info *sta)
158 klp_shadow_free(sta, PS_LOCK, NULL);
163 In-flight parent objects
164 ------------------------
166 Sometimes it may not be convenient or possible to allocate shadow
167 variables alongside their parent objects. Or a livepatch fix may
168 require shadow variables for only a subset of parent object instances.
169 In these cases, the klp_shadow_get_or_alloc() call can be used to attach
170 shadow variables to parents already in-flight.
172 For commit 1d147bfa6429, a good spot to allocate a shadow spinlock is
173 inside ieee80211_sta_ps_deliver_wakeup()::
175 int ps_lock_shadow_ctor(void *obj, void *shadow_data, void *ctor_data)
177 spinlock_t *lock = shadow_data;
179 spin_lock_init(lock);
184 void ieee80211_sta_ps_deliver_wakeup(struct sta_info *sta)
188 /* sync with ieee80211_tx_h_unicast_ps_buf */
189 ps_lock = klp_shadow_get_or_alloc(sta, PS_LOCK,
190 sizeof(*ps_lock), GFP_ATOMIC,
191 ps_lock_shadow_ctor, NULL);
197 This usage will create a shadow variable, only if needed, otherwise it
198 will use one that was already created for this <obj, id> pair.
200 Like the previous use-case, the shadow spinlock needs to be cleaned up.
201 A shadow variable can be freed just before its parent object is freed,
202 or even when the shadow variable itself is no longer required.
208 Shadow variables can also be used as a flag indicating that a data
209 structure was allocated by new, livepatched code. In this case, it
210 doesn't matter what data value the shadow variable holds, its existence
211 suggests how to handle the parent object.
217 * https://github.com/dynup/kpatch
219 The livepatch implementation is based on the kpatch version of shadow
222 * http://files.mkgnu.net/files/dynamos/doc/papers/dynamos_eurosys_07.pdf
224 Dynamic and Adaptive Updates of Non-Quiescent Subsystems in Commodity
225 Operating System Kernels (Kritis Makris, Kyung Dong Ryu 2007) presented
226 a datatype update technique called "shadow data structures".