1Notes: 2001-09-24
2-----------------
3
4This "description" (if one chooses to call it that) needed some major updating
5so here goes. This update addresses a change being made at the same time to
6OpenSSL, and it pretty much completely restructures the underlying mechanics of
7the "ENGINE" code. So it serves a double purpose of being a "ENGINE internals
8for masochists" document *and* a rather extensive commit log message. (I'd get
9lynched for sticking all this in CHANGES or the commit mails :-).
10
11ENGINE_TABLE underlies this restructuring, as described in the internal header
12"eng_int.h", implemented in eng_table.c, and used in each of the "class" files;
13tb_rsa.c, tb_dsa.c, etc.
14
15However, "EVP_CIPHER" underlies the motivation and design of ENGINE_TABLE so
16I'll mention a bit about that first. EVP_CIPHER (and most of this applies
17equally to EVP_MD for digests) is both a "method" and a algorithm/mode
18identifier that, in the current API, "lingers". These cipher description +
19implementation structures can be defined or obtained directly by applications,
20or can be loaded "en masse" into EVP storage so that they can be catalogued and
21searched in various ways, ie. two ways of encrypting with the "des_cbc"
22algorithm/mode pair are;
23
24(i) directly;
25     const EVP_CIPHER *cipher = EVP_des_cbc();
26     EVP_EncryptInit(&ctx, cipher, key, iv);
27     [ ... use EVP_EncryptUpdate() and EVP_EncryptFinal() ...]
28
29(ii) indirectly; 
30     OpenSSL_add_all_ciphers();
31     cipher = EVP_get_cipherbyname("des_cbc");
32     EVP_EncryptInit(&ctx, cipher, key, iv);
33     [ ... etc ... ]
34
35The latter is more generally used because it also allows ciphers/digests to be
36looked up based on other identifiers which can be useful for automatic cipher
37selection, eg. in SSL/TLS, or by user-controllable configuration.
38
39The important point about this is that EVP_CIPHER definitions and structures are
40passed around with impunity and there is no safe way, without requiring massive
41rewrites of many applications, to assume that EVP_CIPHERs can be reference
42counted. One an EVP_CIPHER is exposed to the caller, neither it nor anything it
43comes from can "safely" be destroyed. Unless of course the way of getting to
44such ciphers is via entirely distinct API calls that didn't exist before.
45However existing API usage cannot be made to understand when an EVP_CIPHER
46pointer, that has been passed to the caller, is no longer being used.
47
48The other problem with the existing API w.r.t. to hooking EVP_CIPHER support
49into ENGINE is storage - the OBJ_NAME-based storage used by EVP to register
50ciphers simultaneously registers cipher *types* and cipher *implementations* -
51they are effectively the same thing, an "EVP_CIPHER" pointer. The problem with
52hooking in ENGINEs is that multiple ENGINEs may implement the same ciphers. The
53solution is necessarily that ENGINE-provided ciphers simply are not registered,
54stored, or exposed to the caller in the same manner as existing ciphers. This is
55especially necessary considering the fact ENGINE uses reference counts to allow
56for cleanup, modularity, and DSO support - yet EVP_CIPHERs, as exposed to
57callers in the current API, support no such controls.
58
59Another sticking point for integrating cipher support into ENGINE is linkage.
60Already there is a problem with the way ENGINE supports RSA, DSA, etc whereby
61they are available *because* they're part of a giant ENGINE called "openssl".
62Ie. all implementations *have* to come from an ENGINE, but we get round that by
63having a giant ENGINE with all the software support encapsulated. This creates
64linker hassles if nothing else - linking a 1-line application that calls 2 basic
65RSA functions (eg. "RSA_free(RSA_new());") will result in large quantities of
66ENGINE code being linked in *and* because of that DSA, DH, and RAND also. If we
67continue with this approach for EVP_CIPHER support (even if it *was* possible)
68we would lose our ability to link selectively by selectively loading certain
69implementations of certain functionality. Touching any part of any kind of
70crypto would result in massive static linkage of everything else. So the
71solution is to change the way ENGINE feeds existing "classes", ie. how the
72hooking to ENGINE works from RSA, DSA, DH, RAND, as well as adding new hooking
73for EVP_CIPHER, and EVP_MD.
74
75The way this is now being done is by mostly reverting back to how things used to
76work prior to ENGINE :-). Ie. RSA now has a "RSA_METHOD" pointer again - this
77was previously replaced by an "ENGINE" pointer and all RSA code that required
78the RSA_METHOD would call ENGINE_get_RSA() each time on its ENGINE handle to
79temporarily get and use the ENGINE's RSA implementation. Apart from being more
80efficient, switching back to each RSA having an RSA_METHOD pointer also allows
81us to conceivably operate with *no* ENGINE. As we'll see, this removes any need
82for a fallback ENGINE that encapsulates default implementations - we can simply
83have our RSA structure pointing its RSA_METHOD pointer to the software
84implementation and have its ENGINE pointer set to NULL.
85
86A look at the EVP_CIPHER hooking is most explanatory, the RSA, DSA (etc) cases
87turn out to be degenerate forms of the same thing. The EVP storage of ciphers,
88and the existing EVP API functions that return "software" implementations and
89descriptions remain untouched. However, the storage takes more meaning in terms
90of "cipher description" and less meaning in terms of "implementation". When an
91EVP_CIPHER_CTX is actually initialised with an EVP_CIPHER method and is about to
92begin en/decryption, the hooking to ENGINE comes into play. What happens is that
93cipher-specific ENGINE code is asked for an ENGINE pointer (a functional
94reference) for any ENGINE that is registered to perform the algo/mode that the
95provided EVP_CIPHER structure represents. Under normal circumstances, that
96ENGINE code will return NULL because no ENGINEs will have had any cipher
97implementations *registered*. As such, a NULL ENGINE pointer is stored in the
98EVP_CIPHER_CTX context, and the EVP_CIPHER structure is left hooked into the
99context and so is used as the implementation. Pretty much how things work now
100except we'd have a redundant ENGINE pointer set to NULL and doing nothing.
101
102Conversely, if an ENGINE *has* been registered to perform the algorithm/mode
103combination represented by the provided EVP_CIPHER, then a functional reference
104to that ENGINE will be returned to the EVP_CIPHER_CTX during initialisation.
105That functional reference will be stored in the context (and released on
106cleanup) - and having that reference provides a *safe* way to use an EVP_CIPHER
107definition that is private to the ENGINE. Ie. the EVP_CIPHER provided by the
108application will actually be replaced by an EVP_CIPHER from the registered
109ENGINE - it will support the same algorithm/mode as the original but will be a
110completely different implementation. Because this EVP_CIPHER isn't stored in the
111EVP storage, nor is it returned to applications from traditional API functions,
112there is no associated problem with it not having reference counts. And of
113course, when one of these "private" cipher implementations is hooked into
114EVP_CIPHER_CTX, it is done whilst the EVP_CIPHER_CTX holds a functional
115reference to the ENGINE that owns it, thus the use of the ENGINE's EVP_CIPHER is
116safe.
117
118The "cipher-specific ENGINE code" I mentioned is implemented in tb_cipher.c but
119in essence it is simply an instantiation of "ENGINE_TABLE" code for use by
120EVP_CIPHER code. tb_digest.c is virtually identical but, of course, it is for
121use by EVP_MD code. Ditto for tb_rsa.c, tb_dsa.c, etc. These instantiations of
122ENGINE_TABLE essentially provide linker-separation of the classes so that even
123if ENGINEs implement *all* possible algorithms, an application using only
124EVP_CIPHER code will link at most code relating to EVP_CIPHER, tb_cipher.c, core
125ENGINE code that is independant of class, and of course the ENGINE
126implementation that the application loaded. It will *not* however link any
127class-specific ENGINE code for digests, RSA, etc nor will it bleed over into
128other APIs, such as the RSA/DSA/etc library code.
129
130ENGINE_TABLE is a little more complicated than may seem necessary but this is
131mostly to avoid a lot of "init()"-thrashing on ENGINEs (that may have to load
132DSOs, and other expensive setup that shouldn't be thrashed unnecessarily) *and*
133to duplicate "default" behaviour. Basically an ENGINE_TABLE instantiation, for
134example tb_cipher.c, implements a hash-table keyed by integer "nid" values.
135These nids provide the uniquenness of an algorithm/mode - and each nid will hash
136to a potentially NULL "ENGINE_PILE". An ENGINE_PILE is essentially a list of
137pointers to ENGINEs that implement that particular 'nid'. Each "pile" uses some
138caching tricks such that requests on that 'nid' will be cached and all future
139requests will return immediately (well, at least with minimal operation) unless
140a change is made to the pile, eg. perhaps an ENGINE was unloaded. The reason is
141that an application could have support for 10 ENGINEs statically linked
142in, and the machine in question may not have any of the hardware those 10
143ENGINEs support. If each of those ENGINEs has a "des_cbc" implementation, we
144want to avoid every EVP_CIPHER_CTX setup from trying (and failing) to initialise
145each of those 10 ENGINEs. Instead, the first such request will try to do that
146and will either return (and cache) a NULL ENGINE pointer or will return a
147functional reference to the first that successfully initialised. In the latter
148case it will also cache an extra functional reference to the ENGINE as a
149"default" for that 'nid'. The caching is acknowledged by a 'uptodate' variable
150that is unset only if un/registration takes place on that pile. Ie. if
151implementations of "des_cbc" are added or removed. This behaviour can be
152tweaked; the ENGINE_TABLE_FLAG_NOINIT value can be passed to
153ENGINE_set_table_flags(), in which case the only ENGINEs that tb_cipher.c will
154try to initialise from the "pile" will be those that are already initialised
155(ie. it's simply an increment of the functional reference count, and no real
156"initialisation" will take place).
157
158RSA, DSA, DH, and RAND all have their own ENGINE_TABLE code as well, and the
159difference is that they all use an implicit 'nid' of 1. Whereas EVP_CIPHERs are
160actually qualitatively different depending on 'nid' (the "des_cbc" EVP_CIPHER is
161not an interoperable implementation of "aes_256_cbc"), RSA_METHODs are
162necessarily interoperable and don't have different flavours, only different
163implementations. In other words, the ENGINE_TABLE for RSA will either be empty,
164or will have a single ENGING_PILE hashed to by the 'nid' 1 and that pile
165represents ENGINEs that implement the single "type" of RSA there is.
166
167Cleanup - the registration and unregistration may pose questions about how
168cleanup works with the ENGINE_PILE doing all this caching nonsense (ie. when the
169application or EVP_CIPHER code releases its last reference to an ENGINE, the
170ENGINE_PILE code may still have references and thus those ENGINEs will stay
171hooked in forever). The way this is handled is via "unregistration". With these
172new ENGINE changes, an abstract ENGINE can be loaded and initialised, but that
173is an algorithm-agnostic process. Even if initialised, it will not have
174registered any of its implementations (to do so would link all class "table"
175code despite the fact the application may use only ciphers, for example). This
176is deliberately a distinct step. Moreover, registration and unregistration has
177nothing to do with whether an ENGINE is *functional* or not (ie. you can even
178register an ENGINE and its implementations without it being operational, you may
179not even have the drivers to make it operate). What actually happens with
180respect to cleanup is managed inside eng_lib.c with the "engine_cleanup_***"
181functions. These functions are internal-only and each part of ENGINE code that
182could require cleanup will, upon performing its first allocation, register a
183callback with the "engine_cleanup" code. The other part of this that makes it
184tick is that the ENGINE_TABLE instantiations (tb_***.c) use NULL as their
185initialised state. So if RSA code asks for an ENGINE and no ENGINE has
186registered an implementation, the code will simply return NULL and the tb_rsa.c
187state will be unchanged. Thus, no cleanup is required unless registration takes
188place. ENGINE_cleanup() will simply iterate across a list of registered cleanup
189callbacks calling each in turn, and will then internally delete its own storage
190(a STACK). When a cleanup callback is next registered (eg. if the cleanup() is
191part of a gracefull restart and the application wants to cleanup all state then
192start again), the internal STACK storage will be freshly allocated. This is much
193the same as the situation in the ENGINE_TABLE instantiations ... NULL is the
194initialised state, so only modification operations (not queries) will cause that
195code to have to register a cleanup.
196
197What else? The bignum callbacks and associated ENGINE functions have been
198removed for two obvious reasons; (i) there was no way to generalise them to the
199mechanism now used by RSA/DSA/..., because there's no such thing as a BIGNUM
200method, and (ii) because of (i), there was no meaningful way for library or
201application code to automatically hook and use ENGINE supplied bignum functions
202anyway. Also, ENGINE_cpy() has been removed (although an internal-only version
203exists) - the idea of providing an ENGINE_cpy() function probably wasn't a good
204one and now certainly doesn't make sense in any generalised way. Some of the
205RSA, DSA, DH, and RAND functions that were fiddled during the original ENGINE
206changes have now, as a consequence, been reverted back. This is because the
207hooking of ENGINE is now automatic (and passive, it can interally use a NULL
208ENGINE pointer to simply ignore ENGINE from then on).
209
210Hell, that should be enough for now ... comments welcome: geoff@openssl.org
211
212