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