1=pod
2
3=head1 NAME
4
5lh_new, lh_free, lh_insert, lh_delete, lh_retrieve, lh_doall, lh_doall_arg, lh_error - dynamic hash table
6
7=head1 SYNOPSIS
8
9 #include <openssl/lhash.h>
10
11 DECLARE_LHASH_OF(<type>);
12
13 LHASH *lh_<type>_new();
14 void lh_<type>_free(LHASH_OF(<type> *table);
15
16 <type> *lh_<type>_insert(LHASH_OF(<type> *table, <type> *data);
17 <type> *lh_<type>_delete(LHASH_OF(<type> *table, <type> *data);
18 <type> *lh_retrieve(LHASH_OF<type> *table, <type> *data);
19
20 void lh_<type>_doall(LHASH_OF(<type> *table, LHASH_DOALL_FN_TYPE func);
21 void lh_<type>_doall_arg(LHASH_OF(<type> *table, LHASH_DOALL_ARG_FN_TYPE func,
22          <type2>, <type2> *arg);
23
24 int lh_<type>_error(LHASH_OF(<type> *table);
25
26 typedef int (*LHASH_COMP_FN_TYPE)(const void *, const void *);
27 typedef unsigned long (*LHASH_HASH_FN_TYPE)(const void *);
28 typedef void (*LHASH_DOALL_FN_TYPE)(const void *);
29 typedef void (*LHASH_DOALL_ARG_FN_TYPE)(const void *, const void *);
30
31=head1 DESCRIPTION
32
33This library implements type-checked dynamic hash tables. The hash
34table entries can be arbitrary structures. Usually they consist of key
35and value fields.
36
37lh_<type>_new() creates a new B<LHASH_OF(<type>> structure to store
38arbitrary data entries, and provides the 'hash' and 'compare'
39callbacks to be used in organising the table's entries.  The B<hash>
40callback takes a pointer to a table entry as its argument and returns
41an unsigned long hash value for its key field.  The hash value is
42normally truncated to a power of 2, so make sure that your hash
43function returns well mixed low order bits.  The B<compare> callback
44takes two arguments (pointers to two hash table entries), and returns
450 if their keys are equal, non-zero otherwise.  If your hash table
46will contain items of some particular type and the B<hash> and
47B<compare> callbacks hash/compare these types, then the
48B<DECLARE_LHASH_HASH_FN> and B<IMPLEMENT_LHASH_COMP_FN> macros can be
49used to create callback wrappers of the prototypes required by
50lh_<type>_new().  These provide per-variable casts before calling the
51type-specific callbacks written by the application author.  These
52macros, as well as those used for the "doall" callbacks, are defined
53as;
54
55 #define DECLARE_LHASH_HASH_FN(name, o_type) \
56	 unsigned long name##_LHASH_HASH(const void *);
57 #define IMPLEMENT_LHASH_HASH_FN(name, o_type) \
58	 unsigned long name##_LHASH_HASH(const void *arg) { \
59		 const o_type *a = arg; \
60		 return name##_hash(a); }
61 #define LHASH_HASH_FN(name) name##_LHASH_HASH
62
63 #define DECLARE_LHASH_COMP_FN(name, o_type) \
64	 int name##_LHASH_COMP(const void *, const void *);
65 #define IMPLEMENT_LHASH_COMP_FN(name, o_type) \
66	 int name##_LHASH_COMP(const void *arg1, const void *arg2) { \
67		 const o_type *a = arg1;		    \
68		 const o_type *b = arg2; \
69		 return name##_cmp(a,b); }
70 #define LHASH_COMP_FN(name) name##_LHASH_COMP
71
72 #define DECLARE_LHASH_DOALL_FN(name, o_type) \
73	 void name##_LHASH_DOALL(void *);
74 #define IMPLEMENT_LHASH_DOALL_FN(name, o_type) \
75	 void name##_LHASH_DOALL(void *arg) { \
76		 o_type *a = arg; \
77		 name##_doall(a); }
78 #define LHASH_DOALL_FN(name) name##_LHASH_DOALL
79
80 #define DECLARE_LHASH_DOALL_ARG_FN(name, o_type, a_type) \
81	 void name##_LHASH_DOALL_ARG(void *, void *);
82 #define IMPLEMENT_LHASH_DOALL_ARG_FN(name, o_type, a_type) \
83	 void name##_LHASH_DOALL_ARG(void *arg1, void *arg2) { \
84		 o_type *a = arg1; \
85		 a_type *b = arg2; \
86		 name##_doall_arg(a, b); }
87 #define LHASH_DOALL_ARG_FN(name) name##_LHASH_DOALL_ARG
88
89 An example of a hash table storing (pointers to) structures of type 'STUFF'
90 could be defined as follows;
91
92 /* Calculates the hash value of 'tohash' (implemented elsewhere) */
93 unsigned long STUFF_hash(const STUFF *tohash);
94 /* Orders 'arg1' and 'arg2' (implemented elsewhere) */
95 int stuff_cmp(const STUFF *arg1, const STUFF *arg2);
96 /* Create the type-safe wrapper functions for use in the LHASH internals */
97 static IMPLEMENT_LHASH_HASH_FN(stuff, STUFF);
98 static IMPLEMENT_LHASH_COMP_FN(stuff, STUFF);
99 /* ... */
100 int main(int argc, char *argv[]) {
101         /* Create the new hash table using the hash/compare wrappers */
102         LHASH_OF(STUFF) *hashtable = lh_STUFF_new(LHASH_HASH_FN(STUFF_hash),
103                                   LHASH_COMP_FN(STUFF_cmp));
104	 /* ... */
105 }
106
107lh_<type>_free() frees the B<LHASH_OF(<type>> structure
108B<table>. Allocated hash table entries will not be freed; consider
109using lh_<type>_doall() to deallocate any remaining entries in the
110hash table (see below).
111
112lh_<type>_insert() inserts the structure pointed to by B<data> into
113B<table>.  If there already is an entry with the same key, the old
114value is replaced. Note that lh_<type>_insert() stores pointers, the
115data are not copied.
116
117lh_<type>_delete() deletes an entry from B<table>.
118
119lh_<type>_retrieve() looks up an entry in B<table>. Normally, B<data>
120is a structure with the key field(s) set; the function will return a
121pointer to a fully populated structure.
122
123lh_<type>_doall() will, for every entry in the hash table, call
124B<func> with the data item as its parameter.  For lh_<type>_doall()
125and lh_<type>_doall_arg(), function pointer casting should be avoided
126in the callbacks (see B<NOTE>) - instead use the declare/implement
127macros to create type-checked wrappers that cast variables prior to
128calling your type-specific callbacks.  An example of this is
129illustrated here where the callback is used to cleanup resources for
130items in the hash table prior to the hashtable itself being
131deallocated:
132
133 /* Cleans up resources belonging to 'a' (this is implemented elsewhere) */
134 void STUFF_cleanup_doall(STUFF *a);
135 /* Implement a prototype-compatible wrapper for "STUFF_cleanup" */
136 IMPLEMENT_LHASH_DOALL_FN(STUFF_cleanup, STUFF)
137         /* ... then later in the code ... */
138 /* So to run "STUFF_cleanup" against all items in a hash table ... */
139 lh_STUFF_doall(hashtable, LHASH_DOALL_FN(STUFF_cleanup));
140 /* Then the hash table itself can be deallocated */
141 lh_STUFF_free(hashtable);
142
143When doing this, be careful if you delete entries from the hash table
144in your callbacks: the table may decrease in size, moving the item
145that you are currently on down lower in the hash table - this could
146cause some entries to be skipped during the iteration.  The second
147best solution to this problem is to set hash-E<gt>down_load=0 before
148you start (which will stop the hash table ever decreasing in size).
149The best solution is probably to avoid deleting items from the hash
150table inside a "doall" callback!
151
152lh_<type>_doall_arg() is the same as lh_<type>_doall() except that
153B<func> will be called with B<arg> as the second argument and B<func>
154should be of type B<LHASH_DOALL_ARG_FN_TYPE> (a callback prototype
155that is passed both the table entry and an extra argument).  As with
156lh_doall(), you can instead choose to declare your callback with a
157prototype matching the types you are dealing with and use the
158declare/implement macros to create compatible wrappers that cast
159variables before calling your type-specific callbacks.  An example of
160this is demonstrated here (printing all hash table entries to a BIO
161that is provided by the caller):
162
163 /* Prints item 'a' to 'output_bio' (this is implemented elsewhere) */
164 void STUFF_print_doall_arg(const STUFF *a, BIO *output_bio);
165 /* Implement a prototype-compatible wrapper for "STUFF_print" */
166 static IMPLEMENT_LHASH_DOALL_ARG_FN(STUFF, const STUFF, BIO)
167         /* ... then later in the code ... */
168 /* Print out the entire hashtable to a particular BIO */
169 lh_STUFF_doall_arg(hashtable, LHASH_DOALL_ARG_FN(STUFF_print), BIO,
170                    logging_bio);
171
172lh_<type>_error() can be used to determine if an error occurred in the last
173operation. lh_<type>_error() is a macro.
174
175=head1 RETURN VALUES
176
177lh_<type>_new() returns B<NULL> on error, otherwise a pointer to the new
178B<LHASH> structure.
179
180When a hash table entry is replaced, lh_<type>_insert() returns the value
181being replaced. B<NULL> is returned on normal operation and on error.
182
183lh_<type>_delete() returns the entry being deleted.  B<NULL> is returned if
184there is no such value in the hash table.
185
186lh_<type>_retrieve() returns the hash table entry if it has been found,
187B<NULL> otherwise.
188
189lh_<type>_error() returns 1 if an error occurred in the last operation, 0
190otherwise.
191
192lh_<type>_free(), lh_<type>_doall() and lh_<type>_doall_arg() return no values.
193
194=head1 NOTE
195
196The various LHASH macros and callback types exist to make it possible
197to write type-checked code without resorting to function-prototype
198casting - an evil that makes application code much harder to
199audit/verify and also opens the window of opportunity for stack
200corruption and other hard-to-find bugs.  It also, apparently, violates
201ANSI-C.
202
203The LHASH code regards table entries as constant data.  As such, it
204internally represents lh_insert()'d items with a "const void *"
205pointer type.  This is why callbacks such as those used by lh_doall()
206and lh_doall_arg() declare their prototypes with "const", even for the
207parameters that pass back the table items' data pointers - for
208consistency, user-provided data is "const" at all times as far as the
209LHASH code is concerned.  However, as callers are themselves providing
210these pointers, they can choose whether they too should be treating
211all such parameters as constant.
212
213As an example, a hash table may be maintained by code that, for
214reasons of encapsulation, has only "const" access to the data being
215indexed in the hash table (ie. it is returned as "const" from
216elsewhere in their code) - in this case the LHASH prototypes are
217appropriate as-is.  Conversely, if the caller is responsible for the
218life-time of the data in question, then they may well wish to make
219modifications to table item passed back in the lh_doall() or
220lh_doall_arg() callbacks (see the "STUFF_cleanup" example above).  If
221so, the caller can either cast the "const" away (if they're providing
222the raw callbacks themselves) or use the macros to declare/implement
223the wrapper functions without "const" types.
224
225Callers that only have "const" access to data they're indexing in a
226table, yet declare callbacks without constant types (or cast the
227"const" away themselves), are therefore creating their own risks/bugs
228without being encouraged to do so by the API.  On a related note,
229those auditing code should pay special attention to any instances of
230DECLARE/IMPLEMENT_LHASH_DOALL_[ARG_]_FN macros that provide types
231without any "const" qualifiers.
232
233=head1 BUGS
234
235lh_<type>_insert() returns B<NULL> both for success and error.
236
237=head1 INTERNALS
238
239The following description is based on the SSLeay documentation:
240
241The B<lhash> library implements a hash table described in the
242I<Communications of the ACM> in 1991.  What makes this hash table
243different is that as the table fills, the hash table is increased (or
244decreased) in size via OPENSSL_realloc().  When a 'resize' is done, instead of
245all hashes being redistributed over twice as many 'buckets', one
246bucket is split.  So when an 'expand' is done, there is only a minimal
247cost to redistribute some values.  Subsequent inserts will cause more
248single 'bucket' redistributions but there will never be a sudden large
249cost due to redistributing all the 'buckets'.
250
251The state for a particular hash table is kept in the B<LHASH> structure.
252The decision to increase or decrease the hash table size is made
253depending on the 'load' of the hash table.  The load is the number of
254items in the hash table divided by the size of the hash table.  The
255default values are as follows.  If (hash->up_load E<lt> load) =E<gt>
256expand.  if (hash-E<gt>down_load E<gt> load) =E<gt> contract.  The
257B<up_load> has a default value of 1 and B<down_load> has a default value
258of 2.  These numbers can be modified by the application by just
259playing with the B<up_load> and B<down_load> variables.  The 'load' is
260kept in a form which is multiplied by 256.  So
261hash-E<gt>up_load=8*256; will cause a load of 8 to be set.
262
263If you are interested in performance the field to watch is
264num_comp_calls.  The hash library keeps track of the 'hash' value for
265each item so when a lookup is done, the 'hashes' are compared, if
266there is a match, then a full compare is done, and
267hash-E<gt>num_comp_calls is incremented.  If num_comp_calls is not equal
268to num_delete plus num_retrieve it means that your hash function is
269generating hashes that are the same for different values.  It is
270probably worth changing your hash function if this is the case because
271even if your hash table has 10 items in a 'bucket', it can be searched
272with 10 B<unsigned long> compares and 10 linked list traverses.  This
273will be much less expensive that 10 calls to your compare function.
274
275lh_strhash() is a demo string hashing function:
276
277 unsigned long lh_strhash(const char *c);
278
279Since the B<LHASH> routines would normally be passed structures, this
280routine would not normally be passed to lh_<type>_new(), rather it would be
281used in the function passed to lh_<type>_new().
282
283=head1 SEE ALSO
284
285L<lh_stats(3)|lh_stats(3)>
286
287=head1 HISTORY
288
289The B<lhash> library is available in all versions of SSLeay and OpenSSL.
290lh_error() was added in SSLeay 0.9.1b.
291
292This manpage is derived from the SSLeay documentation.
293
294In OpenSSL 0.9.7, all lhash functions that were passed function pointers
295were changed for better type safety, and the function types LHASH_COMP_FN_TYPE,
296LHASH_HASH_FN_TYPE, LHASH_DOALL_FN_TYPE and LHASH_DOALL_ARG_FN_TYPE
297became available.
298
299In OpenSSL 1.0.0, the lhash interface was revamped for even better
300type checking.
301
302=cut
303