1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * lib/textsearch.c Generic text search interface
4 *
5 * Authors: Thomas Graf <tgraf@suug.ch>
6 * Pablo Neira Ayuso <pablo@netfilter.org>
7 *
8 * ==========================================================================
9 */
10
11/**
12 * DOC: ts_intro
13 * INTRODUCTION
14 *
15 * The textsearch infrastructure provides text searching facilities for
16 * both linear and non-linear data. Individual search algorithms are
17 * implemented in modules and chosen by the user.
18 *
19 * ARCHITECTURE
20 *
21 * .. code-block:: none
22 *
23 * User
24 * +----------------+
25 * | finish()|<--------------(6)-----------------+
26 * |get_next_block()|<--------------(5)---------------+ |
27 * | | Algorithm | |
28 * | | +------------------------------+
29 * | | | init() find() destroy() |
30 * | | +------------------------------+
31 * | | Core API ^ ^ ^
32 * | | +---------------+ (2) (4) (8)
33 * | (1)|----->| prepare() |---+ | |
34 * | (3)|----->| find()/next() |-----------+ |
35 * | (7)|----->| destroy() |----------------------+
36 * +----------------+ +---------------+
37 *
38 * (1) User configures a search by calling textsearch_prepare() specifying
39 * the search parameters such as the pattern and algorithm name.
40 * (2) Core requests the algorithm to allocate and initialize a search
41 * configuration according to the specified parameters.
42 * (3) User starts the search(es) by calling textsearch_find() or
43 * textsearch_next() to fetch subsequent occurrences. A state variable
44 * is provided to the algorithm to store persistent variables.
45 * (4) Core eventually resets the search offset and forwards the find()
46 * request to the algorithm.
47 * (5) Algorithm calls get_next_block() provided by the user continuously
48 * to fetch the data to be searched in block by block.
49 * (6) Algorithm invokes finish() after the last call to get_next_block
50 * to clean up any leftovers from get_next_block. (Optional)
51 * (7) User destroys the configuration by calling textsearch_destroy().
52 * (8) Core notifies the algorithm to destroy algorithm specific
53 * allocations. (Optional)
54 *
55 * USAGE
56 *
57 * Before a search can be performed, a configuration must be created
58 * by calling textsearch_prepare() specifying the searching algorithm,
59 * the pattern to look for and flags. As a flag, you can set TS_IGNORECASE
60 * to perform case insensitive matching. But it might slow down
61 * performance of algorithm, so you should use it at own your risk.
62 * The returned configuration may then be used for an arbitrary
63 * amount of times and even in parallel as long as a separate struct
64 * ts_state variable is provided to every instance.
65 *
66 * The actual search is performed by either calling
67 * textsearch_find_continuous() for linear data or by providing
68 * an own get_next_block() implementation and
69 * calling textsearch_find(). Both functions return
70 * the position of the first occurrence of the pattern or UINT_MAX if
71 * no match was found. Subsequent occurrences can be found by calling
72 * textsearch_next() regardless of the linearity of the data.
73 *
74 * Once you're done using a configuration it must be given back via
75 * textsearch_destroy.
76 *
77 * EXAMPLE::
78 *
79 * int pos;
80 * struct ts_config *conf;
81 * struct ts_state state;
82 * const char *pattern = "chicken";
83 * const char *example = "We dance the funky chicken";
84 *
85 * conf = textsearch_prepare("kmp", pattern, strlen(pattern),
86 * GFP_KERNEL, TS_AUTOLOAD);
87 * if (IS_ERR(conf)) {
88 * err = PTR_ERR(conf);
89 * goto errout;
90 * }
91 *
92 * pos = textsearch_find_continuous(conf, &state, example, strlen(example));
93 * if (pos != UINT_MAX)
94 * panic("Oh my god, dancing chickens at %d\n", pos);
95 *
96 * textsearch_destroy(conf);
97 */
98/* ========================================================================== */
99
100#include <linux/module.h>
101#include <linux/types.h>
102#include <linux/string.h>
103#include <linux/init.h>
104#include <linux/rculist.h>
105#include <linux/rcupdate.h>
106#include <linux/err.h>
107#include <linux/textsearch.h>
108#include <linux/slab.h>
109
110static LIST_HEAD(ts_ops);
111static DEFINE_SPINLOCK(ts_mod_lock);
112
113static inline struct ts_ops *lookup_ts_algo(const char *name)
114{
115 struct ts_ops *o;
116
117 rcu_read_lock();
118 list_for_each_entry_rcu(o, &ts_ops, list) {
119 if (!strcmp(name, o->name)) {
120 if (!try_module_get(module: o->owner))
121 o = NULL;
122 rcu_read_unlock();
123 return o;
124 }
125 }
126 rcu_read_unlock();
127
128 return NULL;
129}
130
131/**
132 * textsearch_register - register a textsearch module
133 * @ops: operations lookup table
134 *
135 * This function must be called by textsearch modules to announce
136 * their presence. The specified &@ops must have %name set to a
137 * unique identifier and the callbacks find(), init(), get_pattern(),
138 * and get_pattern_len() must be implemented.
139 *
140 * Returns 0 or -EEXISTS if another module has already registered
141 * with same name.
142 */
143int textsearch_register(struct ts_ops *ops)
144{
145 int err = -EEXIST;
146 struct ts_ops *o;
147
148 if (ops->name == NULL || ops->find == NULL || ops->init == NULL ||
149 ops->get_pattern == NULL || ops->get_pattern_len == NULL)
150 return -EINVAL;
151
152 spin_lock(lock: &ts_mod_lock);
153 list_for_each_entry(o, &ts_ops, list) {
154 if (!strcmp(ops->name, o->name))
155 goto errout;
156 }
157
158 list_add_tail_rcu(new: &ops->list, head: &ts_ops);
159 err = 0;
160errout:
161 spin_unlock(lock: &ts_mod_lock);
162 return err;
163}
164EXPORT_SYMBOL(textsearch_register);
165
166/**
167 * textsearch_unregister - unregister a textsearch module
168 * @ops: operations lookup table
169 *
170 * This function must be called by textsearch modules to announce
171 * their disappearance for examples when the module gets unloaded.
172 * The &ops parameter must be the same as the one during the
173 * registration.
174 *
175 * Returns 0 on success or -ENOENT if no matching textsearch
176 * registration was found.
177 */
178int textsearch_unregister(struct ts_ops *ops)
179{
180 int err = 0;
181 struct ts_ops *o;
182
183 spin_lock(lock: &ts_mod_lock);
184 list_for_each_entry(o, &ts_ops, list) {
185 if (o == ops) {
186 list_del_rcu(entry: &o->list);
187 goto out;
188 }
189 }
190
191 err = -ENOENT;
192out:
193 spin_unlock(lock: &ts_mod_lock);
194 return err;
195}
196EXPORT_SYMBOL(textsearch_unregister);
197
198struct ts_linear_state
199{
200 unsigned int len;
201 const void *data;
202};
203
204static unsigned int get_linear_data(unsigned int consumed, const u8 **dst,
205 struct ts_config *conf,
206 struct ts_state *state)
207{
208 struct ts_linear_state *st = (struct ts_linear_state *) state->cb;
209
210 if (likely(consumed < st->len)) {
211 *dst = st->data + consumed;
212 return st->len - consumed;
213 }
214
215 return 0;
216}
217
218/**
219 * textsearch_find_continuous - search a pattern in continuous/linear data
220 * @conf: search configuration
221 * @state: search state
222 * @data: data to search in
223 * @len: length of data
224 *
225 * A simplified version of textsearch_find() for continuous/linear data.
226 * Call textsearch_next() to retrieve subsequent matches.
227 *
228 * Returns the position of first occurrence of the pattern or
229 * %UINT_MAX if no occurrence was found.
230 */
231unsigned int textsearch_find_continuous(struct ts_config *conf,
232 struct ts_state *state,
233 const void *data, unsigned int len)
234{
235 struct ts_linear_state *st = (struct ts_linear_state *) state->cb;
236
237 conf->get_next_block = get_linear_data;
238 st->data = data;
239 st->len = len;
240
241 return textsearch_find(conf, state);
242}
243EXPORT_SYMBOL(textsearch_find_continuous);
244
245/**
246 * textsearch_prepare - Prepare a search
247 * @algo: name of search algorithm
248 * @pattern: pattern data
249 * @len: length of pattern
250 * @gfp_mask: allocation mask
251 * @flags: search flags
252 *
253 * Looks up the search algorithm module and creates a new textsearch
254 * configuration for the specified pattern.
255 *
256 * Note: The format of the pattern may not be compatible between
257 * the various search algorithms.
258 *
259 * Returns a new textsearch configuration according to the specified
260 * parameters or a ERR_PTR(). If a zero length pattern is passed, this
261 * function returns EINVAL.
262 */
263struct ts_config *textsearch_prepare(const char *algo, const void *pattern,
264 unsigned int len, gfp_t gfp_mask, int flags)
265{
266 int err = -ENOENT;
267 struct ts_config *conf;
268 struct ts_ops *ops;
269
270 if (len == 0)
271 return ERR_PTR(error: -EINVAL);
272
273 ops = lookup_ts_algo(name: algo);
274#ifdef CONFIG_MODULES
275 /*
276 * Why not always autoload you may ask. Some users are
277 * in a situation where requesting a module may deadlock,
278 * especially when the module is located on a NFS mount.
279 */
280 if (ops == NULL && flags & TS_AUTOLOAD) {
281 request_module("ts_%s", algo);
282 ops = lookup_ts_algo(name: algo);
283 }
284#endif
285
286 if (ops == NULL)
287 goto errout;
288
289 conf = ops->init(pattern, len, gfp_mask, flags);
290 if (IS_ERR(ptr: conf)) {
291 err = PTR_ERR(ptr: conf);
292 goto errout;
293 }
294
295 conf->ops = ops;
296 return conf;
297
298errout:
299 if (ops)
300 module_put(module: ops->owner);
301
302 return ERR_PTR(error: err);
303}
304EXPORT_SYMBOL(textsearch_prepare);
305
306/**
307 * textsearch_destroy - destroy a search configuration
308 * @conf: search configuration
309 *
310 * Releases all references of the configuration and frees
311 * up the memory.
312 */
313void textsearch_destroy(struct ts_config *conf)
314{
315 if (conf->ops) {
316 if (conf->ops->destroy)
317 conf->ops->destroy(conf);
318 module_put(module: conf->ops->owner);
319 }
320
321 kfree(objp: conf);
322}
323EXPORT_SYMBOL(textsearch_destroy);
324

source code of linux/lib/textsearch.c