blob: ea3438e5ed540ed999ada6a4efd898064a201415 [file] [log] [blame]
William Lallemand41db4602017-10-30 11:15:51 +01001/*
2 * Cache management
3 *
4 * Copyright 2017 HAProxy Technologies
5 * William Lallemand <wlallemand@haproxy.com>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
Willy Tarreaub2551052020-06-09 09:07:15 +020013#include <import/eb32tree.h>
14#include <import/sha1.h>
15
Willy Tarreau122eba92020-06-04 10:15:32 +020016#include <haproxy/action-t.h>
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020017#include <haproxy/api.h>
Willy Tarreau6be78492020-06-05 00:00:29 +020018#include <haproxy/cfgparse.h>
Willy Tarreauf1d32c42020-06-04 21:07:02 +020019#include <haproxy/channel.h>
Willy Tarreau83487a82020-06-04 20:19:54 +020020#include <haproxy/cli.h>
Christopher Faulet908628c2022-03-25 16:43:49 +010021#include <haproxy/conn_stream.h>
22#include <haproxy/cs_utils.h>
Willy Tarreau36979d92020-06-05 17:27:29 +020023#include <haproxy/errors.h>
Willy Tarreauc7babd82020-06-04 21:29:29 +020024#include <haproxy/filters.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020025#include <haproxy/hash.h>
Remi Tricot-Le Bretondbb65b52020-10-22 10:40:04 +020026#include <haproxy/http.h>
Willy Tarreauc2b1ff02020-06-04 21:21:03 +020027#include <haproxy/http_ana.h>
Willy Tarreau87735332020-06-04 09:08:41 +020028#include <haproxy/http_htx.h>
Willy Tarreauc761f842020-06-04 11:40:28 +020029#include <haproxy/http_rules.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020030#include <haproxy/htx.h>
31#include <haproxy/net_helper.h>
Willy Tarreaua264d962020-06-04 22:29:18 +020032#include <haproxy/proxy.h>
Remi Tricot-Le Bretonbf971212020-10-27 11:55:57 +010033#include <haproxy/sample.h>
Willy Tarreau334099c2020-06-03 18:38:48 +020034#include <haproxy/shctx.h>
Willy Tarreaudfd3de82020-06-04 23:46:14 +020035#include <haproxy/stream.h>
Willy Tarreau5e539c92020-06-04 20:45:39 +020036#include <haproxy/stream_interface.h>
Willy Tarreauce6700a2021-05-08 13:03:55 +020037#include <haproxy/tools.h>
William Lallemand41db4602017-10-30 11:15:51 +010038
Christopher Faulet27d93c32018-12-15 22:32:02 +010039#define CACHE_FLT_F_IMPLICIT_DECL 0x00000001 /* The cache filtre was implicitly declared (ie without
Christopher Faulet99a17a22018-12-11 09:18:27 +010040 * the filter keyword) */
Tim Duesterhusd7c6e6a2020-09-14 18:01:33 +020041#define CACHE_FLT_INIT 0x00000002 /* Whether the cache name was freed. */
Christopher Fauletafd819c2018-12-11 08:57:45 +010042
Christopher Fauletf4a4ef72018-12-07 17:39:53 +010043const char *cache_store_flt_id = "cache store filter";
William Lallemand41db4602017-10-30 11:15:51 +010044
Willy Tarreau2231b632019-03-29 18:26:52 +010045extern struct applet http_cache_applet;
William Lallemand41db4602017-10-30 11:15:51 +010046
47struct flt_ops cache_ops;
48
49struct cache {
Willy Tarreaufd5efb52017-11-26 08:54:31 +010050 struct list list; /* cache linked list */
William Lallemand41db4602017-10-30 11:15:51 +010051 struct eb_root entries; /* head of cache entries based on keys */
Willy Tarreaufd5efb52017-11-26 08:54:31 +010052 unsigned int maxage; /* max-age */
53 unsigned int maxblocks;
Frédéric Lécaille4eba5442018-10-25 20:29:31 +020054 unsigned int maxobjsz; /* max-object-size (in bytes) */
Remi Tricot-Le Breton5853c0c2020-12-10 17:58:43 +010055 unsigned int max_secondary_entries; /* maximum number of secondary entries with the same primary hash */
Remi Tricot-Le Breton754b2422020-11-16 15:56:10 +010056 uint8_t vary_processing_enabled; /* boolean : manage Vary header (disabled by default) */
Willy Tarreaufd5efb52017-11-26 08:54:31 +010057 char id[33]; /* cache name */
William Lallemand41db4602017-10-30 11:15:51 +010058};
59
Christopher Faulet95220e22018-12-07 17:34:39 +010060/* cache config for filters */
61struct cache_flt_conf {
62 union {
63 struct cache *cache; /* cache used by the filter */
64 char *name; /* cache name used during conf parsing */
65 } c;
66 unsigned int flags; /* CACHE_FLT_F_* */
67};
68
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +010069
70/*
71 * Vary-related structures and functions
72 */
73enum vary_header_bit {
74 VARY_ACCEPT_ENCODING = (1 << 0),
75 VARY_REFERER = (1 << 1),
76 VARY_LAST /* should always be last */
77};
78
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +010079/*
80 * Encoding list extracted from
81 * https://www.iana.org/assignments/http-parameters/http-parameters.xhtml
82 * and RFC7231#5.3.4.
83 */
84enum vary_encoding {
85 VARY_ENCODING_GZIP = (1 << 0),
86 VARY_ENCODING_DEFLATE = (1 << 1),
87 VARY_ENCODING_BR = (1 << 2),
88 VARY_ENCODING_COMPRESS = (1 << 3),
89 VARY_ENCODING_AES128GCM = (1 << 4),
90 VARY_ENCODING_EXI = (1 << 5),
91 VARY_ENCODING_PACK200_GZIP = (1 << 6),
92 VARY_ENCODING_ZSTD = (1 << 7),
93 VARY_ENCODING_IDENTITY = (1 << 8),
94 VARY_ENCODING_STAR = (1 << 9),
95 VARY_ENCODING_OTHER = (1 << 10)
96};
97
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +010098struct vary_hashing_information {
99 struct ist hdr_name; /* Header name */
Ilya Shipitsinf38a0182020-12-21 01:16:17 +0500100 enum vary_header_bit value; /* Bit representing the header in a vary signature */
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100101 unsigned int hash_length; /* Size of the sub hash for this header's value */
Remi Tricot-Le Breton6a34b2b2020-12-23 18:13:47 +0100102 int(*norm_fn)(struct htx*,struct ist hdr_name,char* buf,unsigned int* buf_len); /* Normalization function */
Tim Duesterhused84d842021-01-18 13:41:17 +0100103 int(*cmp_fn)(const void *ref, const void *new, unsigned int len); /* Comparison function, should return 0 if the hashes are alike */
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100104};
105
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100106static int http_request_prebuild_full_secondary_key(struct stream *s);
107static int http_request_build_secondary_key(struct stream *s, int vary_signature);
108static int http_request_reduce_secondary_key(unsigned int vary_signature,
109 char prebuilt_key[HTTP_CACHE_SEC_KEY_LEN]);
110
111static int parse_encoding_value(struct ist value, unsigned int *encoding_value,
112 unsigned int *has_null_weight);
113
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +0100114static int accept_encoding_normalizer(struct htx *htx, struct ist hdr_name,
115 char *buf, unsigned int *buf_len);
116static int default_normalizer(struct htx *htx, struct ist hdr_name,
117 char *buf, unsigned int *buf_len);
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100118
Tim Duesterhused84d842021-01-18 13:41:17 +0100119static int accept_encoding_bitmap_cmp(const void *ref, const void *new, unsigned int len);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100120
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100121/* Warning : do not forget to update HTTP_CACHE_SEC_KEY_LEN when new items are
122 * added to this array. */
123const struct vary_hashing_information vary_information[] = {
Tim Duesterhused84d842021-01-18 13:41:17 +0100124 { IST("accept-encoding"), VARY_ACCEPT_ENCODING, sizeof(uint32_t), &accept_encoding_normalizer, &accept_encoding_bitmap_cmp },
Remi Tricot-Le Breton6a34b2b2020-12-23 18:13:47 +0100125 { IST("referer"), VARY_REFERER, sizeof(int), &default_normalizer, NULL },
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100126};
127
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100128
William Lallemand41db4602017-10-30 11:15:51 +0100129/*
130 * cache ctx for filters
131 */
132struct cache_st {
William Lallemand41db4602017-10-30 11:15:51 +0100133 struct shared_block *first_block;
134};
135
Remi Tricot-Le Breton5853c0c2020-12-10 17:58:43 +0100136#define DEFAULT_MAX_SECONDARY_ENTRY 10
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100137
William Lallemand41db4602017-10-30 11:15:51 +0100138struct cache_entry {
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +0100139 unsigned int complete; /* An entry won't be valid until complete is not null. */
William Lallemand41db4602017-10-30 11:15:51 +0100140 unsigned int latest_validation; /* latest validation date */
141 unsigned int expire; /* expiration date */
Frédéric Lécaillee7a770c2018-10-26 14:29:22 +0200142 unsigned int age; /* Origin server "Age" header value */
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100143
William Lallemand41db4602017-10-30 11:15:51 +0100144 struct eb32_node eb; /* ebtree node used to hold the cache object */
William Lallemandf528fff2017-11-23 19:43:17 +0100145 char hash[20];
Remi Tricot-Le Bretondbb65b52020-10-22 10:40:04 +0200146
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +0100147 char secondary_key[HTTP_CACHE_SEC_KEY_LEN]; /* Optional secondary key. */
148 unsigned int secondary_key_signature; /* Bitfield of the HTTP headers that should be used
149 * to build secondary keys for this cache entry. */
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100150 unsigned int secondary_entries_count; /* Should only be filled in the last entry of a list of dup entries */
Remi Tricot-Le Breton73be7962020-12-10 17:58:42 +0100151 unsigned int last_clear_ts; /* Timestamp of the last call to clear_expired_duplicates. */
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +0100152
Remi Tricot-Le Bretondbb65b52020-10-22 10:40:04 +0200153 unsigned int etag_length; /* Length of the ETag value (if one was found in the response). */
154 unsigned int etag_offset; /* Offset of the ETag value in the data buffer. */
155
Remi Tricot Le Breton27091b42020-10-23 10:51:27 +0200156 time_t last_modified; /* Origin server "Last-Modified" header value converted in
157 * seconds since epoch. If no "Last-Modified"
158 * header is found, use "Date" header value,
159 * otherwise use reception time. This field will
160 * be used in case of an "If-Modified-Since"-based
161 * conditional request. */
162
William Lallemand41db4602017-10-30 11:15:51 +0100163 unsigned char data[0];
164};
165
166#define CACHE_BLOCKSIZE 1024
Willy Tarreau96062a12018-11-11 14:00:28 +0100167#define CACHE_ENTRY_MAX_AGE 2147483648U
William Lallemand41db4602017-10-30 11:15:51 +0100168
169static struct list caches = LIST_HEAD_INIT(caches);
William Lallemandd1d1e222019-08-28 15:22:49 +0200170static struct list caches_config = LIST_HEAD_INIT(caches_config); /* cache config to init */
William Lallemand41db4602017-10-30 11:15:51 +0100171static struct cache *tmp_cache_config = NULL;
172
Willy Tarreau8ceae722018-11-26 11:58:30 +0100173DECLARE_STATIC_POOL(pool_head_cache_st, "cache_st", sizeof(struct cache_st));
174
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100175static struct eb32_node *insert_entry(struct cache *cache, struct cache_entry *new_entry);
176static void delete_entry(struct cache_entry *del_entry);
177
William Lallemandf528fff2017-11-23 19:43:17 +0100178struct cache_entry *entry_exist(struct cache *cache, char *hash)
William Lallemand4da3f8a2017-10-31 14:33:34 +0100179{
180 struct eb32_node *node;
181 struct cache_entry *entry;
182
Willy Tarreau8b507582020-02-25 09:35:07 +0100183 node = eb32_lookup(&cache->entries, read_u32(hash));
William Lallemand4da3f8a2017-10-31 14:33:34 +0100184 if (!node)
185 return NULL;
186
187 entry = eb32_entry(node, struct cache_entry, eb);
William Lallemandf528fff2017-11-23 19:43:17 +0100188
189 /* if that's not the right node */
190 if (memcmp(entry->hash, hash, sizeof(entry->hash)))
191 return NULL;
192
William Lallemand08727662017-11-21 20:01:27 +0100193 if (entry->expire > now.tv_sec) {
William Lallemand4da3f8a2017-10-31 14:33:34 +0100194 return entry;
William Lallemand08727662017-11-21 20:01:27 +0100195 } else {
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100196 delete_entry(entry);
William Lallemand08727662017-11-21 20:01:27 +0100197 entry->eb.key = 0;
198 }
William Lallemand4da3f8a2017-10-31 14:33:34 +0100199 return NULL;
200
201}
202
Remi Tricot-Le Breton6a34b2b2020-12-23 18:13:47 +0100203
204/*
205 * Compare a newly built secondary key to the one found in a cache_entry.
206 * Every sub-part of the key is compared to the reference through the dedicated
207 * comparison function of the sub-part (that might do more than a simple
208 * memcmp).
209 * Returns 0 if the keys are alike.
210 */
211static int secondary_key_cmp(const char *ref_key, const char *new_key)
212{
213 int retval = 0;
Tim Duesterhus5897cfe2021-01-18 13:41:18 +0100214 size_t idx = 0;
215 unsigned int offset = 0;
Remi Tricot-Le Breton6a34b2b2020-12-23 18:13:47 +0100216 const struct vary_hashing_information *info;
217
218 for (idx = 0; idx < sizeof(vary_information)/sizeof(*vary_information) && !retval; ++idx) {
219 info = &vary_information[idx];
220
221 if (info->cmp_fn)
222 retval = info->cmp_fn(&ref_key[offset], &new_key[offset], info->hash_length);
223 else
224 retval = memcmp(&ref_key[offset], &new_key[offset], info->hash_length);
225
226 offset += info->hash_length;
227 }
228
229 return retval;
230}
231
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +0100232/*
233 * There can be multiple entries with the same primary key in the ebtree so in
234 * order to get the proper one out of the list, we use a secondary_key.
235 * This function simply iterates over all the entries with the same primary_key
236 * until it finds the right one.
237 * Returns the cache_entry in case of success, NULL otherwise.
238 */
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100239struct cache_entry *secondary_entry_exist(struct cache *cache, struct cache_entry *entry,
Remi Tricot-Le Breton6a34b2b2020-12-23 18:13:47 +0100240 const char *secondary_key)
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100241{
242 struct eb32_node *node = &entry->eb;
243
244 if (!entry->secondary_key_signature)
245 return NULL;
246
Remi Tricot-Le Breton6a34b2b2020-12-23 18:13:47 +0100247 while (entry && secondary_key_cmp(entry->secondary_key, secondary_key) != 0) {
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100248 node = eb32_next_dup(node);
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100249
250 /* Make the best use of this iteration and clear expired entries
251 * when we find them. Calling delete_entry would be too costly
252 * so we simply call eb32_delete. The secondary_entry count will
253 * be updated when we try to insert a new entry to this list. */
254 if (entry->expire <= now.tv_sec) {
255 eb32_delete(&entry->eb);
256 entry->eb.key = 0;
257 }
258
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100259 entry = node ? eb32_entry(node, struct cache_entry, eb) : NULL;
260 }
261
262 /* Expired entry */
263 if (entry && entry->expire <= now.tv_sec) {
264 eb32_delete(&entry->eb);
265 entry->eb.key = 0;
266 entry = NULL;
267 }
268
269 return entry;
270}
271
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100272
Remi Tricot-Le Breton73be7962020-12-10 17:58:42 +0100273/*
274 * Remove all expired entries from a list of duplicates.
275 * Return the number of alive entries in the list and sets dup_tail to the
276 * current last item of the list.
277 */
278static unsigned int clear_expired_duplicates(struct eb32_node **dup_tail)
279{
280 unsigned int entry_count = 0;
281 struct cache_entry *entry = NULL;
282 struct eb32_node *prev = *dup_tail;
283 struct eb32_node *tail = NULL;
284
285 while (prev) {
286 entry = container_of(prev, struct cache_entry, eb);
287 prev = eb32_prev_dup(prev);
288 if (entry->expire <= now.tv_sec) {
289 eb32_delete(&entry->eb);
290 entry->eb.key = 0;
291 }
292 else {
293 if (!tail)
294 tail = &entry->eb;
295 ++entry_count;
296 }
297 }
298
299 *dup_tail = tail;
300
301 return entry_count;
302}
303
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100304
305/*
306 * This function inserts a cache_entry in the cache's ebtree. In case of
307 * duplicate entries (vary), it then checks that the number of entries did not
308 * reach the max number of secondary entries. If this entry should not have been
309 * created, remove it.
310 * In the regular case (unique entries), this function does not do more than a
311 * simple insert. In case of secondary entries, it will at most cost an
312 * insertion+max_sec_entries time checks and entry deletion.
313 * Returns the newly inserted node in case of success, NULL otherwise.
314 */
315static struct eb32_node *insert_entry(struct cache *cache, struct cache_entry *new_entry)
316{
317 struct eb32_node *prev = NULL;
318 struct cache_entry *entry = NULL;
319 unsigned int entry_count = 0;
Remi Tricot-Le Breton73be7962020-12-10 17:58:42 +0100320 unsigned int last_clear_ts = now.tv_sec;
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100321
322 struct eb32_node *node = eb32_insert(&cache->entries, &new_entry->eb);
323
324 /* We should not have multiple entries with the same primary key unless
325 * the entry has a non null vary signature. */
326 if (!new_entry->secondary_key_signature)
327 return node;
328
329 prev = eb32_prev_dup(node);
330 if (prev != NULL) {
331 /* The last entry of a duplicate list should contain the current
332 * number of entries in the list. */
333 entry = container_of(prev, struct cache_entry, eb);
334 entry_count = entry->secondary_entries_count;
Remi Tricot-Le Breton73be7962020-12-10 17:58:42 +0100335 last_clear_ts = entry->last_clear_ts;
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100336
Remi Tricot-Le Breton5853c0c2020-12-10 17:58:43 +0100337 if (entry_count >= cache->max_secondary_entries) {
Remi Tricot-Le Breton73be7962020-12-10 17:58:42 +0100338 /* Some entries of the duplicate list might be expired so
339 * we will iterate over all the items in order to free some
340 * space. In order to avoid going over the same list too
341 * often, we first check the timestamp of the last check
342 * performed. */
343 if (last_clear_ts == now.tv_sec) {
344 /* Too many entries for this primary key, clear the
345 * one that was inserted. */
346 eb32_delete(node);
347 node->key = 0;
348 return NULL;
349 }
350
351 entry_count = clear_expired_duplicates(&prev);
Remi Tricot-Le Breton5853c0c2020-12-10 17:58:43 +0100352 if (entry_count >= cache->max_secondary_entries) {
Remi Tricot-Le Breton73be7962020-12-10 17:58:42 +0100353 /* Still too many entries for this primary key, delete
354 * the newly inserted one. */
355 entry = container_of(prev, struct cache_entry, eb);
356 entry->last_clear_ts = now.tv_sec;
357 eb32_delete(node);
358 node->key = 0;
359 return NULL;
360 }
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100361 }
362 }
363
364 new_entry->secondary_entries_count = entry_count + 1;
Remi Tricot-Le Breton73be7962020-12-10 17:58:42 +0100365 new_entry->last_clear_ts = last_clear_ts;
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100366
367 return node;
368}
369
370
371/*
372 * This function removes an entry from the ebtree. If the entry was a duplicate
373 * (in case of Vary), it updates the secondary entry counter in another
374 * duplicate entry (the last entry of the dup list).
375 */
376static void delete_entry(struct cache_entry *del_entry)
377{
378 struct eb32_node *prev = NULL, *next = NULL;
379 struct cache_entry *entry = NULL;
380 struct eb32_node *last = NULL;
381
382 if (del_entry->secondary_key_signature) {
383 next = &del_entry->eb;
384
385 /* Look for last entry of the duplicates list. */
386 while ((next = eb32_next_dup(next))) {
387 last = next;
388 }
389
390 if (last) {
391 entry = container_of(last, struct cache_entry, eb);
392 --entry->secondary_entries_count;
393 }
394 else {
395 /* The current entry is the last one, look for the
396 * previous one to update its counter. */
397 prev = eb32_prev_dup(&del_entry->eb);
398 if (prev) {
399 entry = container_of(prev, struct cache_entry, eb);
400 entry->secondary_entries_count = del_entry->secondary_entries_count - 1;
401 }
402 }
403 }
404 eb32_delete(&del_entry->eb);
405 del_entry->eb.key = 0;
406}
407
408
William Lallemand4da3f8a2017-10-31 14:33:34 +0100409static inline struct shared_context *shctx_ptr(struct cache *cache)
410{
411 return (struct shared_context *)((unsigned char *)cache - ((struct shared_context *)NULL)->data);
412}
413
William Lallemand77c11972017-10-31 20:43:01 +0100414static inline struct shared_block *block_ptr(struct cache_entry *entry)
415{
416 return (struct shared_block *)((unsigned char *)entry - ((struct shared_block *)NULL)->data);
417}
418
419
420
William Lallemand41db4602017-10-30 11:15:51 +0100421static int
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100422cache_store_init(struct proxy *px, struct flt_conf *fconf)
William Lallemand41db4602017-10-30 11:15:51 +0100423{
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100424 fconf->flags |= FLT_CFG_FL_HTX;
William Lallemand41db4602017-10-30 11:15:51 +0100425 return 0;
426}
427
Christopher Faulet95220e22018-12-07 17:34:39 +0100428static void
429cache_store_deinit(struct proxy *px, struct flt_conf *fconf)
430{
431 struct cache_flt_conf *cconf = fconf->conf;
432
Tim Duesterhusd7c6e6a2020-09-14 18:01:33 +0200433 if (!(cconf->flags & CACHE_FLT_INIT))
434 free(cconf->c.name);
Christopher Faulet95220e22018-12-07 17:34:39 +0100435 free(cconf);
436}
437
William Lallemand4da3f8a2017-10-31 14:33:34 +0100438static int
Christopher Faulet95220e22018-12-07 17:34:39 +0100439cache_store_check(struct proxy *px, struct flt_conf *fconf)
440{
441 struct cache_flt_conf *cconf = fconf->conf;
Christopher Fauletafd819c2018-12-11 08:57:45 +0100442 struct flt_conf *f;
Christopher Faulet95220e22018-12-07 17:34:39 +0100443 struct cache *cache;
Christopher Faulet27d93c32018-12-15 22:32:02 +0100444 int comp = 0;
Christopher Faulet95220e22018-12-07 17:34:39 +0100445
William Lallemandd1d1e222019-08-28 15:22:49 +0200446 /* Find the cache corresponding to the name in the filter config. The
447 * cache will not be referenced now in the filter config because it is
448 * not fully allocated. This step will be performed during the cache
449 * post_check.
450 */
451 list_for_each_entry(cache, &caches_config, list) {
Tim Duesterhuse5ff1412021-01-02 22:31:53 +0100452 if (strcmp(cache->id, cconf->c.name) == 0)
Christopher Faulet95220e22018-12-07 17:34:39 +0100453 goto found;
Christopher Faulet95220e22018-12-07 17:34:39 +0100454 }
455
456 ha_alert("config: %s '%s': unable to find the cache '%s' referenced by the filter 'cache'.\n",
457 proxy_type_str(px), px->id, (char *)cconf->c.name);
458 return 1;
459
460 found:
Christopher Fauletafd819c2018-12-11 08:57:45 +0100461 /* Here <cache> points on the cache the filter must use and <cconf>
462 * points on the cache filter configuration. */
463
464 /* Check all filters for proxy <px> to know if the compression is
Christopher Faulet27d93c32018-12-15 22:32:02 +0100465 * enabled and if it is after the cache. When the compression is before
466 * the cache, an error is returned. Also check if the cache filter must
467 * be explicitly declaired or not. */
Christopher Fauletafd819c2018-12-11 08:57:45 +0100468 list_for_each_entry(f, &px->filter_configs, list) {
469 if (f == fconf) {
Christopher Faulet27d93c32018-12-15 22:32:02 +0100470 /* The compression filter must be evaluated after the cache. */
471 if (comp) {
472 ha_alert("config: %s '%s': unable to enable the compression filter before "
473 "the cache '%s'.\n", proxy_type_str(px), px->id, cache->id);
474 return 1;
475 }
Christopher Faulet99a17a22018-12-11 09:18:27 +0100476 }
Christopher Faulet8f7fe1c2019-07-15 15:08:25 +0200477 else if (f->id == http_comp_flt_id)
Christopher Faulet27d93c32018-12-15 22:32:02 +0100478 comp = 1;
Christopher Faulet78fbb9f2019-08-11 23:11:03 +0200479 else if (f->id == fcgi_flt_id)
480 continue;
Christopher Faulet27d93c32018-12-15 22:32:02 +0100481 else if ((f->id != fconf->id) && (cconf->flags & CACHE_FLT_F_IMPLICIT_DECL)) {
482 /* Implicit declaration is only allowed with the
Christopher Faulet78fbb9f2019-08-11 23:11:03 +0200483 * compression and fcgi. For other filters, an implicit
Christopher Faulet27d93c32018-12-15 22:32:02 +0100484 * declaration is required. */
485 ha_alert("config: %s '%s': require an explicit filter declaration "
486 "to use the cache '%s'.\n", proxy_type_str(px), px->id, cache->id);
487 return 1;
488 }
489
Christopher Fauletafd819c2018-12-11 08:57:45 +0100490 }
Christopher Faulet95220e22018-12-07 17:34:39 +0100491 return 0;
492}
493
494static int
Christopher Faulet65554e12020-03-06 14:52:06 +0100495cache_store_strm_init(struct stream *s, struct filter *filter)
William Lallemand4da3f8a2017-10-31 14:33:34 +0100496{
Christopher Faulet65554e12020-03-06 14:52:06 +0100497 struct cache_st *st;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100498
Willy Tarreauacc5b012021-03-22 15:00:49 +0100499 st = pool_alloc(pool_head_cache_st);
Christopher Faulet65554e12020-03-06 14:52:06 +0100500 if (st == NULL)
501 return -1;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100502
Christopher Faulet65554e12020-03-06 14:52:06 +0100503 st->first_block = NULL;
504 filter->ctx = st;
Christopher Faulet839791a2019-01-07 16:12:07 +0100505
Christopher Faulet65554e12020-03-06 14:52:06 +0100506 /* Register post-analyzer on AN_RES_WAIT_HTTP */
507 filter->post_analyzers |= AN_RES_WAIT_HTTP;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100508 return 1;
509}
510
Christopher Faulet65554e12020-03-06 14:52:06 +0100511static void
512cache_store_strm_deinit(struct stream *s, struct filter *filter)
William Lallemand49dc0482017-11-24 14:33:54 +0100513{
514 struct cache_st *st = filter->ctx;
Christopher Faulet95220e22018-12-07 17:34:39 +0100515 struct cache_flt_conf *cconf = FLT_CONF(filter);
516 struct cache *cache = cconf->c.cache;
William Lallemand49dc0482017-11-24 14:33:54 +0100517 struct shared_context *shctx = shctx_ptr(cache);
518
William Lallemand49dc0482017-11-24 14:33:54 +0100519 /* Everything should be released in the http_end filter, but we need to do it
520 * there too, in case of errors */
William Lallemand49dc0482017-11-24 14:33:54 +0100521 if (st && st->first_block) {
William Lallemand49dc0482017-11-24 14:33:54 +0100522 shctx_lock(shctx);
523 shctx_row_dec_hot(shctx, st->first_block);
524 shctx_unlock(shctx);
William Lallemand49dc0482017-11-24 14:33:54 +0100525 }
526 if (st) {
Willy Tarreaubafbe012017-11-24 17:34:44 +0100527 pool_free(pool_head_cache_st, st);
William Lallemand49dc0482017-11-24 14:33:54 +0100528 filter->ctx = NULL;
529 }
William Lallemand49dc0482017-11-24 14:33:54 +0100530}
531
Christopher Faulet839791a2019-01-07 16:12:07 +0100532static int
533cache_store_post_analyze(struct stream *s, struct filter *filter, struct channel *chn,
534 unsigned an_bit)
535{
536 struct http_txn *txn = s->txn;
537 struct http_msg *msg = &txn->rsp;
538 struct cache_st *st = filter->ctx;
539
540 if (an_bit != AN_RES_WAIT_HTTP)
541 goto end;
542
543 /* Here we need to check if any compression filter precedes the cache
544 * filter. This is only possible when the compression is configured in
545 * the frontend while the cache filter is configured on the
546 * backend. This case cannot be detected during HAProxy startup. So in
547 * such cases, the cache is disabled.
548 */
549 if (st && (msg->flags & HTTP_MSGF_COMPRESSING)) {
550 pool_free(pool_head_cache_st, st);
551 filter->ctx = NULL;
552 }
553
554 end:
555 return 1;
556}
William Lallemand49dc0482017-11-24 14:33:54 +0100557
558static int
William Lallemand4da3f8a2017-10-31 14:33:34 +0100559cache_store_http_headers(struct stream *s, struct filter *filter, struct http_msg *msg)
560{
561 struct cache_st *st = filter->ctx;
562
William Lallemand4da3f8a2017-10-31 14:33:34 +0100563 if (!(msg->chn->flags & CF_ISRESP) || !st)
564 return 1;
565
Christopher Faulet95e7ea32019-07-15 21:01:29 +0200566 if (st->first_block)
Christopher Faulet67658c92018-12-06 21:59:39 +0100567 register_data_filter(s, msg->chn, filter);
William Lallemand4da3f8a2017-10-31 14:33:34 +0100568 return 1;
569}
570
Frédéric Lécaille8df65ae2018-10-22 18:01:48 +0200571static inline void disable_cache_entry(struct cache_st *st,
572 struct filter *filter, struct shared_context *shctx)
573{
574 struct cache_entry *object;
575
576 object = (struct cache_entry *)st->first_block->data;
577 filter->ctx = NULL; /* disable cache */
578 shctx_lock(shctx);
579 shctx_row_dec_hot(shctx, st->first_block);
Remi Tricot-Le Breton964caaf2020-12-15 14:30:12 +0100580 eb32_delete(&object->eb);
Frédéric Lécaille8df65ae2018-10-22 18:01:48 +0200581 object->eb.key = 0;
582 shctx_unlock(shctx);
583 pool_free(pool_head_cache_st, st);
584}
585
William Lallemand4da3f8a2017-10-31 14:33:34 +0100586static int
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100587cache_store_http_payload(struct stream *s, struct filter *filter, struct http_msg *msg,
588 unsigned int offset, unsigned int len)
589{
Christopher Faulet95220e22018-12-07 17:34:39 +0100590 struct cache_flt_conf *cconf = FLT_CONF(filter);
591 struct shared_context *shctx = shctx_ptr(cconf->c.cache);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100592 struct cache_st *st = filter->ctx;
593 struct htx *htx = htxbuf(&msg->chn->buf);
594 struct htx_blk *blk;
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200595 struct shared_block *fb;
Christopher Faulet497c7592020-03-02 16:19:50 +0100596 struct htx_ret htxret;
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200597 unsigned int orig_len, to_forward;
598 int ret;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100599
600 if (!len)
601 return len;
602
603 if (!st->first_block) {
604 unregister_data_filter(s, msg->chn, filter);
605 return len;
606 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100607
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200608 chunk_reset(&trash);
609 orig_len = len;
610 to_forward = 0;
Christopher Faulet497c7592020-03-02 16:19:50 +0100611
612 htxret = htx_find_offset(htx, offset);
613 blk = htxret.blk;
614 offset = htxret.ret;
615 for (; blk && len; blk = htx_get_next_blk(htx, blk)) {
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100616 enum htx_blk_type type = htx_get_blk_type(blk);
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200617 uint32_t info, sz = htx_get_blksz(blk);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100618 struct ist v;
619
620 switch (type) {
621 case HTX_BLK_UNUSED:
622 break;
623
624 case HTX_BLK_DATA:
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100625 v = htx_get_blk_value(htx, blk);
Tim Duesterhus154374c2021-03-02 18:57:27 +0100626 v = istadv(v, offset);
Tim Duesterhus2471f5c2021-11-08 09:05:01 +0100627 v = isttrim(v, len);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100628
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200629 info = (type << 28) + v.len;
630 chunk_memcat(&trash, (char *)&info, sizeof(info));
Tim Duesterhus77508502022-03-15 13:11:06 +0100631 chunk_istcat(&trash, v);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100632 to_forward += v.len;
633 len -= v.len;
634 break;
635
636 default:
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200637 /* Here offset must always be 0 because only
638 * DATA blocks can be partially transferred. */
639 if (offset)
640 goto no_cache;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100641 if (sz > len)
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200642 goto end;
643
644 chunk_memcat(&trash, (char *)&blk->info, sizeof(blk->info));
645 chunk_memcat(&trash, htx_get_blk_ptr(htx, blk), sz);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100646 to_forward += sz;
647 len -= sz;
648 break;
649 }
650
651 offset = 0;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100652 }
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200653
654 end:
655 shctx_lock(shctx);
656 fb = shctx_row_reserve_hot(shctx, st->first_block, trash.data);
657 if (!fb) {
658 shctx_unlock(shctx);
659 goto no_cache;
660 }
661 shctx_unlock(shctx);
662
663 ret = shctx_row_data_append(shctx, st->first_block, st->first_block->last_append,
664 (unsigned char *)b_head(&trash), b_data(&trash));
665 if (ret < 0)
666 goto no_cache;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100667
668 return to_forward;
669
670 no_cache:
671 disable_cache_entry(st, filter, shctx);
672 unregister_data_filter(s, msg->chn, filter);
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200673 return orig_len;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +0100674}
675
676static int
William Lallemand4da3f8a2017-10-31 14:33:34 +0100677cache_store_http_end(struct stream *s, struct filter *filter,
678 struct http_msg *msg)
679{
680 struct cache_st *st = filter->ctx;
Christopher Faulet95220e22018-12-07 17:34:39 +0100681 struct cache_flt_conf *cconf = FLT_CONF(filter);
682 struct cache *cache = cconf->c.cache;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100683 struct shared_context *shctx = shctx_ptr(cache);
684 struct cache_entry *object;
685
686 if (!(msg->chn->flags & CF_ISRESP))
687 return 1;
688
689 if (st && st->first_block) {
690
691 object = (struct cache_entry *)st->first_block->data;
692
William Lallemand4da3f8a2017-10-31 14:33:34 +0100693 shctx_lock(shctx);
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +0100694 /* The whole payload was cached, the entry can now be used. */
695 object->complete = 1;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100696 /* remove from the hotlist */
William Lallemand4da3f8a2017-10-31 14:33:34 +0100697 shctx_row_dec_hot(shctx, st->first_block);
698 shctx_unlock(shctx);
699
700 }
701 if (st) {
Willy Tarreaubafbe012017-11-24 17:34:44 +0100702 pool_free(pool_head_cache_st, st);
William Lallemand4da3f8a2017-10-31 14:33:34 +0100703 filter->ctx = NULL;
704 }
705
706 return 1;
707}
708
709 /*
710 * This intends to be used when checking HTTP headers for some
711 * word=value directive. Return a pointer to the first character of value, if
Willy Tarreau94a01e12021-01-06 17:35:12 +0100712 * the word was not found or if there wasn't any value assigned to it return NULL
William Lallemand4da3f8a2017-10-31 14:33:34 +0100713 */
714char *directive_value(const char *sample, int slen, const char *word, int wlen)
715{
716 int st = 0;
717
718 if (slen < wlen)
719 return 0;
720
721 while (wlen) {
722 char c = *sample ^ *word;
723 if (c && c != ('A' ^ 'a'))
724 return NULL;
725 sample++;
726 word++;
727 slen--;
728 wlen--;
729 }
730
731 while (slen) {
732 if (st == 0) {
733 if (*sample != '=')
734 return NULL;
735 sample++;
736 slen--;
737 st = 1;
738 continue;
739 } else {
740 return (char *)sample;
741 }
742 }
743
744 return NULL;
745}
746
747/*
748 * Return the maxage in seconds of an HTTP response.
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100749 * The returned value will always take the cache's configuration into account
750 * (cache->maxage) but the actual max age of the response will be set in the
751 * true_maxage parameter. It will be used to determine if a response is already
752 * stale or not.
William Lallemand4da3f8a2017-10-31 14:33:34 +0100753 * Compute the maxage using either:
754 * - the assigned max-age of the cache
755 * - the s-maxage directive
756 * - the max-age directive
757 * - (Expires - Data) headers
758 * - the default-max-age of the cache
759 *
760 */
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100761int http_calc_maxage(struct stream *s, struct cache *cache, int *true_maxage)
William Lallemand4da3f8a2017-10-31 14:33:34 +0100762{
Christopher Faulet95e7ea32019-07-15 21:01:29 +0200763 struct htx *htx = htxbuf(&s->res.buf);
764 struct http_hdr_ctx ctx = { .blk = NULL };
Remi Tricot-Le Bretonfcea3742020-12-03 18:19:30 +0100765 long smaxage = -1;
766 long maxage = -1;
Remi Tricot-Le Bretona6476112020-10-28 17:52:53 +0100767 int expires = -1;
768 struct tm tm = {};
769 time_t expires_val = 0;
Remi Tricot-Le Bretonfcea3742020-12-03 18:19:30 +0100770 char *endptr = NULL;
771 int offset = 0;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100772
Remi Tricot-Le Bretonfcea3742020-12-03 18:19:30 +0100773 /* The Cache-Control max-age and s-maxage directives should be followed by
774 * a positive numerical value (see RFC 7234#5.2.1.1). According to the
775 * specs, a sender "should not" generate a quoted-string value but we will
776 * still accept this format since it isn't strictly forbidden. */
Christopher Faulet95e7ea32019-07-15 21:01:29 +0200777 while (http_find_header(htx, ist("cache-control"), &ctx, 0)) {
778 char *value;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100779
Christopher Faulet95e7ea32019-07-15 21:01:29 +0200780 value = directive_value(ctx.value.ptr, ctx.value.len, "s-maxage", 8);
781 if (value) {
782 struct buffer *chk = get_trash_chunk();
William Lallemand4da3f8a2017-10-31 14:33:34 +0100783
Willy Tarreau49b04822021-11-08 11:44:47 +0100784 chunk_memcat(chk, value, ctx.value.len - 8 + 1);
785 chunk_memcat(chk, "", 1);
Remi Tricot-Le Bretonfcea3742020-12-03 18:19:30 +0100786 offset = (*chk->area == '"') ? 1 : 0;
787 smaxage = strtol(chk->area + offset, &endptr, 10);
Willy Tarreau1f38bdb2021-11-08 12:09:27 +0100788 if (unlikely(smaxage < 0 || endptr == chk->area + offset))
Remi Tricot-Le Bretonfcea3742020-12-03 18:19:30 +0100789 return -1;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100790 }
791
Christopher Faulet95e7ea32019-07-15 21:01:29 +0200792 value = directive_value(ctx.value.ptr, ctx.value.len, "max-age", 7);
793 if (value) {
794 struct buffer *chk = get_trash_chunk();
Christopher Faulet5f2c49f2019-07-15 20:49:46 +0200795
Willy Tarreau49b04822021-11-08 11:44:47 +0100796 chunk_memcat(chk, value, ctx.value.len - 7 + 1);
797 chunk_memcat(chk, "", 1);
Remi Tricot-Le Bretonfcea3742020-12-03 18:19:30 +0100798 offset = (*chk->area == '"') ? 1 : 0;
799 maxage = strtol(chk->area + offset, &endptr, 10);
Willy Tarreau1f38bdb2021-11-08 12:09:27 +0100800 if (unlikely(maxage < 0 || endptr == chk->area + offset))
Remi Tricot-Le Bretonfcea3742020-12-03 18:19:30 +0100801 return -1;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100802 }
803 }
804
Remi Tricot-Le Bretona6476112020-10-28 17:52:53 +0100805 /* Look for Expires header if no s-maxage or max-age Cache-Control data
806 * was found. */
807 if (maxage == -1 && smaxage == -1) {
808 ctx.blk = NULL;
809 if (http_find_header(htx, ist("expires"), &ctx, 1)) {
810 if (parse_http_date(istptr(ctx.value), istlen(ctx.value), &tm)) {
811 expires_val = my_timegm(&tm);
812 /* A request having an expiring date earlier
813 * than the current date should be considered as
814 * stale. */
815 expires = (expires_val >= now.tv_sec) ?
816 (expires_val - now.tv_sec) : 0;
817 }
818 else {
819 /* Following RFC 7234#5.3, an invalid date
820 * format must be treated as a date in the past
821 * so the cache entry must be seen as already
822 * expired. */
823 expires = 0;
824 }
825 }
826 }
William Lallemand4da3f8a2017-10-31 14:33:34 +0100827
828
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100829 if (smaxage > 0) {
830 if (true_maxage)
831 *true_maxage = smaxage;
William Lallemand49b44532017-11-24 18:53:43 +0100832 return MIN(smaxage, cache->maxage);
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100833 }
William Lallemand4da3f8a2017-10-31 14:33:34 +0100834
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100835 if (maxage > 0) {
836 if (true_maxage)
837 *true_maxage = maxage;
William Lallemand49b44532017-11-24 18:53:43 +0100838 return MIN(maxage, cache->maxage);
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100839 }
William Lallemand4da3f8a2017-10-31 14:33:34 +0100840
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100841 if (expires >= 0) {
842 if (true_maxage)
843 *true_maxage = expires;
Remi Tricot-Le Bretona6476112020-10-28 17:52:53 +0100844 return MIN(expires, cache->maxage);
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100845 }
Remi Tricot-Le Bretona6476112020-10-28 17:52:53 +0100846
William Lallemand49b44532017-11-24 18:53:43 +0100847 return cache->maxage;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100848
849}
850
851
William Lallemanda400a3a2017-11-20 19:13:12 +0100852static void cache_free_blocks(struct shared_block *first, struct shared_block *block)
853{
Willy Tarreau5bd37fa2018-04-04 20:17:03 +0200854 struct cache_entry *object = (struct cache_entry *)block->data;
855
856 if (first == block && object->eb.key)
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +0100857 delete_entry(object);
Willy Tarreau5bd37fa2018-04-04 20:17:03 +0200858 object->eb.key = 0;
William Lallemanda400a3a2017-11-20 19:13:12 +0100859}
860
Remi Tricot Le Breton27091b42020-10-23 10:51:27 +0200861
862/* As per RFC 7234#4.3.2, in case of "If-Modified-Since" conditional request, the
863 * date value should be compared to a date determined by in a previous response (for
864 * the same entity). This date could either be the "Last-Modified" value, or the "Date"
865 * value of the response's reception time (by decreasing order of priority). */
866static time_t get_last_modified_time(struct htx *htx)
867{
868 time_t last_modified = 0;
869 struct http_hdr_ctx ctx = { .blk = NULL };
870 struct tm tm = {};
871
872 if (http_find_header(htx, ist("last-modified"), &ctx, 1)) {
873 if (parse_http_date(istptr(ctx.value), istlen(ctx.value), &tm)) {
874 last_modified = my_timegm(&tm);
875 }
876 }
877
878 if (!last_modified) {
879 ctx.blk = NULL;
880 if (http_find_header(htx, ist("date"), &ctx, 1)) {
881 if (parse_http_date(istptr(ctx.value), istlen(ctx.value), &tm)) {
882 last_modified = my_timegm(&tm);
883 }
884 }
885 }
886
887 /* Fallback on the current time if no "Last-Modified" or "Date" header
888 * was found. */
889 if (!last_modified)
890 last_modified = now.tv_sec;
891
892 return last_modified;
893}
894
William Lallemand41db4602017-10-30 11:15:51 +0100895/*
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100896 * Checks the vary header's value. The headers on which vary should be applied
Ilya Shipitsinf38a0182020-12-21 01:16:17 +0500897 * must be explicitly supported in the vary_information array (see cache.c). If
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100898 * any other header is mentioned, we won't store the response.
899 * Returns 1 if Vary-based storage can work, 0 otherwise.
900 */
901static int http_check_vary_header(struct htx *htx, unsigned int *vary_signature)
902{
903 unsigned int vary_idx;
904 unsigned int vary_info_count;
905 const struct vary_hashing_information *vary_info;
906 struct http_hdr_ctx ctx = { .blk = NULL };
907
908 int retval = 1;
909
910 *vary_signature = 0;
911
912 vary_info_count = sizeof(vary_information)/sizeof(*vary_information);
913 while (retval && http_find_header(htx, ist("Vary"), &ctx, 0)) {
914 for (vary_idx = 0; vary_idx < vary_info_count; ++vary_idx) {
915 vary_info = &vary_information[vary_idx];
916 if (isteqi(ctx.value, vary_info->hdr_name)) {
917 *vary_signature |= vary_info->value;
918 break;
919 }
920 }
921 retval = (vary_idx < vary_info_count);
922 }
923
924 return retval;
925}
926
927
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100928/*
929 * Look for the accept-encoding part of the secondary_key and replace the
930 * encoding bitmap part of the hash with the actual encoding of the response,
931 * extracted from the content-encoding header value.
Remi Tricot-Le Breton6ca89162021-01-07 14:50:51 +0100932 * Responses that have an unknown encoding will not be cached if they also
933 * "vary" on the accept-encoding value.
934 * Returns 0 if we found a known encoding in the response, -1 otherwise.
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100935 */
Remi Tricot-Le Breton6ca89162021-01-07 14:50:51 +0100936static int set_secondary_key_encoding(struct htx *htx, char *secondary_key)
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100937{
938 unsigned int resp_encoding_bitmap = 0;
939 const struct vary_hashing_information *info = vary_information;
940 unsigned int offset = 0;
941 unsigned int count = 0;
942 unsigned int hash_info_count = sizeof(vary_information)/sizeof(*vary_information);
943 unsigned int encoding_value;
944 struct http_hdr_ctx ctx = { .blk = NULL };
945
946 /* Look for the accept-encoding part of the secondary_key. */
947 while (count < hash_info_count && info->value != VARY_ACCEPT_ENCODING) {
948 offset += info->hash_length;
949 ++info;
950 ++count;
951 }
952
953 if (count == hash_info_count)
Remi Tricot-Le Breton6ca89162021-01-07 14:50:51 +0100954 return -1;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100955
956 while (http_find_header(htx, ist("content-encoding"), &ctx, 0)) {
Remi Tricot-Le Breton6ca89162021-01-07 14:50:51 +0100957 if (parse_encoding_value(ctx.value, &encoding_value, NULL))
958 return -1; /* Do not store responses with an unknown encoding */
959 resp_encoding_bitmap |= encoding_value;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100960 }
961
962 if (!resp_encoding_bitmap)
963 resp_encoding_bitmap |= VARY_ENCODING_IDENTITY;
964
965 /* Rewrite the bitmap part of the hash with the new bitmap that only
Ilya Shipitsinb8888ab2021-01-06 21:20:16 +0500966 * corresponds the the response's encoding. */
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100967 write_u32(secondary_key + offset, resp_encoding_bitmap);
Remi Tricot-Le Breton6ca89162021-01-07 14:50:51 +0100968
969 return 0;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +0100970}
971
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +0100972
973/*
Ilya Shipitsin6fb0f212020-04-02 15:25:26 +0500974 * This function will store the headers of the response in a buffer and then
William Lallemand41db4602017-10-30 11:15:51 +0100975 * register a filter to store the data
976 */
977enum act_return http_action_store_cache(struct act_rule *rule, struct proxy *px,
Christopher Faulet8f3c2562019-06-03 22:19:18 +0200978 struct session *sess, struct stream *s, int flags)
William Lallemand41db4602017-10-30 11:15:51 +0100979{
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +0100980 int effective_maxage = 0;
981 int true_maxage = 0;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100982 struct http_txn *txn = s->txn;
983 struct http_msg *msg = &txn->rsp;
984 struct filter *filter;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100985 struct shared_block *first = NULL;
Christopher Faulet95220e22018-12-07 17:34:39 +0100986 struct cache_flt_conf *cconf = rule->arg.act.p[0];
Remi Tricot-Le Breton754b2422020-11-16 15:56:10 +0100987 struct cache *cache = cconf->c.cache;
988 struct shared_context *shctx = shctx_ptr(cache);
Christopher Faulet839791a2019-01-07 16:12:07 +0100989 struct cache_st *cache_ctx = NULL;
990 struct cache_entry *object, *old;
Willy Tarreau8b507582020-02-25 09:35:07 +0100991 unsigned int key = read_u32(txn->cache_hash);
Christopher Faulet95e7ea32019-07-15 21:01:29 +0200992 struct htx *htx;
993 struct http_hdr_ctx ctx;
Christopher Fauletb0667472019-09-03 22:22:12 +0200994 size_t hdrs_len = 0;
Christopher Faulet95e7ea32019-07-15 21:01:29 +0200995 int32_t pos;
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +0100996 unsigned int vary_signature = 0;
William Lallemand4da3f8a2017-10-31 14:33:34 +0100997
William Lallemand4da3f8a2017-10-31 14:33:34 +0100998 /* Don't cache if the response came from a cache */
999 if ((obj_type(s->target) == OBJ_TYPE_APPLET) &&
1000 s->target == &http_cache_applet.obj_type) {
1001 goto out;
1002 }
1003
1004 /* cache only HTTP/1.1 */
1005 if (!(txn->req.flags & HTTP_MSGF_VER_11))
1006 goto out;
1007
Willy Tarreau6905d182019-10-01 17:59:17 +02001008 /* cache only GET method */
Remi Tricot-Le Breton72cffaf2020-12-03 18:19:31 +01001009 if (txn->meth != HTTP_METH_GET) {
1010 /* In case of successful unsafe method on a stored resource, the
1011 * cached entry must be invalidated (see RFC7234#4.4).
1012 * A "non-error response" is one with a 2xx (Successful) or 3xx
1013 * (Redirection) status code. */
1014 if (txn->status >= 200 && txn->status < 400) {
1015 switch (txn->meth) {
1016 case HTTP_METH_OPTIONS:
1017 case HTTP_METH_GET:
1018 case HTTP_METH_HEAD:
1019 case HTTP_METH_TRACE:
1020 break;
1021
1022 default: /* Any unsafe method */
Ilya Shipitsinf38a0182020-12-21 01:16:17 +05001023 /* Discard any corresponding entry in case of successful
Remi Tricot-Le Breton72cffaf2020-12-03 18:19:31 +01001024 * unsafe request (such as PUT, POST or DELETE). */
1025 shctx_lock(shctx);
1026
1027 old = entry_exist(cconf->c.cache, txn->cache_hash);
1028 if (old) {
1029 eb32_delete(&old->eb);
1030 old->eb.key = 0;
1031 }
1032 shctx_unlock(shctx);
1033 }
1034 }
William Lallemand4da3f8a2017-10-31 14:33:34 +01001035 goto out;
Remi Tricot-Le Breton72cffaf2020-12-03 18:19:31 +01001036 }
William Lallemand4da3f8a2017-10-31 14:33:34 +01001037
Willy Tarreauc9036c02019-01-11 19:38:25 +01001038 /* cache key was not computed */
1039 if (!key)
1040 goto out;
1041
William Lallemand4da3f8a2017-10-31 14:33:34 +01001042 /* cache only 200 status code */
1043 if (txn->status != 200)
1044 goto out;
1045
Christopher Faulet839791a2019-01-07 16:12:07 +01001046 /* Find the corresponding filter instance for the current stream */
1047 list_for_each_entry(filter, &s->strm_flt.filters, list) {
1048 if (FLT_ID(filter) == cache_store_flt_id && FLT_CONF(filter) == cconf) {
1049 /* No filter ctx, don't cache anything */
1050 if (!filter->ctx)
1051 goto out;
1052 cache_ctx = filter->ctx;
1053 break;
1054 }
1055 }
1056
1057 /* from there, cache_ctx is always defined */
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001058 htx = htxbuf(&s->res.buf);
William Lallemand4da3f8a2017-10-31 14:33:34 +01001059
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001060 /* Do not cache too big objects. */
1061 if ((msg->flags & HTTP_MSGF_CNT_LEN) && shctx->max_obj_size > 0 &&
1062 htx->data + htx->extra > shctx->max_obj_size)
1063 goto out;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001064
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001065 /* Only a subset of headers are supported in our Vary implementation. If
1066 * any other header is present in the Vary header value, we won't be
Remi Tricot-Le Breton754b2422020-11-16 15:56:10 +01001067 * able to use the cache. Likewise, if Vary header support is disabled,
1068 * avoid caching responses that contain such a header. */
1069 ctx.blk = NULL;
1070 if (cache->vary_processing_enabled) {
1071 if (!http_check_vary_header(htx, &vary_signature))
1072 goto out;
Remi Tricot-Le Breton2b5c5cb2020-12-23 18:13:45 +01001073 if (vary_signature) {
1074 /* If something went wrong during the secondary key
1075 * building, do not store the response. */
1076 if (!(txn->flags & TX_CACHE_HAS_SEC_KEY))
1077 goto out;
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001078 http_request_reduce_secondary_key(vary_signature, txn->cache_secondary_hash);
Remi Tricot-Le Breton2b5c5cb2020-12-23 18:13:45 +01001079 }
Remi Tricot-Le Breton754b2422020-11-16 15:56:10 +01001080 }
1081 else if (http_find_header(htx, ist("Vary"), &ctx, 0)) {
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001082 goto out;
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001083 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001084
Christopher Fauletfc9cfe42019-07-16 14:54:53 +02001085 http_check_response_for_cacheability(s, &s->res);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001086
Remi Tricot-Le Bretoncc9bf2e2020-11-12 11:14:41 +01001087 if (!(txn->flags & TX_CACHEABLE) || !(txn->flags & TX_CACHE_COOK) || (txn->flags & TX_CACHE_IGNORE))
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001088 goto out;
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001089
1090 shctx_lock(shctx);
1091 old = entry_exist(cache, txn->cache_hash);
1092 if (old) {
1093 if (vary_signature)
1094 old = secondary_entry_exist(cconf->c.cache, old,
1095 txn->cache_secondary_hash);
1096 if (old) {
1097 if (!old->complete) {
1098 /* An entry with the same primary key is already being
1099 * created, we should not try to store the current
1100 * response because it will waste space in the cache. */
1101 shctx_unlock(shctx);
1102 goto out;
1103 }
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +01001104 delete_entry(old);
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001105 old->eb.key = 0;
1106 }
1107 }
1108 first = shctx_row_reserve_hot(shctx, NULL, sizeof(struct cache_entry));
1109 if (!first) {
1110 shctx_unlock(shctx);
1111 goto out;
1112 }
1113 /* the received memory is not initialized, we need at least to mark
1114 * the object as not indexed yet.
1115 */
1116 object = (struct cache_entry *)first->data;
1117 memset(object, 0, sizeof(*object));
1118 object->eb.key = key;
1119 object->secondary_key_signature = vary_signature;
1120 /* We need to temporarily set a valid expiring time until the actual one
1121 * is set by the end of this function (in case of concurrent accesses to
1122 * the same resource). This way the second access will find an existing
1123 * but not yet usable entry in the tree and will avoid storing its data. */
1124 object->expire = now.tv_sec + 2;
1125
1126 memcpy(object->hash, txn->cache_hash, sizeof(object->hash));
1127 if (vary_signature)
1128 memcpy(object->secondary_key, txn->cache_secondary_hash, HTTP_CACHE_SEC_KEY_LEN);
1129
1130 /* Insert the entry in the tree even if the payload is not cached yet. */
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +01001131 if (insert_entry(cache, object) != &object->eb) {
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001132 object->eb.key = 0;
1133 shctx_unlock(shctx);
1134 goto out;
1135 }
1136 shctx_unlock(shctx);
1137
1138 /* reserve space for the cache_entry structure */
1139 first->len = sizeof(struct cache_entry);
1140 first->last_append = NULL;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001141
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +01001142 /* Determine the entry's maximum age (taking into account the cache's
1143 * configuration) as well as the response's explicit max age (extracted
1144 * from cache-control directives or the expires header). */
1145 effective_maxage = http_calc_maxage(s, cconf->c.cache, &true_maxage);
1146
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001147 ctx.blk = NULL;
1148 if (http_find_header(htx, ist("Age"), &ctx, 0)) {
Tim Duesterhusc2942842021-01-02 22:47:17 +01001149 long long hdr_age;
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001150 if (!strl2llrc(ctx.value.ptr, ctx.value.len, &hdr_age) && hdr_age > 0) {
1151 if (unlikely(hdr_age > CACHE_ENTRY_MAX_AGE))
1152 hdr_age = CACHE_ENTRY_MAX_AGE;
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +01001153 /* A response with an Age value greater than its
1154 * announced max age is stale and should not be stored. */
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001155 object->age = hdr_age;
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +01001156 if (unlikely(object->age > true_maxage))
1157 goto out;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001158 }
Remi Tricot-Le Breton51058d62020-12-03 18:19:32 +01001159 else
1160 goto out;
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001161 http_remove_header(htx, &ctx);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001162 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001163
Remi Tricot Le Breton27091b42020-10-23 10:51:27 +02001164 /* Build a last-modified time that will be stored in the cache_entry and
1165 * compared to a future If-Modified-Since client header. */
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001166 object->last_modified = get_last_modified_time(htx);
Remi Tricot Le Breton27091b42020-10-23 10:51:27 +02001167
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001168 chunk_reset(&trash);
1169 for (pos = htx_get_first(htx); pos != -1; pos = htx_get_next(htx, pos)) {
1170 struct htx_blk *blk = htx_get_blk(htx, pos);
1171 enum htx_blk_type type = htx_get_blk_type(blk);
1172 uint32_t sz = htx_get_blksz(blk);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001173
Christopher Fauletb0667472019-09-03 22:22:12 +02001174 hdrs_len += sizeof(*blk) + sz;
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001175 chunk_memcat(&trash, (char *)&blk->info, sizeof(blk->info));
1176 chunk_memcat(&trash, htx_get_blk_ptr(htx, blk), sz);
Remi Tricot-Le Bretondbb65b52020-10-22 10:40:04 +02001177
1178 /* Look for optional ETag header.
1179 * We need to store the offset of the ETag value in order for
1180 * future conditional requests to be able to perform ETag
1181 * comparisons. */
1182 if (type == HTX_BLK_HDR) {
Tim Duesterhuse2fff102021-01-02 22:47:16 +01001183 struct ist header_name = htx_get_blk_name(htx, blk);
Remi Tricot-Le Bretondbb65b52020-10-22 10:40:04 +02001184 if (isteq(header_name, ist("etag"))) {
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001185 object->etag_length = sz - istlen(header_name);
1186 object->etag_offset = sizeof(struct cache_entry) + b_data(&trash) - sz + istlen(header_name);
Remi Tricot-Le Bretondbb65b52020-10-22 10:40:04 +02001187 }
1188 }
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001189 if (type == HTX_BLK_EOH)
1190 break;
Frédéric Lécaillee7a770c2018-10-26 14:29:22 +02001191 }
1192
Christopher Fauletb0667472019-09-03 22:22:12 +02001193 /* Do not cache objects if the headers are too big. */
1194 if (hdrs_len > htx->size - global.tune.maxrewrite)
1195 goto out;
1196
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01001197 /* If the response has a secondary_key, fill its key part related to
1198 * encodings with the actual encoding of the response. This way any
1199 * subsequent request having the same primary key will have its accepted
Remi Tricot-Le Breton6ca89162021-01-07 14:50:51 +01001200 * encodings tested upon the cached response's one.
1201 * We will not cache a response that has an unknown encoding (not
Ilya Shipitsin7704b0e2021-01-23 02:11:59 +05001202 * explicitly supported in parse_encoding_value function). */
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01001203 if (cache->vary_processing_enabled && vary_signature)
Remi Tricot-Le Breton6ca89162021-01-07 14:50:51 +01001204 if (set_secondary_key_encoding(htx, object->secondary_key))
1205 goto out;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01001206
William Lallemand4da3f8a2017-10-31 14:33:34 +01001207 shctx_lock(shctx);
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001208 if (!shctx_row_reserve_hot(shctx, first, trash.data)) {
William Lallemand4da3f8a2017-10-31 14:33:34 +01001209 shctx_unlock(shctx);
1210 goto out;
1211 }
1212 shctx_unlock(shctx);
1213
William Lallemand4da3f8a2017-10-31 14:33:34 +01001214 /* cache the headers in a http action because it allows to chose what
1215 * to cache, for example you might want to cache a response before
1216 * modifying some HTTP headers, or on the contrary after modifying
1217 * those headers.
1218 */
William Lallemand4da3f8a2017-10-31 14:33:34 +01001219 /* does not need to be locked because it's in the "hot" list,
1220 * copy the headers */
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001221 if (shctx_row_data_append(shctx, first, NULL, (unsigned char *)trash.area, trash.data) < 0)
1222 goto out;
William Lallemand4da3f8a2017-10-31 14:33:34 +01001223
1224 /* register the buffer in the filter ctx for filling it with data*/
Christopher Faulet839791a2019-01-07 16:12:07 +01001225 if (cache_ctx) {
1226 cache_ctx->first_block = first;
Christopher Faulet839791a2019-01-07 16:12:07 +01001227 /* store latest value and expiration time */
1228 object->latest_validation = now.tv_sec;
Remi Tricot-Le Breton795e1412020-12-03 18:19:29 +01001229 object->expire = now.tv_sec + effective_maxage;
Christopher Faulet839791a2019-01-07 16:12:07 +01001230 return ACT_RET_CONT;
William Lallemand4da3f8a2017-10-31 14:33:34 +01001231 }
1232
1233out:
1234 /* if does not cache */
1235 if (first) {
1236 shctx_lock(shctx);
William Lallemand08727662017-11-21 20:01:27 +01001237 first->len = 0;
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001238 if (object->eb.key)
Remi Tricot-Le Breton65904e42020-12-10 17:58:41 +01001239 delete_entry(object);
William Lallemand08727662017-11-21 20:01:27 +01001240 object->eb.key = 0;
William Lallemand4da3f8a2017-10-31 14:33:34 +01001241 shctx_row_dec_hot(shctx, first);
1242 shctx_unlock(shctx);
1243 }
1244
William Lallemand41db4602017-10-30 11:15:51 +01001245 return ACT_RET_CONT;
1246}
1247
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001248#define HTX_CACHE_INIT 0 /* Initial state. */
1249#define HTX_CACHE_HEADER 1 /* Cache entry headers forwarding */
1250#define HTX_CACHE_DATA 2 /* Cache entry data forwarding */
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001251#define HTX_CACHE_EOM 3 /* Cache entry completely forwarded. Finish the HTX message */
1252#define HTX_CACHE_END 4 /* Cache entry treatment terminated */
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001253
William Lallemandecb73b12017-11-24 14:33:55 +01001254static void http_cache_applet_release(struct appctx *appctx)
1255{
Christopher Faulet95220e22018-12-07 17:34:39 +01001256 struct cache_flt_conf *cconf = appctx->rule->arg.act.p[0];
William Lallemandecb73b12017-11-24 14:33:55 +01001257 struct cache_entry *cache_ptr = appctx->ctx.cache.entry;
Christopher Faulet95220e22018-12-07 17:34:39 +01001258 struct cache *cache = cconf->c.cache;
William Lallemandecb73b12017-11-24 14:33:55 +01001259 struct shared_block *first = block_ptr(cache_ptr);
1260
1261 shctx_lock(shctx_ptr(cache));
1262 shctx_row_dec_hot(shctx_ptr(cache), first);
1263 shctx_unlock(shctx_ptr(cache));
1264}
1265
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001266
1267static unsigned int htx_cache_dump_blk(struct appctx *appctx, struct htx *htx, enum htx_blk_type type,
1268 uint32_t info, struct shared_block *shblk, unsigned int offset)
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001269{
Christopher Faulet95220e22018-12-07 17:34:39 +01001270 struct cache_flt_conf *cconf = appctx->rule->arg.act.p[0];
1271 struct shared_context *shctx = shctx_ptr(cconf->c.cache);
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001272 struct htx_blk *blk;
Christopher Faulet15a4ce82019-09-03 22:11:52 +02001273 char *ptr;
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001274 unsigned int max, total;
1275 uint32_t blksz;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001276
Christopher Faulet908628c2022-03-25 16:43:49 +01001277 max = htx_get_max_blksz(htx, channel_htx_recv_max(cs_ic(appctx->owner), htx));
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001278 if (!max)
1279 return 0;
Christopher Faulet2d7c5392019-06-03 10:41:26 +02001280 blksz = ((type == HTX_BLK_HDR || type == HTX_BLK_TLR)
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001281 ? (info & 0xff) + ((info >> 8) & 0xfffff)
1282 : info & 0xfffffff);
1283 if (blksz > max)
1284 return 0;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001285
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001286 blk = htx_add_blk(htx, type, blksz);
1287 if (!blk)
1288 return 0;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001289
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001290 blk->info = info;
1291 total = 4;
Christopher Faulet15a4ce82019-09-03 22:11:52 +02001292 ptr = htx_get_blk_ptr(htx, blk);
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001293 while (blksz) {
1294 max = MIN(blksz, shctx->block_size - offset);
Christopher Faulet15a4ce82019-09-03 22:11:52 +02001295 memcpy(ptr, (const char *)shblk->data + offset, max);
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001296 offset += max;
1297 blksz -= max;
1298 total += max;
Christopher Faulet15a4ce82019-09-03 22:11:52 +02001299 ptr += max;
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001300 if (blksz || offset == shctx->block_size) {
1301 shblk = LIST_NEXT(&shblk->list, typeof(shblk), list);
1302 offset = 0;
1303 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001304 }
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001305 appctx->ctx.cache.offset = offset;
1306 appctx->ctx.cache.next = shblk;
1307 appctx->ctx.cache.sent += total;
1308 return total;
1309}
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001310
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001311static unsigned int htx_cache_dump_data_blk(struct appctx *appctx, struct htx *htx,
1312 uint32_t info, struct shared_block *shblk, unsigned int offset)
1313{
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001314
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001315 struct cache_flt_conf *cconf = appctx->rule->arg.act.p[0];
1316 struct shared_context *shctx = shctx_ptr(cconf->c.cache);
1317 unsigned int max, total, rem_data;
1318 uint32_t blksz;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001319
Christopher Faulet908628c2022-03-25 16:43:49 +01001320 max = htx_get_max_blksz(htx, channel_htx_recv_max(cs_ic(appctx->owner), htx));
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001321 if (!max)
1322 return 0;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001323
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001324 rem_data = 0;
Christopher Fauletbda83972019-06-11 09:58:09 +02001325 if (appctx->ctx.cache.rem_data) {
1326 blksz = appctx->ctx.cache.rem_data;
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001327 total = 0;
Christopher Fauletbda83972019-06-11 09:58:09 +02001328 }
1329 else {
1330 blksz = (info & 0xfffffff);
1331 total = 4;
1332 }
1333 if (blksz > max) {
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001334 rem_data = blksz - max;
1335 blksz = max;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001336 }
1337
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001338 while (blksz) {
1339 size_t sz;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001340
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001341 max = MIN(blksz, shctx->block_size - offset);
1342 sz = htx_add_data(htx, ist2(shblk->data + offset, max));
1343 offset += sz;
1344 blksz -= sz;
1345 total += sz;
1346 if (sz < max)
1347 break;
1348 if (blksz || offset == shctx->block_size) {
1349 shblk = LIST_NEXT(&shblk->list, typeof(shblk), list);
1350 offset = 0;
1351 }
1352 }
1353
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001354 appctx->ctx.cache.offset = offset;
1355 appctx->ctx.cache.next = shblk;
1356 appctx->ctx.cache.sent += total;
1357 appctx->ctx.cache.rem_data = rem_data + blksz;
1358 return total;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001359}
1360
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001361static size_t htx_cache_dump_msg(struct appctx *appctx, struct htx *htx, unsigned int len,
1362 enum htx_blk_type mark)
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001363{
Christopher Faulet95220e22018-12-07 17:34:39 +01001364 struct cache_flt_conf *cconf = appctx->rule->arg.act.p[0];
1365 struct shared_context *shctx = shctx_ptr(cconf->c.cache);
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001366 struct shared_block *shblk;
1367 unsigned int offset, sz;
1368 unsigned int ret, total = 0;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001369
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001370 while (len) {
1371 enum htx_blk_type type;
1372 uint32_t info;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001373
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001374 shblk = appctx->ctx.cache.next;
1375 offset = appctx->ctx.cache.offset;
1376 if (appctx->ctx.cache.rem_data) {
1377 type = HTX_BLK_DATA;
1378 info = 0;
1379 goto add_data_blk;
1380 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001381
Ilya Shipitsin6fb0f212020-04-02 15:25:26 +05001382 /* Get info of the next HTX block. May be split on 2 shblk */
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001383 sz = MIN(4, shctx->block_size - offset);
1384 memcpy((char *)&info, (const char *)shblk->data + offset, sz);
1385 offset += sz;
1386 if (sz < 4) {
1387 shblk = LIST_NEXT(&shblk->list, typeof(shblk), list);
1388 memcpy(((char *)&info)+sz, (const char *)shblk->data, 4 - sz);
1389 offset = (4 - sz);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001390 }
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001391
1392 /* Get payload of the next HTX block and insert it. */
1393 type = (info >> 28);
1394 if (type != HTX_BLK_DATA)
1395 ret = htx_cache_dump_blk(appctx, htx, type, info, shblk, offset);
1396 else {
1397 add_data_blk:
1398 ret = htx_cache_dump_data_blk(appctx, htx, info, shblk, offset);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001399 }
1400
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001401 if (!ret)
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001402 break;
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001403 total += ret;
1404 len -= ret;
1405
1406 if (appctx->ctx.cache.rem_data || type == mark)
1407 break;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001408 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001409
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001410 return total;
1411}
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001412
1413static int htx_cache_add_age_hdr(struct appctx *appctx, struct htx *htx)
1414{
1415 struct cache_entry *cache_ptr = appctx->ctx.cache.entry;
1416 unsigned int age;
1417 char *end;
1418
1419 chunk_reset(&trash);
1420 age = MAX(0, (int)(now.tv_sec - cache_ptr->latest_validation)) + cache_ptr->age;
1421 if (unlikely(age > CACHE_ENTRY_MAX_AGE))
1422 age = CACHE_ENTRY_MAX_AGE;
1423 end = ultoa_o(age, b_head(&trash), b_size(&trash));
1424 b_set_data(&trash, end - b_head(&trash));
1425 if (!http_add_header(htx, ist("Age"), ist2(b_head(&trash), b_data(&trash))))
1426 return 0;
1427 return 1;
1428}
1429
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001430static void http_cache_io_handler(struct appctx *appctx)
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001431{
1432 struct cache_entry *cache_ptr = appctx->ctx.cache.entry;
1433 struct shared_block *first = block_ptr(cache_ptr);
Christopher Faulet908628c2022-03-25 16:43:49 +01001434 struct conn_stream *cs = appctx->owner;
1435 struct channel *req = cs_oc(cs);
1436 struct channel *res = cs_ic(cs);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001437 struct htx *req_htx, *res_htx;
1438 struct buffer *errmsg;
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001439 unsigned int len;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001440 size_t ret, total = 0;
1441
Christopher Faulet8b1eed12022-03-07 16:44:30 +01001442 res_htx = htx_from_buf(&res->buf);
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001443 total = res_htx->data;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001444
Christopher Faulet62e75742022-03-31 09:16:34 +02001445 if (unlikely(cs->state == CS_ST_DIS || cs->state == CS_ST_CLO))
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001446 goto out;
1447
Ilya Shipitsin6fb0f212020-04-02 15:25:26 +05001448 /* Check if the input buffer is available. */
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001449 if (!b_size(&res->buf)) {
Christopher Faulet908628c2022-03-25 16:43:49 +01001450 si_rx_room_blk(cs->si);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001451 goto out;
1452 }
1453
Willy Tarreauefef3232018-12-16 00:37:45 +01001454 if (res->flags & (CF_SHUTW|CF_SHUTR|CF_SHUTW_NOW))
Willy Tarreau273e9642018-12-16 00:35:15 +01001455 appctx->st0 = HTX_CACHE_END;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001456
1457 if (appctx->st0 == HTX_CACHE_INIT) {
1458 appctx->ctx.cache.next = block_ptr(cache_ptr);
1459 appctx->ctx.cache.offset = sizeof(*cache_ptr);
1460 appctx->ctx.cache.sent = 0;
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001461 appctx->ctx.cache.rem_data = 0;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001462 appctx->st0 = HTX_CACHE_HEADER;
1463 }
1464
1465 if (appctx->st0 == HTX_CACHE_HEADER) {
1466 /* Headers must be dump at once. Otherwise it is an error */
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001467 len = first->len - sizeof(*cache_ptr) - appctx->ctx.cache.sent;
1468 ret = htx_cache_dump_msg(appctx, res_htx, len, HTX_BLK_EOH);
1469 if (!ret || (htx_get_tail_type(res_htx) != HTX_BLK_EOH) ||
1470 !htx_cache_add_age_hdr(appctx, res_htx))
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001471 goto error;
1472
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001473 /* In case of a conditional request, we might want to send a
1474 * "304 Not Modified" response instead of the stored data. */
Tim Duesterhuse0142342020-10-22 21:15:06 +02001475 if (appctx->ctx.cache.send_notmodified) {
1476 if (!http_replace_res_status(res_htx, ist("304"), ist("Not Modified"))) {
1477 /* If replacing the status code fails we need to send the full response. */
1478 appctx->ctx.cache.send_notmodified = 0;
1479 }
1480 }
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001481
1482 /* Skip response body for HEAD requests or in case of "304 Not
1483 * Modified" response. */
Christopher Faulet908628c2022-03-25 16:43:49 +01001484 if (__cs_strm(cs)->txn->meth == HTTP_METH_HEAD || appctx->ctx.cache.send_notmodified)
Christopher Fauletf0dd0372019-02-25 11:08:34 +01001485 appctx->st0 = HTX_CACHE_EOM;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001486 else
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001487 appctx->st0 = HTX_CACHE_DATA;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001488 }
1489
1490 if (appctx->st0 == HTX_CACHE_DATA) {
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001491 len = first->len - sizeof(*cache_ptr) - appctx->ctx.cache.sent;
1492 if (len) {
Christopher Fauletd1ac2b92020-12-02 19:12:22 +01001493 ret = htx_cache_dump_msg(appctx, res_htx, len, HTX_BLK_UNUSED);
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001494 if (ret < len) {
Christopher Faulet908628c2022-03-25 16:43:49 +01001495 si_rx_room_blk(cs->si);
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001496 goto out;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001497 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001498 }
Christopher Fauletd1ac2b92020-12-02 19:12:22 +01001499 appctx->st0 = HTX_CACHE_EOM;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001500 }
1501
1502 if (appctx->st0 == HTX_CACHE_EOM) {
Christopher Fauletd1ac2b92020-12-02 19:12:22 +01001503 /* no more data are expected. */
1504 res_htx->flags |= HTX_FL_EOM;
Christopher Faulet908628c2022-03-25 16:43:49 +01001505 cs->endp->flags |= CS_EP_EOI;
Christopher Fauletdbf1e882022-03-07 15:53:57 +01001506 res->flags |= CF_EOI;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001507 appctx->st0 = HTX_CACHE_END;
1508 }
1509
1510 end:
Christopher Fauletadb36312019-02-25 11:40:49 +01001511 if (!(res->flags & CF_SHUTR) && appctx->st0 == HTX_CACHE_END) {
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001512 res->flags |= CF_READ_NULL;
Christopher Fauletda098e62022-03-31 17:44:45 +02001513 cs_shutr(cs);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001514 }
1515
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001516 out:
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001517 total = res_htx->data - total;
Christopher Faulet61123912019-01-02 14:10:01 +01001518 if (total)
1519 channel_add_input(res, total);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001520 htx_to_buf(res_htx, &res->buf);
Christopher Fauletadb36312019-02-25 11:40:49 +01001521
1522 /* eat the whole request */
1523 if (co_data(req)) {
1524 req_htx = htx_from_buf(&req->buf);
1525 co_htx_skip(req, req_htx, co_data(req));
1526 htx_to_buf(req_htx, &req->buf);
1527 }
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001528 return;
1529
1530 error:
1531 /* Sent and HTTP error 500 */
1532 b_reset(&res->buf);
Christopher Fauletf7346382019-07-17 22:02:08 +02001533 errmsg = &http_err_chunks[HTTP_ERR_500];
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001534 res->buf.data = b_data(errmsg);
1535 memcpy(res->buf.area, b_head(errmsg), b_data(errmsg));
1536 res_htx = htx_from_buf(&res->buf);
1537
Christopher Faulet8f3c2562019-06-03 22:19:18 +02001538 total = 0;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001539 appctx->st0 = HTX_CACHE_END;
1540 goto end;
1541}
1542
1543
Christopher Faulet95220e22018-12-07 17:34:39 +01001544static int parse_cache_rule(struct proxy *proxy, const char *name, struct act_rule *rule, char **err)
William Lallemand41db4602017-10-30 11:15:51 +01001545{
1546 struct flt_conf *fconf;
Christopher Faulet95220e22018-12-07 17:34:39 +01001547 struct cache_flt_conf *cconf = NULL;
William Lallemand41db4602017-10-30 11:15:51 +01001548
Christopher Faulet95220e22018-12-07 17:34:39 +01001549 if (!*name || strcmp(name, "if") == 0 || strcmp(name, "unless") == 0) {
William Lallemand41db4602017-10-30 11:15:51 +01001550 memprintf(err, "expects a cache name");
Christopher Faulet95220e22018-12-07 17:34:39 +01001551 goto err;
William Lallemand41db4602017-10-30 11:15:51 +01001552 }
1553
1554 /* check if a cache filter was already registered with this cache
1555 * name, if that's the case, must use it. */
1556 list_for_each_entry(fconf, &proxy->filter_configs, list) {
Christopher Faulet95220e22018-12-07 17:34:39 +01001557 if (fconf->id == cache_store_flt_id) {
1558 cconf = fconf->conf;
Tim Duesterhuse5ff1412021-01-02 22:31:53 +01001559 if (cconf && strcmp((char *)cconf->c.name, name) == 0) {
Christopher Faulet95220e22018-12-07 17:34:39 +01001560 rule->arg.act.p[0] = cconf;
1561 return 1;
1562 }
William Lallemand41db4602017-10-30 11:15:51 +01001563 }
1564 }
1565
Christopher Faulet95220e22018-12-07 17:34:39 +01001566 /* Create the filter cache config */
1567 cconf = calloc(1, sizeof(*cconf));
1568 if (!cconf) {
1569 memprintf(err, "out of memory\n");
1570 goto err;
1571 }
Christopher Faulet99a17a22018-12-11 09:18:27 +01001572 cconf->flags = CACHE_FLT_F_IMPLICIT_DECL;
Christopher Faulet95220e22018-12-07 17:34:39 +01001573 cconf->c.name = strdup(name);
1574 if (!cconf->c.name) {
1575 memprintf(err, "out of memory\n");
William Lallemand41db4602017-10-30 11:15:51 +01001576 goto err;
1577 }
Christopher Faulet95220e22018-12-07 17:34:39 +01001578
William Lallemand41db4602017-10-30 11:15:51 +01001579 /* register a filter to fill the cache buffer */
1580 fconf = calloc(1, sizeof(*fconf));
1581 if (!fconf) {
Christopher Faulet95220e22018-12-07 17:34:39 +01001582 memprintf(err, "out of memory\n");
William Lallemand41db4602017-10-30 11:15:51 +01001583 goto err;
1584 }
Christopher Faulet95220e22018-12-07 17:34:39 +01001585 fconf->id = cache_store_flt_id;
1586 fconf->conf = cconf;
William Lallemand41db4602017-10-30 11:15:51 +01001587 fconf->ops = &cache_ops;
Willy Tarreau2b718102021-04-21 07:32:39 +02001588 LIST_APPEND(&proxy->filter_configs, &fconf->list);
William Lallemand41db4602017-10-30 11:15:51 +01001589
Christopher Faulet95220e22018-12-07 17:34:39 +01001590 rule->arg.act.p[0] = cconf;
1591 return 1;
William Lallemand41db4602017-10-30 11:15:51 +01001592
Christopher Faulet95220e22018-12-07 17:34:39 +01001593 err:
1594 free(cconf);
1595 return 0;
1596}
1597
1598enum act_parse_ret parse_cache_store(const char **args, int *orig_arg, struct proxy *proxy,
1599 struct act_rule *rule, char **err)
1600{
1601 rule->action = ACT_CUSTOM;
1602 rule->action_ptr = http_action_store_cache;
1603
1604 if (!parse_cache_rule(proxy, args[*orig_arg], rule, err))
1605 return ACT_RET_PRS_ERR;
William Lallemand41db4602017-10-30 11:15:51 +01001606
Christopher Faulet95220e22018-12-07 17:34:39 +01001607 (*orig_arg)++;
1608 return ACT_RET_PRS_OK;
William Lallemand41db4602017-10-30 11:15:51 +01001609}
1610
Baptiste Assmanndb92a832019-08-05 16:55:32 +02001611/* This produces a sha1 hash of the concatenation of the HTTP method,
1612 * the first occurrence of the Host header followed by the path component
1613 * if it begins with a slash ('/'). */
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001614int sha1_hosturi(struct stream *s)
William Lallemandf528fff2017-11-23 19:43:17 +01001615{
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001616 struct http_txn *txn = s->txn;
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001617 struct htx *htx = htxbuf(&s->req.buf);
1618 struct htx_sl *sl;
1619 struct http_hdr_ctx ctx;
Willy Tarreauccc61d82019-10-17 09:28:28 +02001620 struct ist uri;
William Lallemandf528fff2017-11-23 19:43:17 +01001621 blk_SHA_CTX sha1_ctx;
Willy Tarreau83061a82018-07-13 11:56:34 +02001622 struct buffer *trash;
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001623
William Lallemandf528fff2017-11-23 19:43:17 +01001624 trash = get_trash_chunk();
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001625 ctx.blk = NULL;
Baptiste Assmanndb92a832019-08-05 16:55:32 +02001626
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001627 sl = http_get_stline(htx);
Willy Tarreauccc61d82019-10-17 09:28:28 +02001628 uri = htx_sl_req_uri(sl); // whole uri
1629 if (!uri.len)
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001630 return 0;
Willy Tarreauccc61d82019-10-17 09:28:28 +02001631
1632 /* In HTTP/1, most URIs are seen in origin form ('/path/to/resource'),
1633 * unless haproxy is deployed in front of an outbound cache. In HTTP/2,
1634 * URIs are almost always sent in absolute form with their scheme. In
1635 * this case, the scheme is almost always "https". In order to support
1636 * sharing of cache objects between H1 and H2, we'll hash the absolute
1637 * URI whenever known, or prepend "https://" + the Host header for
1638 * relative URIs. The difference will only appear on absolute HTTP/1
1639 * requests sent to an origin server, which practically is never met in
1640 * the real world so we don't care about the ability to share the same
1641 * key here.URIs are normalized from the absolute URI to an origin form as
1642 * well.
1643 */
1644 if (!(sl->flags & HTX_SL_F_HAS_AUTHORITY)) {
Willy Tarreau20020ae2019-10-29 13:02:15 +01001645 chunk_istcat(trash, ist("https://"));
Willy Tarreauccc61d82019-10-17 09:28:28 +02001646 if (!http_find_header(htx, ist("Host"), &ctx, 0))
1647 return 0;
Willy Tarreau20020ae2019-10-29 13:02:15 +01001648 chunk_istcat(trash, ctx.value);
Willy Tarreauccc61d82019-10-17 09:28:28 +02001649 }
1650
Tim Duesterhus9f7ed8a2021-11-08 09:05:04 +01001651 chunk_istcat(trash, uri);
William Lallemandf528fff2017-11-23 19:43:17 +01001652
1653 /* hash everything */
1654 blk_SHA1_Init(&sha1_ctx);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001655 blk_SHA1_Update(&sha1_ctx, trash->area, trash->data);
William Lallemandf528fff2017-11-23 19:43:17 +01001656 blk_SHA1_Final((unsigned char *)txn->cache_hash, &sha1_ctx);
1657
1658 return 1;
1659}
1660
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001661/* Looks for "If-None-Match" headers in the request and compares their value
1662 * with the one that might have been stored in the cache_entry. If any of them
1663 * matches, a "304 Not Modified" response should be sent instead of the cached
1664 * data.
1665 * Although unlikely in a GET/HEAD request, the "If-None-Match: *" syntax is
Remi Tricot-Le Breton53161d82020-10-23 10:51:28 +02001666 * valid and should receive a "304 Not Modified" response (RFC 7234#4.3.2).
1667 *
1668 * If no "If-None-Match" header was found, look for an "If-Modified-Since"
1669 * header and compare its value (date) to the one stored in the cache_entry.
1670 * If the request's date is later than the cached one, we also send a
1671 * "304 Not Modified" response (see RFCs 7232#3.3 and 7234#4.3.2).
1672 *
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001673 * Returns 1 if "304 Not Modified" should be sent, 0 otherwise.
1674 */
1675static int should_send_notmodified_response(struct cache *cache, struct htx *htx,
1676 struct cache_entry *entry)
1677{
1678 int retval = 0;
1679
1680 struct http_hdr_ctx ctx = { .blk = NULL };
1681 struct ist cache_entry_etag = IST_NULL;
1682 struct buffer *etag_buffer = NULL;
Remi Tricot-Le Breton53161d82020-10-23 10:51:28 +02001683 int if_none_match_found = 0;
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001684
Remi Tricot-Le Breton53161d82020-10-23 10:51:28 +02001685 struct tm tm = {};
1686 time_t if_modified_since = 0;
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001687
1688 /* If we find a "If-None-Match" header in the request, rebuild the
Remi Tricot-Le Breton53161d82020-10-23 10:51:28 +02001689 * cache_entry's ETag in order to perform comparisons.
1690 * There could be multiple "if-none-match" header lines. */
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001691 while (http_find_header(htx, ist("if-none-match"), &ctx, 0)) {
Remi Tricot-Le Breton53161d82020-10-23 10:51:28 +02001692 if_none_match_found = 1;
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001693
1694 /* A '*' matches everything. */
1695 if (isteq(ctx.value, ist("*")) != 0) {
1696 retval = 1;
1697 break;
1698 }
1699
Remi Tricot-Le Breton53161d82020-10-23 10:51:28 +02001700 /* No need to rebuild an etag if none was stored in the cache. */
1701 if (entry->etag_length == 0)
1702 break;
1703
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001704 /* Rebuild the stored ETag. */
1705 if (etag_buffer == NULL) {
1706 etag_buffer = get_trash_chunk();
1707
1708 if (shctx_row_data_get(shctx_ptr(cache), block_ptr(entry),
1709 (unsigned char*)b_orig(etag_buffer),
1710 entry->etag_offset, entry->etag_length) == 0) {
1711 cache_entry_etag = ist2(b_orig(etag_buffer), entry->etag_length);
1712 } else {
1713 /* We could not rebuild the ETag in one go, we
1714 * won't send a "304 Not Modified" response. */
1715 break;
1716 }
1717 }
1718
1719 if (http_compare_etags(cache_entry_etag, ctx.value) == 1) {
1720 retval = 1;
1721 break;
1722 }
1723 }
1724
Remi Tricot-Le Breton53161d82020-10-23 10:51:28 +02001725 /* If the request did not contain an "If-None-Match" header, we look for
1726 * an "If-Modified-Since" header (see RFC 7232#3.3). */
1727 if (retval == 0 && if_none_match_found == 0) {
1728 ctx.blk = NULL;
1729 if (http_find_header(htx, ist("if-modified-since"), &ctx, 1)) {
1730 if (parse_http_date(istptr(ctx.value), istlen(ctx.value), &tm)) {
1731 if_modified_since = my_timegm(&tm);
1732
1733 /* We send a "304 Not Modified" response if the
1734 * entry's last modified date is earlier than
1735 * the one found in the "If-Modified-Since"
1736 * header. */
1737 retval = (entry->last_modified <= if_modified_since);
1738 }
1739 }
1740 }
1741
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001742 return retval;
1743}
1744
William Lallemand41db4602017-10-30 11:15:51 +01001745enum act_return http_action_req_cache_use(struct act_rule *rule, struct proxy *px,
1746 struct session *sess, struct stream *s, int flags)
1747{
William Lallemand77c11972017-10-31 20:43:01 +01001748
Christopher Fauletb3d4bca2019-02-25 10:59:33 +01001749 struct http_txn *txn = s->txn;
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001750 struct cache_entry *res, *sec_entry = NULL;
Christopher Faulet95220e22018-12-07 17:34:39 +01001751 struct cache_flt_conf *cconf = rule->arg.act.p[0];
1752 struct cache *cache = cconf->c.cache;
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001753 struct shared_block *entry_block;
1754
William Lallemand77c11972017-10-31 20:43:01 +01001755
Willy Tarreau6905d182019-10-01 17:59:17 +02001756 /* Ignore cache for HTTP/1.0 requests and for requests other than GET
1757 * and HEAD */
Christopher Fauletb3d4bca2019-02-25 10:59:33 +01001758 if (!(txn->req.flags & HTTP_MSGF_VER_11) ||
Willy Tarreau6905d182019-10-01 17:59:17 +02001759 (txn->meth != HTTP_METH_GET && txn->meth != HTTP_METH_HEAD))
Christopher Fauletb3d4bca2019-02-25 10:59:33 +01001760 txn->flags |= TX_CACHE_IGNORE;
1761
Christopher Fauletfc9cfe42019-07-16 14:54:53 +02001762 http_check_request_for_cacheability(s, &s->req);
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01001763
Remi Tricot-Le Breton72cffaf2020-12-03 18:19:31 +01001764 /* The request's hash has to be calculated for all requests, even POSTs
Ilya Shipitsinf38a0182020-12-21 01:16:17 +05001765 * or PUTs for instance because RFC7234 specifies that a successful
Remi Tricot-Le Breton72cffaf2020-12-03 18:19:31 +01001766 * "unsafe" method on a stored resource must invalidate it
1767 * (see RFC7234#4.4). */
1768 if (!sha1_hosturi(s))
Willy Tarreau504455c2017-12-22 17:47:35 +01001769 return ACT_RET_CONT;
1770
Willy Tarreau504455c2017-12-22 17:47:35 +01001771 if (s->txn->flags & TX_CACHE_IGNORE)
1772 return ACT_RET_CONT;
1773
Willy Tarreaua1214a52018-12-14 14:00:25 +01001774 if (px == strm_fe(s))
Willy Tarreau4781b152021-04-06 13:53:36 +02001775 _HA_ATOMIC_INC(&px->fe_counters.p.http.cache_lookups);
Willy Tarreaua1214a52018-12-14 14:00:25 +01001776 else
Willy Tarreau4781b152021-04-06 13:53:36 +02001777 _HA_ATOMIC_INC(&px->be_counters.p.http.cache_lookups);
Willy Tarreaua1214a52018-12-14 14:00:25 +01001778
William Lallemanda400a3a2017-11-20 19:13:12 +01001779 shctx_lock(shctx_ptr(cache));
William Lallemandf528fff2017-11-23 19:43:17 +01001780 res = entry_exist(cache, s->txn->cache_hash);
Remi Tricot-Le Breton32434472020-11-25 10:09:43 +01001781 /* We must not use an entry that is not complete. */
1782 if (res && res->complete) {
William Lallemand77c11972017-10-31 20:43:01 +01001783 struct appctx *appctx;
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001784 entry_block = block_ptr(res);
1785 shctx_row_inc_hot(shctx_ptr(cache), entry_block);
William Lallemanda400a3a2017-11-20 19:13:12 +01001786 shctx_unlock(shctx_ptr(cache));
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001787
1788 /* In case of Vary, we could have multiple entries with the same
Remi Tricot-Le Breton2b5c5cb2020-12-23 18:13:45 +01001789 * primary hash. We need to calculate the secondary hash in order
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001790 * to find the actual entry we want (if it exists). */
1791 if (res->secondary_key_signature) {
1792 if (!http_request_build_secondary_key(s, res->secondary_key_signature)) {
1793 shctx_lock(shctx_ptr(cache));
1794 sec_entry = secondary_entry_exist(cache, res,
1795 s->txn->cache_secondary_hash);
1796 if (sec_entry && sec_entry != res) {
1797 /* The wrong row was added to the hot list. */
1798 shctx_row_dec_hot(shctx_ptr(cache), entry_block);
1799 entry_block = block_ptr(sec_entry);
1800 shctx_row_inc_hot(shctx_ptr(cache), entry_block);
1801 }
1802 res = sec_entry;
1803 shctx_unlock(shctx_ptr(cache));
1804 }
1805 else
1806 res = NULL;
1807 }
1808
1809 /* We looked for a valid secondary entry and could not find one,
1810 * the request must be forwarded to the server. */
1811 if (!res) {
1812 shctx_lock(shctx_ptr(cache));
1813 shctx_row_dec_hot(shctx_ptr(cache), entry_block);
1814 shctx_unlock(shctx_ptr(cache));
1815 return ACT_RET_CONT;
1816 }
1817
William Lallemand77c11972017-10-31 20:43:01 +01001818 s->target = &http_cache_applet.obj_type;
Christopher Faulet7a58d792021-12-23 13:40:08 +01001819 if ((appctx = si_register_handler(cs_si(s->csb), objt_applet(s->target)))) {
Christopher Faulet95e7ea32019-07-15 21:01:29 +02001820 appctx->st0 = HTX_CACHE_INIT;
William Lallemand77c11972017-10-31 20:43:01 +01001821 appctx->rule = rule;
1822 appctx->ctx.cache.entry = res;
Frédéric Lécaille8df65ae2018-10-22 18:01:48 +02001823 appctx->ctx.cache.next = NULL;
1824 appctx->ctx.cache.sent = 0;
Remi Tricot-Le Breton6cb10382020-10-22 10:40:05 +02001825 appctx->ctx.cache.send_notmodified =
1826 should_send_notmodified_response(cache, htxbuf(&s->req.buf), res);
Willy Tarreaua1214a52018-12-14 14:00:25 +01001827
1828 if (px == strm_fe(s))
Willy Tarreau4781b152021-04-06 13:53:36 +02001829 _HA_ATOMIC_INC(&px->fe_counters.p.http.cache_hits);
Willy Tarreaua1214a52018-12-14 14:00:25 +01001830 else
Willy Tarreau4781b152021-04-06 13:53:36 +02001831 _HA_ATOMIC_INC(&px->be_counters.p.http.cache_hits);
Olivier Houchardfccf8402017-11-01 14:04:02 +01001832 return ACT_RET_CONT;
William Lallemand77c11972017-10-31 20:43:01 +01001833 } else {
William Lallemand55e76742017-11-21 20:01:28 +01001834 shctx_lock(shctx_ptr(cache));
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001835 shctx_row_dec_hot(shctx_ptr(cache), entry_block);
William Lallemand55e76742017-11-21 20:01:28 +01001836 shctx_unlock(shctx_ptr(cache));
Olivier Houchardfccf8402017-11-01 14:04:02 +01001837 return ACT_RET_YIELD;
William Lallemand77c11972017-10-31 20:43:01 +01001838 }
1839 }
William Lallemanda400a3a2017-11-20 19:13:12 +01001840 shctx_unlock(shctx_ptr(cache));
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001841
1842 /* Shared context does not need to be locked while we calculate the
1843 * secondary hash. */
Remi Tricot-Le Breton754b2422020-11-16 15:56:10 +01001844 if (!res && cache->vary_processing_enabled) {
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01001845 /* Build a complete secondary hash until the server response
1846 * tells us which fields should be kept (if any). */
1847 http_request_prebuild_full_secondary_key(s);
1848 }
Olivier Houchardfccf8402017-11-01 14:04:02 +01001849 return ACT_RET_CONT;
William Lallemand41db4602017-10-30 11:15:51 +01001850}
1851
1852
1853enum act_parse_ret parse_cache_use(const char **args, int *orig_arg, struct proxy *proxy,
1854 struct act_rule *rule, char **err)
1855{
William Lallemand41db4602017-10-30 11:15:51 +01001856 rule->action = ACT_CUSTOM;
1857 rule->action_ptr = http_action_req_cache_use;
1858
Christopher Faulet95220e22018-12-07 17:34:39 +01001859 if (!parse_cache_rule(proxy, args[*orig_arg], rule, err))
William Lallemand41db4602017-10-30 11:15:51 +01001860 return ACT_RET_PRS_ERR;
William Lallemand41db4602017-10-30 11:15:51 +01001861
1862 (*orig_arg)++;
1863 return ACT_RET_PRS_OK;
William Lallemand41db4602017-10-30 11:15:51 +01001864}
1865
1866int cfg_parse_cache(const char *file, int linenum, char **args, int kwm)
1867{
1868 int err_code = 0;
1869
1870 if (strcmp(args[0], "cache") == 0) { /* new cache section */
1871
1872 if (!*args[1]) {
Tim Duesterhusff4d86b2020-08-18 22:20:27 +02001873 ha_alert("parsing [%s:%d] : '%s' expects a <name> argument\n",
Christopher Faulet767a84b2017-11-24 16:50:31 +01001874 file, linenum, args[0]);
William Lallemand41db4602017-10-30 11:15:51 +01001875 err_code |= ERR_ALERT | ERR_ABORT;
1876 goto out;
1877 }
1878
1879 if (alertif_too_many_args(1, file, linenum, args, &err_code)) {
1880 err_code |= ERR_ABORT;
1881 goto out;
1882 }
1883
1884 if (tmp_cache_config == NULL) {
Tim Duesterhusff4d86b2020-08-18 22:20:27 +02001885 struct cache *cache_config;
1886
William Lallemand41db4602017-10-30 11:15:51 +01001887 tmp_cache_config = calloc(1, sizeof(*tmp_cache_config));
1888 if (!tmp_cache_config) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001889 ha_alert("parsing [%s:%d]: out of memory.\n", file, linenum);
William Lallemand41db4602017-10-30 11:15:51 +01001890 err_code |= ERR_ALERT | ERR_ABORT;
1891 goto out;
1892 }
1893
1894 strlcpy2(tmp_cache_config->id, args[1], 33);
1895 if (strlen(args[1]) > 32) {
Tim Duesterhusff4d86b2020-08-18 22:20:27 +02001896 ha_warning("parsing [%s:%d]: cache name is limited to 32 characters, truncate to '%s'.\n",
Christopher Faulet767a84b2017-11-24 16:50:31 +01001897 file, linenum, tmp_cache_config->id);
William Lallemand41db4602017-10-30 11:15:51 +01001898 err_code |= ERR_WARN;
1899 }
Tim Duesterhusff4d86b2020-08-18 22:20:27 +02001900
1901 list_for_each_entry(cache_config, &caches_config, list) {
1902 if (strcmp(tmp_cache_config->id, cache_config->id) == 0) {
1903 ha_alert("parsing [%s:%d]: Duplicate cache name '%s'.\n",
1904 file, linenum, tmp_cache_config->id);
1905 err_code |= ERR_ALERT | ERR_ABORT;
1906 goto out;
1907 }
1908 }
1909
William Lallemand49b44532017-11-24 18:53:43 +01001910 tmp_cache_config->maxage = 60;
William Lallemand41db4602017-10-30 11:15:51 +01001911 tmp_cache_config->maxblocks = 0;
Frédéric Lécaillea2219f52018-10-22 16:59:13 +02001912 tmp_cache_config->maxobjsz = 0;
Remi Tricot-Le Breton5853c0c2020-12-10 17:58:43 +01001913 tmp_cache_config->max_secondary_entries = DEFAULT_MAX_SECONDARY_ENTRY;
William Lallemand41db4602017-10-30 11:15:51 +01001914 }
1915 } else if (strcmp(args[0], "total-max-size") == 0) {
Frédéric Lécailleb9b8b6b2018-10-25 20:17:45 +02001916 unsigned long int maxsize;
1917 char *err;
William Lallemand41db4602017-10-30 11:15:51 +01001918
1919 if (alertif_too_many_args(1, file, linenum, args, &err_code)) {
1920 err_code |= ERR_ABORT;
1921 goto out;
1922 }
1923
Frédéric Lécailleb9b8b6b2018-10-25 20:17:45 +02001924 maxsize = strtoul(args[1], &err, 10);
1925 if (err == args[1] || *err != '\0') {
1926 ha_warning("parsing [%s:%d]: total-max-size wrong value '%s'\n",
1927 file, linenum, args[1]);
1928 err_code |= ERR_ABORT;
1929 goto out;
1930 }
1931
1932 if (maxsize > (UINT_MAX >> 20)) {
1933 ha_warning("parsing [%s:%d]: \"total-max-size\" (%s) must not be greater than %u\n",
1934 file, linenum, args[1], UINT_MAX >> 20);
1935 err_code |= ERR_ABORT;
1936 goto out;
1937 }
1938
William Lallemand41db4602017-10-30 11:15:51 +01001939 /* size in megabytes */
Frédéric Lécailleb9b8b6b2018-10-25 20:17:45 +02001940 maxsize *= 1024 * 1024 / CACHE_BLOCKSIZE;
William Lallemand41db4602017-10-30 11:15:51 +01001941 tmp_cache_config->maxblocks = maxsize;
William Lallemand49b44532017-11-24 18:53:43 +01001942 } else if (strcmp(args[0], "max-age") == 0) {
1943 if (alertif_too_many_args(1, file, linenum, args, &err_code)) {
1944 err_code |= ERR_ABORT;
1945 goto out;
1946 }
1947
1948 if (!*args[1]) {
1949 ha_warning("parsing [%s:%d]: '%s' expects an age parameter in seconds.\n",
1950 file, linenum, args[0]);
1951 err_code |= ERR_WARN;
1952 }
1953
1954 tmp_cache_config->maxage = atoi(args[1]);
Frédéric Lécaillea2219f52018-10-22 16:59:13 +02001955 } else if (strcmp(args[0], "max-object-size") == 0) {
Frédéric Lécaille4eba5442018-10-25 20:29:31 +02001956 unsigned int maxobjsz;
1957 char *err;
1958
Frédéric Lécaillea2219f52018-10-22 16:59:13 +02001959 if (alertif_too_many_args(1, file, linenum, args, &err_code)) {
1960 err_code |= ERR_ABORT;
1961 goto out;
1962 }
1963
1964 if (!*args[1]) {
1965 ha_warning("parsing [%s:%d]: '%s' expects a maximum file size parameter in bytes.\n",
1966 file, linenum, args[0]);
1967 err_code |= ERR_WARN;
1968 }
1969
Frédéric Lécaille4eba5442018-10-25 20:29:31 +02001970 maxobjsz = strtoul(args[1], &err, 10);
1971 if (err == args[1] || *err != '\0') {
1972 ha_warning("parsing [%s:%d]: max-object-size wrong value '%s'\n",
1973 file, linenum, args[1]);
1974 err_code |= ERR_ABORT;
1975 goto out;
1976 }
1977 tmp_cache_config->maxobjsz = maxobjsz;
Remi Tricot-Le Breton754b2422020-11-16 15:56:10 +01001978 } else if (strcmp(args[0], "process-vary") == 0) {
1979 if (alertif_too_many_args(1, file, linenum, args, &err_code)) {
1980 err_code |= ERR_ABORT;
1981 goto out;
1982 }
1983
1984 if (!*args[1]) {
Remi Tricot-Le Bretone6cc5b52020-12-23 18:13:53 +01001985 ha_warning("parsing [%s:%d]: '%s' expects \"on\" or \"off\" (enable or disable vary processing).\n",
Remi Tricot-Le Breton754b2422020-11-16 15:56:10 +01001986 file, linenum, args[0]);
1987 err_code |= ERR_WARN;
1988 }
Remi Tricot-Le Bretone6cc5b52020-12-23 18:13:53 +01001989 if (strcmp(args[1], "on") == 0)
1990 tmp_cache_config->vary_processing_enabled = 1;
1991 else if (strcmp(args[1], "off") == 0)
1992 tmp_cache_config->vary_processing_enabled = 0;
1993 else {
1994 ha_warning("parsing [%s:%d]: '%s' expects \"on\" or \"off\" (enable or disable vary processing).\n",
1995 file, linenum, args[0]);
1996 err_code |= ERR_WARN;
1997 }
Remi Tricot-Le Breton5853c0c2020-12-10 17:58:43 +01001998 } else if (strcmp(args[0], "max-secondary-entries") == 0) {
1999 unsigned int max_sec_entries;
2000 char *err;
2001
2002 if (alertif_too_many_args(1, file, linenum, args, &err_code)) {
2003 err_code |= ERR_ABORT;
2004 goto out;
2005 }
2006
2007 if (!*args[1]) {
2008 ha_warning("parsing [%s:%d]: '%s' expects a strictly positive number.\n",
2009 file, linenum, args[0]);
2010 err_code |= ERR_WARN;
2011 }
2012
2013 max_sec_entries = strtoul(args[1], &err, 10);
2014 if (err == args[1] || *err != '\0' || max_sec_entries == 0) {
2015 ha_warning("parsing [%s:%d]: max-secondary-entries wrong value '%s'\n",
2016 file, linenum, args[1]);
2017 err_code |= ERR_ABORT;
2018 goto out;
2019 }
2020 tmp_cache_config->max_secondary_entries = max_sec_entries;
Frédéric Lécaillea2219f52018-10-22 16:59:13 +02002021 }
2022 else if (*args[0] != 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002023 ha_alert("parsing [%s:%d] : unknown keyword '%s' in 'cache' section\n", file, linenum, args[0]);
William Lallemand41db4602017-10-30 11:15:51 +01002024 err_code |= ERR_ALERT | ERR_FATAL;
2025 goto out;
2026 }
2027out:
2028 return err_code;
2029}
2030
2031/* once the cache section is parsed */
2032
2033int cfg_post_parse_section_cache()
2034{
William Lallemand41db4602017-10-30 11:15:51 +01002035 int err_code = 0;
William Lallemand41db4602017-10-30 11:15:51 +01002036
2037 if (tmp_cache_config) {
William Lallemand41db4602017-10-30 11:15:51 +01002038
2039 if (tmp_cache_config->maxblocks <= 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002040 ha_alert("Size not specified for cache '%s'\n", tmp_cache_config->id);
William Lallemand41db4602017-10-30 11:15:51 +01002041 err_code |= ERR_FATAL | ERR_ALERT;
2042 goto out;
2043 }
2044
Frédéric Lécaille4eba5442018-10-25 20:29:31 +02002045 if (!tmp_cache_config->maxobjsz) {
Frédéric Lécaillea2219f52018-10-22 16:59:13 +02002046 /* Default max. file size is a 256th of the cache size. */
2047 tmp_cache_config->maxobjsz =
2048 (tmp_cache_config->maxblocks * CACHE_BLOCKSIZE) >> 8;
Frédéric Lécaille4eba5442018-10-25 20:29:31 +02002049 }
2050 else if (tmp_cache_config->maxobjsz > tmp_cache_config->maxblocks * CACHE_BLOCKSIZE / 2) {
2051 ha_alert("\"max-object-size\" is limited to an half of \"total-max-size\" => %u\n", tmp_cache_config->maxblocks * CACHE_BLOCKSIZE / 2);
2052 err_code |= ERR_FATAL | ERR_ALERT;
2053 goto out;
2054 }
Frédéric Lécaillea2219f52018-10-22 16:59:13 +02002055
William Lallemandd1d1e222019-08-28 15:22:49 +02002056 /* add to the list of cache to init and reinit tmp_cache_config
2057 * for next cache section, if any.
2058 */
Willy Tarreau2b718102021-04-21 07:32:39 +02002059 LIST_APPEND(&caches_config, &tmp_cache_config->list);
William Lallemandd1d1e222019-08-28 15:22:49 +02002060 tmp_cache_config = NULL;
2061 return err_code;
2062 }
2063out:
Willy Tarreau61cfdf42021-02-20 10:46:51 +01002064 ha_free(&tmp_cache_config);
William Lallemandd1d1e222019-08-28 15:22:49 +02002065 return err_code;
2066
2067}
2068
2069int post_check_cache()
2070{
2071 struct proxy *px;
2072 struct cache *back, *cache_config, *cache;
2073 struct shared_context *shctx;
2074 int ret_shctx;
Christopher Fauletfc633b62020-11-06 15:24:23 +01002075 int err_code = ERR_NONE;
William Lallemandd1d1e222019-08-28 15:22:49 +02002076
2077 list_for_each_entry_safe(cache_config, back, &caches_config, list) {
2078
2079 ret_shctx = shctx_init(&shctx, cache_config->maxblocks, CACHE_BLOCKSIZE,
2080 cache_config->maxobjsz, sizeof(struct cache), 1);
William Lallemand4da3f8a2017-10-31 14:33:34 +01002081
Frédéric Lécaillebc584492018-10-25 20:18:59 +02002082 if (ret_shctx <= 0) {
William Lallemand41db4602017-10-30 11:15:51 +01002083 if (ret_shctx == SHCTX_E_INIT_LOCK)
Christopher Faulet767a84b2017-11-24 16:50:31 +01002084 ha_alert("Unable to initialize the lock for the cache.\n");
William Lallemand41db4602017-10-30 11:15:51 +01002085 else
Christopher Faulet767a84b2017-11-24 16:50:31 +01002086 ha_alert("Unable to allocate cache.\n");
William Lallemand41db4602017-10-30 11:15:51 +01002087
2088 err_code |= ERR_FATAL | ERR_ALERT;
2089 goto out;
2090 }
William Lallemanda400a3a2017-11-20 19:13:12 +01002091 shctx->free_block = cache_free_blocks;
William Lallemandd1d1e222019-08-28 15:22:49 +02002092 /* the cache structure is stored in the shctx and added to the
2093 * caches list, we can remove the entry from the caches_config
2094 * list */
2095 memcpy(shctx->data, cache_config, sizeof(struct cache));
William Lallemand41db4602017-10-30 11:15:51 +01002096 cache = (struct cache *)shctx->data;
Remi Tricot-Le Breton1785f3d2020-11-16 15:56:09 +01002097 cache->entries = EB_ROOT;
Willy Tarreau2b718102021-04-21 07:32:39 +02002098 LIST_APPEND(&caches, &cache->list);
2099 LIST_DELETE(&cache_config->list);
William Lallemandd1d1e222019-08-28 15:22:49 +02002100 free(cache_config);
2101
2102 /* Find all references for this cache in the existing filters
2103 * (over all proxies) and reference it in matching filters.
2104 */
2105 for (px = proxies_list; px; px = px->next) {
2106 struct flt_conf *fconf;
2107 struct cache_flt_conf *cconf;
2108
2109 list_for_each_entry(fconf, &px->filter_configs, list) {
2110 if (fconf->id != cache_store_flt_id)
2111 continue;
2112
2113 cconf = fconf->conf;
Tim Duesterhuse5ff1412021-01-02 22:31:53 +01002114 if (strcmp(cache->id, cconf->c.name) == 0) {
William Lallemandd1d1e222019-08-28 15:22:49 +02002115 free(cconf->c.name);
Tim Duesterhusd7c6e6a2020-09-14 18:01:33 +02002116 cconf->flags |= CACHE_FLT_INIT;
William Lallemandd1d1e222019-08-28 15:22:49 +02002117 cconf->c.cache = cache;
2118 break;
2119 }
2120 }
2121 }
William Lallemand41db4602017-10-30 11:15:51 +01002122 }
William Lallemandd1d1e222019-08-28 15:22:49 +02002123
William Lallemand41db4602017-10-30 11:15:51 +01002124out:
William Lallemand41db4602017-10-30 11:15:51 +01002125 return err_code;
2126
William Lallemand41db4602017-10-30 11:15:51 +01002127}
2128
William Lallemand41db4602017-10-30 11:15:51 +01002129struct flt_ops cache_ops = {
2130 .init = cache_store_init,
Christopher Faulet95220e22018-12-07 17:34:39 +01002131 .check = cache_store_check,
2132 .deinit = cache_store_deinit,
William Lallemand41db4602017-10-30 11:15:51 +01002133
Christopher Faulet65554e12020-03-06 14:52:06 +01002134 /* Handle stream init/deinit */
2135 .attach = cache_store_strm_init,
2136 .detach = cache_store_strm_deinit,
2137
William Lallemand4da3f8a2017-10-31 14:33:34 +01002138 /* Handle channels activity */
Christopher Faulet839791a2019-01-07 16:12:07 +01002139 .channel_post_analyze = cache_store_post_analyze,
William Lallemand4da3f8a2017-10-31 14:33:34 +01002140
2141 /* Filter HTTP requests and responses */
2142 .http_headers = cache_store_http_headers,
Christopher Faulet54a8d5a2018-12-07 12:21:11 +01002143 .http_payload = cache_store_http_payload,
William Lallemand4da3f8a2017-10-31 14:33:34 +01002144 .http_end = cache_store_http_end,
William Lallemand41db4602017-10-30 11:15:51 +01002145};
2146
Christopher Faulet99a17a22018-12-11 09:18:27 +01002147
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002148#define CHECK_ENCODING(str, encoding_name, encoding_value) \
2149 ({ \
2150 int retval = 0; \
2151 if (istmatch(str, (struct ist){ .ptr = encoding_name+1, .len = sizeof(encoding_name) - 2 })) { \
2152 retval = encoding_value; \
2153 encoding = istadv(encoding, sizeof(encoding_name) - 2); \
2154 } \
2155 (retval); \
2156 })
2157
2158/*
2159 * Parse the encoding <encoding> and try to match the encoding part upon an
2160 * encoding list of explicitly supported encodings (which all have a specific
2161 * bit in an encoding bitmap). If a weight is included in the value, find out if
2162 * it is null or not. The bit value will be set in the <encoding_value>
2163 * parameter and the <has_null_weight> will be set to 1 if the weight is strictly
2164 * 0, 1 otherwise.
2165 * The encodings list is extracted from
2166 * https://www.iana.org/assignments/http-parameters/http-parameters.xhtml.
2167 * Returns 0 in case of success and -1 in case of error.
2168 */
2169static int parse_encoding_value(struct ist encoding, unsigned int *encoding_value,
2170 unsigned int *has_null_weight)
2171{
2172 int retval = 0;
2173
2174 if (!encoding_value)
2175 return -1;
2176
2177 if (!istlen(encoding))
2178 return -1; /* Invalid encoding */
2179
2180 *encoding_value = 0;
2181 if (has_null_weight)
2182 *has_null_weight = 0;
2183
2184 switch (*encoding.ptr) {
2185 case 'a':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002186 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002187 *encoding_value = CHECK_ENCODING(encoding, "aes128gcm", VARY_ENCODING_AES128GCM);
2188 break;
2189 case 'b':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002190 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002191 *encoding_value = CHECK_ENCODING(encoding, "br", VARY_ENCODING_BR);
2192 break;
2193 case 'c':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002194 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002195 *encoding_value = CHECK_ENCODING(encoding, "compress", VARY_ENCODING_COMPRESS);
2196 break;
2197 case 'd':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002198 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002199 *encoding_value = CHECK_ENCODING(encoding, "deflate", VARY_ENCODING_DEFLATE);
2200 break;
2201 case 'e':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002202 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002203 *encoding_value = CHECK_ENCODING(encoding, "exi", VARY_ENCODING_EXI);
2204 break;
2205 case 'g':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002206 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002207 *encoding_value = CHECK_ENCODING(encoding, "gzip", VARY_ENCODING_GZIP);
2208 break;
2209 case 'i':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002210 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002211 *encoding_value = CHECK_ENCODING(encoding, "identity", VARY_ENCODING_IDENTITY);
2212 break;
2213 case 'p':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002214 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002215 *encoding_value = CHECK_ENCODING(encoding, "pack200-gzip", VARY_ENCODING_PACK200_GZIP);
2216 break;
2217 case 'x':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002218 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002219 *encoding_value = CHECK_ENCODING(encoding, "x-gzip", VARY_ENCODING_GZIP);
2220 if (!*encoding_value)
2221 *encoding_value = CHECK_ENCODING(encoding, "x-compress", VARY_ENCODING_COMPRESS);
2222 break;
2223 case 'z':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002224 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002225 *encoding_value = CHECK_ENCODING(encoding, "zstd", VARY_ENCODING_ZSTD);
2226 break;
2227 case '*':
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002228 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002229 *encoding_value = VARY_ENCODING_STAR;
2230 break;
2231 default:
2232 retval = -1; /* Unmanaged encoding */
2233 break;
2234 }
2235
2236 /* Process the optional weight part of the encoding. */
2237 if (*encoding_value) {
2238 encoding = http_trim_leading_spht(encoding);
2239 if (istlen(encoding)) {
2240 if (*encoding.ptr != ';')
2241 return -1;
2242
2243 if (has_null_weight) {
Tim Duesterhus284fbe12021-11-04 22:35:44 +01002244 encoding = istnext(encoding);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002245
2246 encoding = http_trim_leading_spht(encoding);
2247
2248 *has_null_weight = isteq(encoding, ist("q=0"));
2249 }
2250 }
2251 }
2252
2253 return retval;
2254}
2255
Tim Duesterhus23b29452020-11-24 22:22:56 +01002256#define ACCEPT_ENCODING_MAX_ENTRIES 16
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002257/*
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002258 * Build a bitmap of the accept-encoding header.
2259 *
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002260 * The bitmap is built by matching every sub-part of the accept-encoding value
2261 * with a subset of explicitly supported encodings, which all have their own bit
2262 * in the bitmap. This bitmap will be used to determine if a response can be
2263 * served to a client (that is if it has an encoding that is accepted by the
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002264 * client). Any unknown encodings will be indicated by the VARY_ENCODING_OTHER
2265 * bit.
2266 *
2267 * Returns 0 in case of success and -1 in case of error.
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002268 */
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002269static int accept_encoding_normalizer(struct htx *htx, struct ist hdr_name,
2270 char *buf, unsigned int *buf_len)
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002271{
Tim Duesterhus23b29452020-11-24 22:22:56 +01002272 size_t count = 0;
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002273 uint32_t encoding_bitmap = 0;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002274 unsigned int encoding_bmp_bl = -1;
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002275 struct http_hdr_ctx ctx = { .blk = NULL };
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002276 unsigned int encoding_value;
2277 unsigned int rejected_encoding;
2278
Ilya Shipitsinb8888ab2021-01-06 21:20:16 +05002279 /* A user agent always accepts an unencoded value unless it explicitly
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002280 * refuses it through an "identity;q=0" accept-encoding value. */
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002281 encoding_bitmap |= VARY_ENCODING_IDENTITY;
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002282
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002283 /* Iterate over all the ACCEPT_ENCODING_MAX_ENTRIES first accept-encoding
2284 * values that might span acrosse multiple accept-encoding headers. */
2285 while (http_find_header(htx, hdr_name, &ctx, 0) && count < ACCEPT_ENCODING_MAX_ENTRIES) {
Tim Duesterhus3bc6af42021-06-18 15:09:28 +02002286 count++;
2287
2288 /* As per RFC7231#5.3.4, "An Accept-Encoding header field with a
2289 * combined field-value that is empty implies that the user agent
2290 * does not want any content-coding in response."
2291 *
2292 * We must (and did) count the existence of this empty header to not
2293 * hit the `count == 0` case below, but must ignore the value to not
2294 * include VARY_ENCODING_OTHER into the final bitmap.
2295 */
2296 if (istlen(ctx.value) == 0)
2297 continue;
2298
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002299 /* Turn accept-encoding value to lower case */
2300 ist2bin_lc(istptr(ctx.value), ctx.value);
Tim Duesterhus23b29452020-11-24 22:22:56 +01002301
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002302 /* Try to identify a known encoding and to manage null weights. */
2303 if (!parse_encoding_value(ctx.value, &encoding_value, &rejected_encoding)) {
2304 if (rejected_encoding)
2305 encoding_bmp_bl &= ~encoding_value;
2306 else
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002307 encoding_bitmap |= encoding_value;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002308 }
2309 else {
2310 /* Unknown encoding */
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002311 encoding_bitmap |= VARY_ENCODING_OTHER;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002312 }
Remi Tricot-Le Breton8bb72aa2020-11-30 17:06:03 +01002313 }
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002314
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002315 /* If a "*" was found in the accepted encodings (without a null weight),
Ilya Shipitsinb8888ab2021-01-06 21:20:16 +05002316 * all the encoding are accepted except the ones explicitly rejected. */
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002317 if (encoding_bitmap & VARY_ENCODING_STAR) {
2318 encoding_bitmap = ~0;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002319 }
2320
Ilya Shipitsinb8888ab2021-01-06 21:20:16 +05002321 /* Clear explicitly rejected encodings from the bitmap */
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002322 encoding_bitmap &= encoding_bmp_bl;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002323
2324 /* As per RFC7231#5.3.4, "If no Accept-Encoding field is in the request,
2325 * any content-coding is considered acceptable by the user agent". */
2326 if (count == 0)
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002327 encoding_bitmap = ~0;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002328
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002329 /* A request with more than ACCEPT_ENCODING_MAX_ENTRIES accepted
2330 * encodings might be illegitimate so we will not use it. */
2331 if (count == ACCEPT_ENCODING_MAX_ENTRIES)
2332 return -1;
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002333
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002334 write_u32(buf, encoding_bitmap);
2335 *buf_len = sizeof(encoding_bitmap);
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002336
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002337 /* This function fills the hash buffer correctly even if no header was
2338 * found, hence the 0 return value (success). */
Tim Duesterhus23b29452020-11-24 22:22:56 +01002339 return 0;
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002340}
Tim Duesterhus23b29452020-11-24 22:22:56 +01002341#undef ACCEPT_ENCODING_MAX_ENTRIES
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002342
2343/*
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002344 * Normalizer used by default for the Referer header. It only
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002345 * calculates a simple crc of the whole value.
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002346 * Only the first occurrence of the header will be taken into account in the
2347 * hash.
2348 * Returns 0 in case of success, 1 if the hash buffer should be filled with 0s
2349 * and -1 in case of error.
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002350 */
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002351static int default_normalizer(struct htx *htx, struct ist hdr_name,
2352 char *buf, unsigned int *buf_len)
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002353{
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002354 int retval = 1;
2355 struct http_hdr_ctx ctx = { .blk = NULL };
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002356
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002357 if (http_find_header(htx, hdr_name, &ctx, 1)) {
2358 retval = 0;
2359 write_u32(buf, hash_crc32(istptr(ctx.value), istlen(ctx.value)));
2360 *buf_len = sizeof(int);
2361 }
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002362
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002363 return retval;
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002364}
2365
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002366/*
Tim Duesterhused84d842021-01-18 13:41:17 +01002367 * Accept-Encoding bitmap comparison function.
2368 * Returns 0 if the bitmaps are compatible.
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002369 */
Tim Duesterhused84d842021-01-18 13:41:17 +01002370static int accept_encoding_bitmap_cmp(const void *ref, const void *new, unsigned int len)
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002371{
Tim Duesterhused84d842021-01-18 13:41:17 +01002372 uint32_t ref_bitmap = read_u32(ref);
2373 uint32_t new_bitmap = read_u32(new);
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002374
Tim Duesterhused84d842021-01-18 13:41:17 +01002375 if (!(ref_bitmap & VARY_ENCODING_OTHER)) {
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002376 /* All the bits set in the reference bitmap correspond to the
2377 * stored response' encoding and should all be set in the new
2378 * encoding bitmap in order for the client to be able to manage
Tim Duesterhusdc38bc42020-12-29 12:43:53 +01002379 * the response.
2380 *
2381 * If this is the case the cached response has encodings that
2382 * are accepted by the client. It can be served directly by
2383 * the cache (as far as the accept-encoding part is concerned).
2384 */
2385
Tim Duesterhused84d842021-01-18 13:41:17 +01002386 return (ref_bitmap & new_bitmap) != ref_bitmap;
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002387 }
Tim Duesterhusdc38bc42020-12-29 12:43:53 +01002388 else {
Tim Duesterhus1d66e392021-01-18 13:41:16 +01002389 return 1;
Tim Duesterhusdc38bc42020-12-29 12:43:53 +01002390 }
Remi Tricot-Le Bretonce9e7b22020-12-23 18:13:49 +01002391}
2392
2393
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002394/*
2395 * Pre-calculate the hashes of all the supported headers (in our Vary
2396 * implementation) of a given request. We have to calculate all the hashes
2397 * in advance because the actual Vary signature won't be known until the first
2398 * response.
2399 * Only the first occurrence of every header will be taken into account in the
2400 * hash.
2401 * If the header is not present, the hash portion of the given header will be
2402 * filled with zeros.
2403 * Returns 0 in case of success.
2404 */
2405static int http_request_prebuild_full_secondary_key(struct stream *s)
2406{
Remi Tricot-Le Bretonbba29122020-12-23 18:13:44 +01002407 /* The fake signature (second parameter) will ensure that every part of the
2408 * secondary key is calculated. */
2409 return http_request_build_secondary_key(s, ~0);
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002410}
2411
2412
2413/*
2414 * Calculate the secondary key for a request for which we already have a known
2415 * vary signature. The key is made by aggregating hashes calculated for every
2416 * header mentioned in the vary signature.
2417 * Only the first occurrence of every header will be taken into account in the
2418 * hash.
2419 * If the header is not present, the hash portion of the given header will be
2420 * filled with zeros.
2421 * Returns 0 in case of success.
2422 */
2423static int http_request_build_secondary_key(struct stream *s, int vary_signature)
2424{
2425 struct http_txn *txn = s->txn;
2426 struct htx *htx = htxbuf(&s->req.buf);
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002427
2428 unsigned int idx;
2429 const struct vary_hashing_information *info = NULL;
2430 unsigned int hash_length = 0;
2431 int retval = 0;
2432 int offset = 0;
2433
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002434 for (idx = 0; idx < sizeof(vary_information)/sizeof(*vary_information) && retval >= 0; ++idx) {
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002435 info = &vary_information[idx];
2436
Remi Tricot-Le Bretone4421de2020-12-23 18:13:46 +01002437 /* The normalizing functions will be in charge of getting the
2438 * header values from the htx. This way they can manage multiple
2439 * occurrences of their processed header. */
2440 if ((vary_signature & info->value) && info->norm_fn != NULL &&
2441 !(retval = info->norm_fn(htx, info->hdr_name, &txn->cache_secondary_hash[offset], &hash_length))) {
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002442 offset += hash_length;
2443 }
2444 else {
2445 /* Fill hash with 0s. */
2446 hash_length = info->hash_length;
2447 memset(&txn->cache_secondary_hash[offset], 0, hash_length);
2448 offset += hash_length;
2449 }
2450 }
2451
Remi Tricot-Le Breton2b5c5cb2020-12-23 18:13:45 +01002452 if (retval >= 0)
2453 txn->flags |= TX_CACHE_HAS_SEC_KEY;
2454
2455 return (retval < 0);
Remi Tricot-Le Breton3d082362020-11-16 15:56:08 +01002456}
2457
2458/*
2459 * Build the actual secondary key of a given request out of the prebuilt key and
2460 * the actual vary signature (extracted from the response).
2461 * Returns 0 in case of success.
2462 */
2463static int http_request_reduce_secondary_key(unsigned int vary_signature,
2464 char prebuilt_key[HTTP_CACHE_SEC_KEY_LEN])
2465{
2466 int offset = 0;
2467 int global_offset = 0;
2468 int vary_info_count = 0;
2469 int keep = 0;
2470 unsigned int vary_idx;
2471 const struct vary_hashing_information *vary_info;
2472
2473 vary_info_count = sizeof(vary_information)/sizeof(*vary_information);
2474 for (vary_idx = 0; vary_idx < vary_info_count; ++vary_idx) {
2475 vary_info = &vary_information[vary_idx];
2476 keep = (vary_signature & vary_info->value) ? 0xff : 0;
2477
2478 for (offset = 0; offset < vary_info->hash_length; ++offset,++global_offset) {
2479 prebuilt_key[global_offset] &= keep;
2480 }
2481 }
2482
2483 return 0;
2484}
2485
2486
Christopher Faulet99a17a22018-12-11 09:18:27 +01002487
2488static int
2489parse_cache_flt(char **args, int *cur_arg, struct proxy *px,
2490 struct flt_conf *fconf, char **err, void *private)
2491{
2492 struct flt_conf *f, *back;
Willy Tarreaua73da1e2018-12-14 10:19:28 +01002493 struct cache_flt_conf *cconf = NULL;
Christopher Faulet99a17a22018-12-11 09:18:27 +01002494 char *name = NULL;
2495 int pos = *cur_arg;
2496
Christopher Faulet2a37cdb2020-05-18 11:58:16 +02002497 /* Get the cache filter name. <pos> point on "cache" keyword */
2498 if (!*args[pos + 1]) {
Tim Duesterhusea969f62020-08-18 22:06:51 +02002499 memprintf(err, "%s : expects a <name> argument", args[pos]);
Christopher Faulet2a37cdb2020-05-18 11:58:16 +02002500 goto error;
2501 }
2502 name = strdup(args[pos + 1]);
2503 if (!name) {
2504 memprintf(err, "%s '%s' : out of memory", args[pos], args[pos + 1]);
2505 goto error;
Christopher Faulet99a17a22018-12-11 09:18:27 +01002506 }
Christopher Faulet2a37cdb2020-05-18 11:58:16 +02002507 pos += 2;
Christopher Faulet99a17a22018-12-11 09:18:27 +01002508
2509 /* Check if an implicit filter with the same name already exists. If so,
2510 * we remove the implicit filter to use the explicit one. */
2511 list_for_each_entry_safe(f, back, &px->filter_configs, list) {
2512 if (f->id != cache_store_flt_id)
2513 continue;
2514
2515 cconf = f->conf;
Tim Duesterhuse5ff1412021-01-02 22:31:53 +01002516 if (strcmp(name, cconf->c.name) != 0) {
Christopher Faulet99a17a22018-12-11 09:18:27 +01002517 cconf = NULL;
2518 continue;
2519 }
2520
2521 if (!(cconf->flags & CACHE_FLT_F_IMPLICIT_DECL)) {
2522 cconf = NULL;
2523 memprintf(err, "%s: multiple explicit declarations of the cache filter '%s'",
2524 px->id, name);
Tim Duesterhusd34b1ce2020-01-18 01:46:18 +01002525 goto error;
Christopher Faulet99a17a22018-12-11 09:18:27 +01002526 }
2527
2528 /* Remove the implicit filter. <cconf> is kept for the explicit one */
Willy Tarreau2b718102021-04-21 07:32:39 +02002529 LIST_DELETE(&f->list);
Christopher Faulet99a17a22018-12-11 09:18:27 +01002530 free(f);
2531 free(name);
2532 break;
2533 }
2534
2535 /* No implicit cache filter found, create configuration for the explicit one */
2536 if (!cconf) {
2537 cconf = calloc(1, sizeof(*cconf));
2538 if (!cconf) {
2539 memprintf(err, "%s: out of memory", args[*cur_arg]);
2540 goto error;
2541 }
2542 cconf->c.name = name;
2543 }
2544
2545 cconf->flags = 0;
2546 fconf->id = cache_store_flt_id;
2547 fconf->conf = cconf;
2548 fconf->ops = &cache_ops;
2549
2550 *cur_arg = pos;
2551 return 0;
2552
2553 error:
2554 free(name);
2555 free(cconf);
2556 return -1;
2557}
2558
Aurélien Nephtaliabbf6072018-04-18 13:26:46 +02002559static int cli_parse_show_cache(char **args, char *payload, struct appctx *appctx, void *private)
William Lallemand1f49a362017-11-21 20:01:26 +01002560{
2561 if (!cli_has_level(appctx, ACCESS_LVL_ADMIN))
2562 return 1;
2563
2564 return 0;
2565}
2566
2567static int cli_io_handler_show_cache(struct appctx *appctx)
2568{
2569 struct cache* cache = appctx->ctx.cli.p0;
Christopher Faulet908628c2022-03-25 16:43:49 +01002570 struct conn_stream *cs = appctx->owner;
William Lallemand1f49a362017-11-21 20:01:26 +01002571
William Lallemand1f49a362017-11-21 20:01:26 +01002572 if (cache == NULL) {
2573 cache = LIST_ELEM((caches).n, typeof(struct cache *), list);
2574 }
2575
2576 list_for_each_entry_from(cache, &caches, list) {
2577 struct eb32_node *node = NULL;
2578 unsigned int next_key;
2579 struct cache_entry *entry;
Remi Tricot-Le Bretone3e1e5f2020-11-27 15:48:40 +01002580 unsigned int i;
William Lallemand1f49a362017-11-21 20:01:26 +01002581
William Lallemand1f49a362017-11-21 20:01:26 +01002582 next_key = appctx->ctx.cli.i0;
Willy Tarreauafe1de52018-04-04 11:56:43 +02002583 if (!next_key) {
2584 chunk_printf(&trash, "%p: %s (shctx:%p, available blocks:%d)\n", cache, cache->id, shctx_ptr(cache), shctx_ptr(cache)->nbav);
Christopher Faulet908628c2022-03-25 16:43:49 +01002585 if (ci_putchk(cs_ic(cs), &trash) == -1) {
2586 si_rx_room_blk(cs->si);
Willy Tarreauafe1de52018-04-04 11:56:43 +02002587 return 0;
2588 }
2589 }
William Lallemand1f49a362017-11-21 20:01:26 +01002590
2591 appctx->ctx.cli.p0 = cache;
2592
2593 while (1) {
2594
2595 shctx_lock(shctx_ptr(cache));
Christopher Faulet27f88a92021-11-23 16:03:05 +01002596 node = eb32_lookup_ge(&cache->entries, next_key);
William Lallemand1f49a362017-11-21 20:01:26 +01002597 if (!node) {
2598 shctx_unlock(shctx_ptr(cache));
Willy Tarreauafe1de52018-04-04 11:56:43 +02002599 appctx->ctx.cli.i0 = 0;
William Lallemand1f49a362017-11-21 20:01:26 +01002600 break;
2601 }
2602
2603 entry = container_of(node, struct cache_entry, eb);
William Lallemand1f49a362017-11-21 20:01:26 +01002604 next_key = node->key + 1;
Willy Tarreauf1de1b52022-04-13 11:21:39 +02002605
2606 if (entry->expire > now.tv_sec) {
2607 chunk_printf(&trash, "%p hash:%u vary:0x", entry, read_u32(entry->hash));
2608 for (i = 0; i < HTTP_CACHE_SEC_KEY_LEN; ++i)
2609 chunk_appendf(&trash, "%02x", (unsigned char)entry->secondary_key[i]);
2610 chunk_appendf(&trash, " size:%u (%u blocks), refcount:%u, expire:%d\n",
2611 block_ptr(entry)->len, block_ptr(entry)->block_count,
2612 block_ptr(entry)->refcount, entry->expire - (int)now.tv_sec);
2613 } else {
2614 /* time to remove that one */
2615 delete_entry(entry);
2616 entry->eb.key = 0;
2617 }
2618
William Lallemand1f49a362017-11-21 20:01:26 +01002619 appctx->ctx.cli.i0 = next_key;
2620
2621 shctx_unlock(shctx_ptr(cache));
2622
Christopher Faulet908628c2022-03-25 16:43:49 +01002623 if (ci_putchk(cs_ic(cs), &trash) == -1) {
2624 si_rx_room_blk(cs->si);
William Lallemand1f49a362017-11-21 20:01:26 +01002625 return 0;
2626 }
2627 }
2628
2629 }
2630
2631 return 1;
2632
2633}
2634
Remi Tricot-Le Bretonbf971212020-10-27 11:55:57 +01002635
2636/*
2637 * boolean, returns true if response was built out of a cache entry.
2638 */
2639static int
2640smp_fetch_res_cache_hit(const struct arg *args, struct sample *smp,
2641 const char *kw, void *private)
2642{
2643 smp->data.type = SMP_T_BOOL;
2644 smp->data.u.sint = (smp->strm ? (smp->strm->target == &http_cache_applet.obj_type) : 0);
2645
2646 return 1;
2647}
2648
2649/*
2650 * string, returns cache name (if response came from a cache).
2651 */
2652static int
2653smp_fetch_res_cache_name(const struct arg *args, struct sample *smp,
2654 const char *kw, void *private)
2655{
2656 struct appctx *appctx = NULL;
2657
2658 struct cache_flt_conf *cconf = NULL;
2659 struct cache *cache = NULL;
2660
2661 if (!smp->strm || smp->strm->target != &http_cache_applet.obj_type)
2662 return 0;
2663
Christopher Faulet908628c2022-03-25 16:43:49 +01002664 /* Get appctx from the conn-stream. */
Christopher Faulet95a61e82021-12-22 14:22:03 +01002665 appctx = cs_appctx(smp->strm->csb);
Remi Tricot-Le Bretonbf971212020-10-27 11:55:57 +01002666 if (appctx && appctx->rule) {
2667 cconf = appctx->rule->arg.act.p[0];
2668 if (cconf) {
2669 cache = cconf->c.cache;
2670
2671 smp->data.type = SMP_T_STR;
2672 smp->flags = SMP_F_CONST;
2673 smp->data.u.str.area = cache->id;
2674 smp->data.u.str.data = strlen(cache->id);
2675 return 1;
2676 }
2677 }
2678
2679 return 0;
2680}
2681
Christopher Faulet99a17a22018-12-11 09:18:27 +01002682/* Declare the filter parser for "cache" keyword */
2683static struct flt_kw_list filter_kws = { "CACHE", { }, {
2684 { "cache", parse_cache_flt, NULL },
2685 { NULL, NULL, NULL },
2686 }
2687};
2688
2689INITCALL1(STG_REGISTER, flt_register_keywords, &filter_kws);
2690
William Lallemand1f49a362017-11-21 20:01:26 +01002691static struct cli_kw_list cli_kws = {{},{
Willy Tarreaub205bfd2021-05-07 11:38:37 +02002692 { { "show", "cache", NULL }, "show cache : show cache status", cli_parse_show_cache, cli_io_handler_show_cache, NULL, NULL },
William Lallemande899af82017-11-22 16:41:26 +01002693 {{},}
William Lallemand1f49a362017-11-21 20:01:26 +01002694}};
2695
Willy Tarreau0108d902018-11-25 19:14:37 +01002696INITCALL1(STG_REGISTER, cli_register_kw, &cli_kws);
William Lallemand1f49a362017-11-21 20:01:26 +01002697
William Lallemand41db4602017-10-30 11:15:51 +01002698static struct action_kw_list http_res_actions = {
2699 .kw = {
2700 { "cache-store", parse_cache_store },
2701 { NULL, NULL }
2702 }
2703};
2704
Willy Tarreau0108d902018-11-25 19:14:37 +01002705INITCALL1(STG_REGISTER, http_res_keywords_register, &http_res_actions);
2706
William Lallemand41db4602017-10-30 11:15:51 +01002707static struct action_kw_list http_req_actions = {
2708 .kw = {
2709 { "cache-use", parse_cache_use },
2710 { NULL, NULL }
2711 }
2712};
2713
Willy Tarreau0108d902018-11-25 19:14:37 +01002714INITCALL1(STG_REGISTER, http_req_keywords_register, &http_req_actions);
2715
Willy Tarreau2231b632019-03-29 18:26:52 +01002716struct applet http_cache_applet = {
William Lallemand41db4602017-10-30 11:15:51 +01002717 .obj_type = OBJ_TYPE_APPLET,
2718 .name = "<CACHE>", /* used for logging */
William Lallemand77c11972017-10-31 20:43:01 +01002719 .fct = http_cache_io_handler,
William Lallemandecb73b12017-11-24 14:33:55 +01002720 .release = http_cache_applet_release,
William Lallemand41db4602017-10-30 11:15:51 +01002721};
2722
Willy Tarreaue6552512018-11-26 11:33:13 +01002723/* config parsers for this section */
2724REGISTER_CONFIG_SECTION("cache", cfg_parse_cache, cfg_post_parse_section_cache);
William Lallemandd1d1e222019-08-28 15:22:49 +02002725REGISTER_POST_CHECK(post_check_cache);
Remi Tricot-Le Bretonbf971212020-10-27 11:55:57 +01002726
2727
2728/* Note: must not be declared <const> as its list will be overwritten */
2729static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
2730 { "res.cache_hit", smp_fetch_res_cache_hit, 0, NULL, SMP_T_BOOL, SMP_USE_HRSHP, SMP_VAL_RESPONSE },
2731 { "res.cache_name", smp_fetch_res_cache_name, 0, NULL, SMP_T_STR, SMP_USE_HRSHP, SMP_VAL_RESPONSE },
2732 { /* END */ },
2733 }
2734};
2735
2736INITCALL1(STG_REGISTER, sample_register_fetches, &sample_fetch_keywords);