lustre/obdclass/upcall_cache.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/obdclass/upcall_cache.c
  32  *
  33  * Supplementary groups cache.
  34  */
  35 #define DEBUG_SUBSYSTEM S_SEC
  36
  37 #include <libcfs/libcfs.h>
  38 #include <uapi/linux/lnet/lnet-types.h>
  39 #include <upcall_cache.h>
  40
  41 static struct upcall_cache_entry *alloc_entry(struct upcall_cache *cache,
  42                                               __u64 key, void *args)
  43 {
  44         struct upcall_cache_entry *entry;
  45
  46         LIBCFS_ALLOC(entry, sizeof(*entry));
  47         if (!entry)
  48                 return NULL;
  49
  50         UC_CACHE_SET_NEW(entry);
  51         INIT_LIST_HEAD(&entry->ue_hash);
  52         entry->ue_key = key;
  53         atomic_set(&entry->ue_refcount, 0);
  54         init_waitqueue_head(&entry->ue_waitq);
  55         if (cache->uc_ops->init_entry)
  56                 cache->uc_ops->init_entry(entry, args);
  57         return entry;
  58 }
  59
  60 /* protected by cache lock */
  61 static void free_entry(struct upcall_cache *cache,
  62                        struct upcall_cache_entry *entry)
  63 {
  64         if (cache->uc_ops->free_entry)
  65                 cache->uc_ops->free_entry(cache, entry);
  66
  67         list_del(&entry->ue_hash);
  68         CDEBUG(D_OTHER, "destroy cache entry %p for key %llu\n",
  69                 entry, entry->ue_key);
  70         LIBCFS_FREE(entry, sizeof(*entry));
  71 }
  72
  73 static inline int upcall_compare(struct upcall_cache *cache,
  74                                  struct upcall_cache_entry *entry,
  75                                  __u64 key, void *args)
  76 {
  77         if (entry->ue_key != key)
  78                 return -1;
  79
  80         if (cache->uc_ops->upcall_compare)
  81                 return cache->uc_ops->upcall_compare(cache, entry, key, args);
  82
  83         return 0;
  84 }
  85
  86 static inline int downcall_compare(struct upcall_cache *cache,
  87                                    struct upcall_cache_entry *entry,
  88                                    __u64 key, void *args)
  89 {
  90         if (entry->ue_key != key)
  91                 return -1;
  92
  93         if (cache->uc_ops->downcall_compare)
  94                 return cache->uc_ops->downcall_compare(cache, entry, key, args);
  95
  96         return 0;
  97 }
  98
  99 static inline void get_entry(struct upcall_cache_entry *entry)
 100 {
 101         atomic_inc(&entry->ue_refcount);
 102 }
 103
 104 static inline void put_entry(struct upcall_cache *cache,
 105                              struct upcall_cache_entry *entry)
 106 {
 107         if (atomic_dec_and_test(&entry->ue_refcount) &&
 108             (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry))) {
 109                 free_entry(cache, entry);
 110         }
 111 }
 112
 113 static int check_unlink_entry(struct upcall_cache *cache,
 114                               struct upcall_cache_entry *entry)
 115 {
 116         time64_t now = ktime_get_seconds();
 117
 118         if (UC_CACHE_IS_VALID(entry) && now < entry->ue_expire)
 119                 return 0;
 120
 121         if (UC_CACHE_IS_ACQUIRING(entry)) {
 122                 if (entry->ue_acquire_expire == 0 ||
 123                     now < entry->ue_acquire_expire)
 124                         return 0;
 125
 126                 UC_CACHE_SET_EXPIRED(entry);
 127                 wake_up(&entry->ue_waitq);
 128         } else if (!UC_CACHE_IS_INVALID(entry)) {
 129                 UC_CACHE_SET_EXPIRED(entry);
 130         }
 131
 132         list_del_init(&entry->ue_hash);
 133         if (!atomic_read(&entry->ue_refcount))
 134                 free_entry(cache, entry);
 135         return 1;
 136 }
 137
 138 static inline int refresh_entry(struct upcall_cache *cache,
 139                          struct upcall_cache_entry *entry)
 140 {
 141         LASSERT(cache->uc_ops->do_upcall);
 142         return cache->uc_ops->do_upcall(cache, entry);
 143 }
 144
 145 struct upcall_cache_entry *upcall_cache_get_entry(struct upcall_cache *cache,
 146                                                   __u64 key, void *args)
 147 {
 148         struct upcall_cache_entry *entry = NULL, *new = NULL, *next;
 149         struct list_head *head;
 150         wait_queue_entry_t wait;
 151         int rc, found;
 152         ENTRY;
 153
 154         LASSERT(cache);
 155
 156         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 157 find_again:
 158         found = 0;
 159         spin_lock(&cache->uc_lock);
 160         list_for_each_entry_safe(entry, next, head, ue_hash) {
 161                 /* check invalid & expired items */
 162                 if (check_unlink_entry(cache, entry))
 163                         continue;
 164                 if (upcall_compare(cache, entry, key, args) == 0) {
 165                         found = 1;
 166                         break;
 167                 }
 168         }
 169
 170         if (!found) {
 171                 if (!new) {
 172                         spin_unlock(&cache->uc_lock);
 173                         new = alloc_entry(cache, key, args);
 174                         if (!new) {
 175                                 CERROR("fail to alloc entry\n");
 176                                 RETURN(ERR_PTR(-ENOMEM));
 177                         }
 178                         goto find_again;
 179                 } else {
 180                         list_add(&new->ue_hash, head);
 181                         entry = new;
 182                 }
 183         } else {
 184                 if (new) {
 185                         free_entry(cache, new);
 186                         new = NULL;
 187                 }
 188                 list_move(&entry->ue_hash, head);
 189         }
 190         get_entry(entry);
 191
 192         /* acquire for new one */
 193         if (UC_CACHE_IS_NEW(entry)) {
 194                 UC_CACHE_SET_ACQUIRING(entry);
 195                 UC_CACHE_CLEAR_NEW(entry);
 196                 spin_unlock(&cache->uc_lock);
 197                 rc = refresh_entry(cache, entry);
 198                 spin_lock(&cache->uc_lock);
 199                 entry->ue_acquire_expire = ktime_get_seconds() +
 200                                            cache->uc_acquire_expire;
 201                 if (rc < 0) {
 202                         UC_CACHE_CLEAR_ACQUIRING(entry);
 203                         UC_CACHE_SET_INVALID(entry);
 204                         wake_up(&entry->ue_waitq);
 205                         if (unlikely(rc == -EREMCHG)) {
 206                                 put_entry(cache, entry);
 207                                 GOTO(out, entry = ERR_PTR(rc));
 208                         }
 209                 }
 210         }
 211         /* someone (and only one) is doing upcall upon this item,
 212          * wait it to complete */
 213         if (UC_CACHE_IS_ACQUIRING(entry)) {
 214                 long expiry = (entry == new) ?
 215                               cfs_time_seconds(cache->uc_acquire_expire) :
 216                               MAX_SCHEDULE_TIMEOUT;
 217                 long left;
 218
 219                 init_wait(&wait);
 220                 add_wait_queue(&entry->ue_waitq, &wait);
 221                 set_current_state(TASK_INTERRUPTIBLE);
 222                 spin_unlock(&cache->uc_lock);
 223
 224                 left = schedule_timeout(expiry);
 225
 226                 spin_lock(&cache->uc_lock);
 227                 remove_wait_queue(&entry->ue_waitq, &wait);
 228                 if (UC_CACHE_IS_ACQUIRING(entry)) {
 229                         /* we're interrupted or upcall failed in the middle */
 230                         rc = left > 0 ? -EINTR : -ETIMEDOUT;
 231                         CERROR("acquire for key %llu: error %d\n",
 232                                entry->ue_key, rc);
 233                         put_entry(cache, entry);
 234                         GOTO(out, entry = ERR_PTR(rc));
 235                 }
 236         }
 237
 238         /* invalid means error, don't need to try again */
 239         if (UC_CACHE_IS_INVALID(entry)) {
 240                 put_entry(cache, entry);
 241                 GOTO(out, entry = ERR_PTR(-EIDRM));
 242         }
 243
 244         /* check expired
 245          * We can't refresh the existing one because some
 246          * memory might be shared by multiple processes.
 247          */
 248         if (check_unlink_entry(cache, entry)) {
 249                 /* if expired, try again. but if this entry is
 250                  * created by me but too quickly turn to expired
 251                  * without any error, should at least give a
 252                  * chance to use it once.
 253                  */
 254                 if (entry != new) {
 255                         put_entry(cache, entry);
 256                         spin_unlock(&cache->uc_lock);
 257                         new = NULL;
 258                         goto find_again;
 259                 }
 260         }
 261
 262         /* Now we know it's good */
 263 out:
 264         spin_unlock(&cache->uc_lock);
 265         RETURN(entry);
 266 }
 267 EXPORT_SYMBOL(upcall_cache_get_entry);
 268
 269 void upcall_cache_put_entry(struct upcall_cache *cache,
 270                             struct upcall_cache_entry *entry)
 271 {
 272         ENTRY;
 273
 274         if (!entry) {
 275                 EXIT;
 276                 return;
 277         }
 278
 279         LASSERT(atomic_read(&entry->ue_refcount) > 0);
 280         spin_lock(&cache->uc_lock);
 281         put_entry(cache, entry);
 282         spin_unlock(&cache->uc_lock);
 283         EXIT;
 284 }
 285 EXPORT_SYMBOL(upcall_cache_put_entry);
 286
 287 int upcall_cache_downcall(struct upcall_cache *cache, __u32 err, __u64 key,
 288                           void *args)
 289 {
 290         struct upcall_cache_entry *entry = NULL;
 291         struct list_head *head;
 292         int found = 0, rc = 0;
 293         ENTRY;
 294
 295         LASSERT(cache);
 296
 297         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 298
 299         spin_lock(&cache->uc_lock);
 300         list_for_each_entry(entry, head, ue_hash) {
 301                 if (downcall_compare(cache, entry, key, args) == 0) {
 302                         found = 1;
 303                         get_entry(entry);
 304                         break;
 305                 }
 306         }
 307
 308         if (!found) {
 309                 CDEBUG(D_OTHER, "%s: upcall for key %llu not expected\n",
 310                        cache->uc_name, key);
 311                 /* haven't found, it's possible */
 312                 spin_unlock(&cache->uc_lock);
 313                 RETURN(-EINVAL);
 314         }
 315
 316         if (err) {
 317                 CDEBUG(D_OTHER, "%s: upcall for key %llu returned %d\n",
 318                        cache->uc_name, entry->ue_key, err);
 319                 GOTO(out, rc = -EINVAL);
 320         }
 321
 322         if (!UC_CACHE_IS_ACQUIRING(entry)) {
 323                 CDEBUG(D_RPCTRACE, "%s: found uptodate entry %p (key %llu)"
 324                        "\n", cache->uc_name, entry, entry->ue_key);
 325                 GOTO(out, rc = 0);
 326         }
 327
 328         if (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry)) {
 329                 CERROR("%s: found a stale entry %p (key %llu) in ioctl\n",
 330                        cache->uc_name, entry, entry->ue_key);
 331                 GOTO(out, rc = -EINVAL);
 332         }
 333
 334         spin_unlock(&cache->uc_lock);
 335         if (cache->uc_ops->parse_downcall)
 336                 rc = cache->uc_ops->parse_downcall(cache, entry, args);
 337         spin_lock(&cache->uc_lock);
 338         if (rc)
 339                 GOTO(out, rc);
 340
 341         entry->ue_expire = ktime_get_seconds() + cache->uc_entry_expire;
 342         UC_CACHE_SET_VALID(entry);
 343         CDEBUG(D_OTHER, "%s: created upcall cache entry %p for key %llu\n",
 344                cache->uc_name, entry, entry->ue_key);
 345 out:
 346         if (rc) {
 347                 UC_CACHE_SET_INVALID(entry);
 348                 list_del_init(&entry->ue_hash);
 349         }
 350         UC_CACHE_CLEAR_ACQUIRING(entry);
 351         spin_unlock(&cache->uc_lock);
 352         wake_up(&entry->ue_waitq);
 353         put_entry(cache, entry);
 354
 355         RETURN(rc);
 356 }
 357 EXPORT_SYMBOL(upcall_cache_downcall);
 358
 359 void upcall_cache_flush(struct upcall_cache *cache, int force)
 360 {
 361         struct upcall_cache_entry *entry, *next;
 362         int i;
 363         ENTRY;
 364
 365         spin_lock(&cache->uc_lock);
 366         for (i = 0; i < UC_CACHE_HASH_SIZE; i++) {
 367                 list_for_each_entry_safe(entry, next,
 368                                          &cache->uc_hashtable[i], ue_hash) {
 369                         if (!force && atomic_read(&entry->ue_refcount)) {
 370                                 UC_CACHE_SET_EXPIRED(entry);
 371                                 continue;
 372                         }
 373                         LASSERT(!atomic_read(&entry->ue_refcount));
 374                         free_entry(cache, entry);
 375                 }
 376         }
 377         spin_unlock(&cache->uc_lock);
 378         EXIT;
 379 }
 380 EXPORT_SYMBOL(upcall_cache_flush);
 381
 382 void upcall_cache_flush_one(struct upcall_cache *cache, __u64 key, void *args)
 383 {
 384         struct list_head *head;
 385         struct upcall_cache_entry *entry;
 386         int found = 0;
 387         ENTRY;
 388
 389         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 390
 391         spin_lock(&cache->uc_lock);
 392         list_for_each_entry(entry, head, ue_hash) {
 393                 if (upcall_compare(cache, entry, key, args) == 0) {
 394                         found = 1;
 395                         break;
 396                 }
 397         }
 398
 399         if (found) {
 400                 CWARN("%s: flush entry %p: key %llu, ref %d, fl %x, "
 401                       "cur %lld, ex %lld/%lld\n",
 402                       cache->uc_name, entry, entry->ue_key,
 403                       atomic_read(&entry->ue_refcount), entry->ue_flags,
 404                       ktime_get_real_seconds(), entry->ue_acquire_expire,
 405                       entry->ue_expire);
 406                 UC_CACHE_SET_EXPIRED(entry);
 407                 if (!atomic_read(&entry->ue_refcount))
 408                         free_entry(cache, entry);
 409         }
 410         spin_unlock(&cache->uc_lock);
 411 }
 412 EXPORT_SYMBOL(upcall_cache_flush_one);
 413
 414 struct upcall_cache *upcall_cache_init(const char *name, const char *upcall,
 415                                        struct upcall_cache_ops *ops)
 416 {
 417         struct upcall_cache *cache;
 418         int i;
 419         ENTRY;
 420
 421         LIBCFS_ALLOC(cache, sizeof(*cache));
 422         if (!cache)
 423                 RETURN(ERR_PTR(-ENOMEM));
 424
 425         spin_lock_init(&cache->uc_lock);
 426         init_rwsem(&cache->uc_upcall_rwsem);
 427         for (i = 0; i < UC_CACHE_HASH_SIZE; i++)
 428                 INIT_LIST_HEAD(&cache->uc_hashtable[i]);
 429         strlcpy(cache->uc_name, name, sizeof(cache->uc_name));
 430         /* upcall pathname proc tunable */
 431         strlcpy(cache->uc_upcall, upcall, sizeof(cache->uc_upcall));
 432         cache->uc_entry_expire = 20 * 60;
 433         cache->uc_acquire_expire = 30;
 434         cache->uc_ops = ops;
 435
 436         RETURN(cache);
 437 }
 438 EXPORT_SYMBOL(upcall_cache_init);
 439
 440 void upcall_cache_cleanup(struct upcall_cache *cache)
 441 {
 442         if (!cache)
 443                 return;
 444         upcall_cache_flush_all(cache);
 445         LIBCFS_FREE(cache, sizeof(*cache));
 446 }
 447 EXPORT_SYMBOL(upcall_cache_cleanup);