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         bool failedacquiring = false;
 150         struct list_head *head;
 151         wait_queue_entry_t wait;
 152         int rc, found;
 153         ENTRY;
 154
 155         LASSERT(cache);
 156
 157         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 158 find_again:
 159         found = 0;
 160         spin_lock(&cache->uc_lock);
 161         list_for_each_entry_safe(entry, next, head, ue_hash) {
 162                 /* check invalid & expired items */
 163                 if (check_unlink_entry(cache, entry))
 164                         continue;
 165                 if (upcall_compare(cache, entry, key, args) == 0) {
 166                         found = 1;
 167                         break;
 168                 }
 169         }
 170
 171         if (!found) {
 172                 if (!new) {
 173                         spin_unlock(&cache->uc_lock);
 174                         new = alloc_entry(cache, key, args);
 175                         if (!new) {
 176                                 CERROR("fail to alloc entry\n");
 177                                 RETURN(ERR_PTR(-ENOMEM));
 178                         }
 179                         goto find_again;
 180                 } else {
 181                         list_add(&new->ue_hash, head);
 182                         entry = new;
 183                 }
 184         } else {
 185                 if (new) {
 186                         free_entry(cache, new);
 187                         new = NULL;
 188                 }
 189                 list_move(&entry->ue_hash, head);
 190         }
 191         get_entry(entry);
 192
 193         /* acquire for new one */
 194         if (UC_CACHE_IS_NEW(entry)) {
 195                 UC_CACHE_SET_ACQUIRING(entry);
 196                 UC_CACHE_CLEAR_NEW(entry);
 197                 spin_unlock(&cache->uc_lock);
 198                 rc = refresh_entry(cache, entry);
 199                 spin_lock(&cache->uc_lock);
 200                 entry->ue_acquire_expire = ktime_get_seconds() +
 201                                            cache->uc_acquire_expire;
 202                 if (rc < 0) {
 203                         UC_CACHE_CLEAR_ACQUIRING(entry);
 204                         UC_CACHE_SET_INVALID(entry);
 205                         wake_up(&entry->ue_waitq);
 206                         if (unlikely(rc == -EREMCHG)) {
 207                                 put_entry(cache, entry);
 208                                 GOTO(out, entry = ERR_PTR(rc));
 209                         }
 210                 }
 211         }
 212         /* someone (and only one) is doing upcall upon this item,
 213          * wait it to complete */
 214         if (UC_CACHE_IS_ACQUIRING(entry)) {
 215                 long expiry = (entry == new) ?
 216                               cfs_time_seconds(cache->uc_acquire_expire) :
 217                               MAX_SCHEDULE_TIMEOUT;
 218                 long left;
 219
 220                 init_wait(&wait);
 221                 add_wait_queue(&entry->ue_waitq, &wait);
 222                 set_current_state(TASK_INTERRUPTIBLE);
 223                 spin_unlock(&cache->uc_lock);
 224
 225                 left = schedule_timeout(expiry);
 226
 227                 spin_lock(&cache->uc_lock);
 228                 remove_wait_queue(&entry->ue_waitq, &wait);
 229                 if (UC_CACHE_IS_ACQUIRING(entry)) {
 230                         /* we're interrupted or upcall failed in the middle */
 231                         rc = left > 0 ? -EINTR : -ETIMEDOUT;
 232                         put_entry(cache, entry);
 233                         if (!failedacquiring) {
 234                                 spin_unlock(&cache->uc_lock);
 235                                 failedacquiring = true;
 236                                 new = NULL;
 237                                 CDEBUG(D_OTHER,
 238                                        "retry acquire for key %llu (got %d)\n",
 239                                        entry->ue_key, rc);
 240                                 goto find_again;
 241                         }
 242                         CERROR("acquire for key %llu: error %d\n",
 243                                entry->ue_key, rc);
 244                         GOTO(out, entry = ERR_PTR(rc));
 245                 }
 246         }
 247
 248         /* invalid means error, don't need to try again */
 249         if (UC_CACHE_IS_INVALID(entry)) {
 250                 put_entry(cache, entry);
 251                 GOTO(out, entry = ERR_PTR(-EIDRM));
 252         }
 253
 254         /* check expired
 255          * We can't refresh the existing one because some
 256          * memory might be shared by multiple processes.
 257          */
 258         if (check_unlink_entry(cache, entry)) {
 259                 /* if expired, try again. but if this entry is
 260                  * created by me but too quickly turn to expired
 261                  * without any error, should at least give a
 262                  * chance to use it once.
 263                  */
 264                 if (entry != new) {
 265                         put_entry(cache, entry);
 266                         spin_unlock(&cache->uc_lock);
 267                         new = NULL;
 268                         goto find_again;
 269                 }
 270         }
 271
 272         /* Now we know it's good */
 273 out:
 274         spin_unlock(&cache->uc_lock);
 275         RETURN(entry);
 276 }
 277 EXPORT_SYMBOL(upcall_cache_get_entry);
 278
 279 void upcall_cache_put_entry(struct upcall_cache *cache,
 280                             struct upcall_cache_entry *entry)
 281 {
 282         ENTRY;
 283
 284         if (!entry) {
 285                 EXIT;
 286                 return;
 287         }
 288
 289         LASSERT(atomic_read(&entry->ue_refcount) > 0);
 290         spin_lock(&cache->uc_lock);
 291         put_entry(cache, entry);
 292         spin_unlock(&cache->uc_lock);
 293         EXIT;
 294 }
 295 EXPORT_SYMBOL(upcall_cache_put_entry);
 296
 297 int upcall_cache_downcall(struct upcall_cache *cache, __u32 err, __u64 key,
 298                           void *args)
 299 {
 300         struct upcall_cache_entry *entry = NULL;
 301         struct list_head *head;
 302         int found = 0, rc = 0;
 303         ENTRY;
 304
 305         LASSERT(cache);
 306
 307         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 308
 309         spin_lock(&cache->uc_lock);
 310         list_for_each_entry(entry, head, ue_hash) {
 311                 if (downcall_compare(cache, entry, key, args) == 0) {
 312                         found = 1;
 313                         get_entry(entry);
 314                         break;
 315                 }
 316         }
 317
 318         if (!found) {
 319                 CDEBUG(D_OTHER, "%s: upcall for key %llu not expected\n",
 320                        cache->uc_name, key);
 321                 /* haven't found, it's possible */
 322                 spin_unlock(&cache->uc_lock);
 323                 RETURN(-EINVAL);
 324         }
 325
 326         if (err) {
 327                 CDEBUG(D_OTHER, "%s: upcall for key %llu returned %d\n",
 328                        cache->uc_name, entry->ue_key, err);
 329                 GOTO(out, rc = -EINVAL);
 330         }
 331
 332         if (!UC_CACHE_IS_ACQUIRING(entry)) {
 333                 CDEBUG(D_RPCTRACE, "%s: found uptodate entry %p (key %llu)"
 334                        "\n", cache->uc_name, entry, entry->ue_key);
 335                 GOTO(out, rc = 0);
 336         }
 337
 338         if (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry)) {
 339                 CERROR("%s: found a stale entry %p (key %llu) in ioctl\n",
 340                        cache->uc_name, entry, entry->ue_key);
 341                 GOTO(out, rc = -EINVAL);
 342         }
 343
 344         spin_unlock(&cache->uc_lock);
 345         if (cache->uc_ops->parse_downcall)
 346                 rc = cache->uc_ops->parse_downcall(cache, entry, args);
 347         spin_lock(&cache->uc_lock);
 348         if (rc)
 349                 GOTO(out, rc);
 350
 351         entry->ue_expire = ktime_get_seconds() + cache->uc_entry_expire;
 352         UC_CACHE_SET_VALID(entry);
 353         CDEBUG(D_OTHER, "%s: created upcall cache entry %p for key %llu\n",
 354                cache->uc_name, entry, entry->ue_key);
 355 out:
 356         if (rc) {
 357                 UC_CACHE_SET_INVALID(entry);
 358                 list_del_init(&entry->ue_hash);
 359         }
 360         UC_CACHE_CLEAR_ACQUIRING(entry);
 361         spin_unlock(&cache->uc_lock);
 362         wake_up(&entry->ue_waitq);
 363         put_entry(cache, entry);
 364
 365         RETURN(rc);
 366 }
 367 EXPORT_SYMBOL(upcall_cache_downcall);
 368
 369 void upcall_cache_flush(struct upcall_cache *cache, int force)
 370 {
 371         struct upcall_cache_entry *entry, *next;
 372         int i;
 373         ENTRY;
 374
 375         spin_lock(&cache->uc_lock);
 376         for (i = 0; i < UC_CACHE_HASH_SIZE; i++) {
 377                 list_for_each_entry_safe(entry, next,
 378                                          &cache->uc_hashtable[i], ue_hash) {
 379                         if (!force && atomic_read(&entry->ue_refcount)) {
 380                                 UC_CACHE_SET_EXPIRED(entry);
 381                                 continue;
 382                         }
 383                         LASSERT(!atomic_read(&entry->ue_refcount));
 384                         free_entry(cache, entry);
 385                 }
 386         }
 387         spin_unlock(&cache->uc_lock);
 388         EXIT;
 389 }
 390 EXPORT_SYMBOL(upcall_cache_flush);
 391
 392 void upcall_cache_flush_one(struct upcall_cache *cache, __u64 key, void *args)
 393 {
 394         struct list_head *head;
 395         struct upcall_cache_entry *entry;
 396         int found = 0;
 397         ENTRY;
 398
 399         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 400
 401         spin_lock(&cache->uc_lock);
 402         list_for_each_entry(entry, head, ue_hash) {
 403                 if (upcall_compare(cache, entry, key, args) == 0) {
 404                         found = 1;
 405                         break;
 406                 }
 407         }
 408
 409         if (found) {
 410                 CWARN("%s: flush entry %p: key %llu, ref %d, fl %x, "
 411                       "cur %lld, ex %lld/%lld\n",
 412                       cache->uc_name, entry, entry->ue_key,
 413                       atomic_read(&entry->ue_refcount), entry->ue_flags,
 414                       ktime_get_real_seconds(), entry->ue_acquire_expire,
 415                       entry->ue_expire);
 416                 UC_CACHE_SET_EXPIRED(entry);
 417                 if (!atomic_read(&entry->ue_refcount))
 418                         free_entry(cache, entry);
 419         }
 420         spin_unlock(&cache->uc_lock);
 421 }
 422 EXPORT_SYMBOL(upcall_cache_flush_one);
 423
 424 struct upcall_cache *upcall_cache_init(const char *name, const char *upcall,
 425                                        struct upcall_cache_ops *ops)
 426 {
 427         struct upcall_cache *cache;
 428         int i;
 429         ENTRY;
 430
 431         LIBCFS_ALLOC(cache, sizeof(*cache));
 432         if (!cache)
 433                 RETURN(ERR_PTR(-ENOMEM));
 434
 435         spin_lock_init(&cache->uc_lock);
 436         init_rwsem(&cache->uc_upcall_rwsem);
 437         for (i = 0; i < UC_CACHE_HASH_SIZE; i++)
 438                 INIT_LIST_HEAD(&cache->uc_hashtable[i]);
 439         strlcpy(cache->uc_name, name, sizeof(cache->uc_name));
 440         /* upcall pathname proc tunable */
 441         strlcpy(cache->uc_upcall, upcall, sizeof(cache->uc_upcall));
 442         cache->uc_entry_expire = 20 * 60;
 443         cache->uc_acquire_expire = 30;
 444         cache->uc_ops = ops;
 445
 446         RETURN(cache);
 447 }
 448 EXPORT_SYMBOL(upcall_cache_init);
 449
 450 void upcall_cache_cleanup(struct upcall_cache *cache)
 451 {
 452         if (!cache)
 453                 return;
 454         upcall_cache_flush_all(cache);
 455         LIBCFS_FREE(cache, sizeof(*cache));
 456 }
 457 EXPORT_SYMBOL(upcall_cache_cleanup);