lustre/lvfs/upcall_cache.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Supplementary groups cache.
   5  *
   6  *  Copyright (c) 2004 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_SEC
  25
  26 #ifndef AUTOCONF_INCLUDED
  27 #include <linux/config.h>
  28 #endif
  29 #include <linux/module.h>
  30 #include <linux/kernel.h>
  31 #include <linux/mm.h>
  32 #include <linux/kmod.h>
  33 #include <linux/string.h>
  34 #include <linux/stat.h>
  35 #include <linux/errno.h>
  36 #include <linux/version.h>
  37 #include <linux/unistd.h>
  38
  39 #include <asm/system.h>
  40 #include <asm/uaccess.h>
  41
  42 #include <linux/fs.h>
  43 #include <linux/stat.h>
  44 #include <asm/uaccess.h>
  45 #include <linux/slab.h>
  46 #include <asm/segment.h>
  47
  48 #include <obd_support.h>
  49 #include <lustre_lib.h>
  50
  51 static struct upcall_cache_entry *alloc_entry(struct upcall_cache *cache,
  52                                               __u64 key, void *args)
  53 {
  54         struct upcall_cache_entry *entry;
  55
  56         OBD_ALLOC_PTR(entry);
  57         if (!entry)
  58                 return NULL;
  59
  60         UC_CACHE_SET_NEW(entry);
  61         INIT_LIST_HEAD(&entry->ue_hash);
  62         entry->ue_key = key;
  63         atomic_set(&entry->ue_refcount, 0);
  64         init_waitqueue_head(&entry->ue_waitq);
  65         if (cache->uc_ops->init_entry)
  66                 cache->uc_ops->init_entry(entry, args);
  67         return entry;
  68 }
  69
  70 /* protected by cache lock */
  71 static void free_entry(struct upcall_cache *cache,
  72                        struct upcall_cache_entry *entry)
  73 {
  74         if (cache->uc_ops->free_entry)
  75                 cache->uc_ops->free_entry(cache, entry);
  76
  77         list_del(&entry->ue_hash);
  78         CDEBUG(D_OTHER, "destroy cache entry %p for key "LPU64"\n",
  79                entry, entry->ue_key);
  80         OBD_FREE_PTR(entry);
  81 }
  82
  83 static inline int upcall_compare(struct upcall_cache *cache,
  84                                  struct upcall_cache_entry *entry,
  85                                  __u64 key, void *args)
  86 {
  87         if (entry->ue_key != key)
  88                 return -1;
  89
  90         if (cache->uc_ops->upcall_compare)
  91                 return cache->uc_ops->upcall_compare(cache, entry, key, args);
  92
  93         return 0;
  94 }
  95
  96 static inline int downcall_compare(struct upcall_cache *cache,
  97                                    struct upcall_cache_entry *entry,
  98                                    __u64 key, void *args)
  99 {
 100         if (entry->ue_key != key)
 101                 return -1;
 102
 103         if (cache->uc_ops->downcall_compare)
 104                 return cache->uc_ops->downcall_compare(cache, entry, key, args);
 105
 106         return 0;
 107 }
 108
 109 static inline void get_entry(struct upcall_cache_entry *entry)
 110 {
 111         atomic_inc(&entry->ue_refcount);
 112 }
 113
 114 static inline void put_entry(struct upcall_cache *cache,
 115                              struct upcall_cache_entry *entry)
 116 {
 117         if (atomic_dec_and_test(&entry->ue_refcount) &&
 118             (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry))) {
 119                 free_entry(cache, entry);
 120         }
 121 }
 122
 123 static int check_unlink_entry(struct upcall_cache *cache,
 124                               struct upcall_cache_entry *entry)
 125 {
 126         if (UC_CACHE_IS_VALID(entry) &&
 127             time_before(jiffies, entry->ue_expire))
 128                 return 0;
 129
 130         if (UC_CACHE_IS_ACQUIRING(entry)) {
 131                 if (time_before(jiffies, entry->ue_acquire_expire))
 132                         return 0;
 133
 134                 UC_CACHE_SET_EXPIRED(entry);
 135                 wake_up_all(&entry->ue_waitq);
 136         } else if (!UC_CACHE_IS_INVALID(entry)) {
 137                 UC_CACHE_SET_EXPIRED(entry);
 138         }
 139
 140         list_del_init(&entry->ue_hash);
 141         if (!atomic_read(&entry->ue_refcount))
 142                 free_entry(cache, entry);
 143         return 1;
 144 }
 145
 146 static inline int refresh_entry(struct upcall_cache *cache,
 147                          struct upcall_cache_entry *entry)
 148 {
 149         LASSERT(cache->uc_ops->do_upcall);
 150         return cache->uc_ops->do_upcall(cache, entry);
 151 }
 152
 153 struct upcall_cache_entry *upcall_cache_get_entry(struct upcall_cache *cache,
 154                                                   __u64 key, void *args)
 155 {
 156         struct upcall_cache_entry *entry = NULL, *new = NULL, *next;
 157         struct list_head *head;
 158         wait_queue_t wait;
 159         int rc, found;
 160         ENTRY;
 161
 162         LASSERT(cache);
 163
 164         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 165 find_again:
 166         found = 0;
 167         spin_lock(&cache->uc_lock);
 168         list_for_each_entry_safe(entry, next, head, ue_hash) {
 169                 /* check invalid & expired items */
 170                 if (check_unlink_entry(cache, entry))
 171                         continue;
 172                 if (upcall_compare(cache, entry, key, args) == 0) {
 173                         found = 1;
 174                         break;
 175                 }
 176         }
 177
 178         if (!found) { /* didn't find it */
 179                 if (!new) {
 180                         spin_unlock(&cache->uc_lock);
 181                         new = alloc_entry(cache, key, args);
 182                         if (!new) {
 183                                 CERROR("fail to alloc entry\n");
 184                                 RETURN(ERR_PTR(-ENOMEM));
 185                         }
 186                         goto find_again;
 187                 } else {
 188                         list_add(&new->ue_hash, head);
 189                         entry = new;
 190                 }
 191         } else {
 192                 if (new) {
 193                         free_entry(cache, new);
 194                         new = NULL;
 195                 }
 196                 list_move(&entry->ue_hash, head);
 197         }
 198         get_entry(entry);
 199
 200         /* acquire for new one */
 201         if (UC_CACHE_IS_NEW(entry)) {
 202                 UC_CACHE_SET_ACQUIRING(entry);
 203                 UC_CACHE_CLEAR_NEW(entry);
 204                 entry->ue_acquire_expire = jiffies + cache->uc_acquire_expire;
 205                 spin_unlock(&cache->uc_lock);
 206                 rc = refresh_entry(cache, entry);
 207                 spin_lock(&cache->uc_lock);
 208                 if (rc < 0) {
 209                         UC_CACHE_CLEAR_ACQUIRING(entry);
 210                         UC_CACHE_SET_INVALID(entry);
 211                 }
 212                 /* fall through */
 213         }
 214         /* someone (and only one) is doing upcall upon
 215          * this item, just wait it complete
 216          */
 217         if (UC_CACHE_IS_ACQUIRING(entry)) {
 218                 unsigned long expiry = jiffies + cache->uc_acquire_expire;
 219
 220                 init_waitqueue_entry(&wait, current);
 221                 add_wait_queue(&entry->ue_waitq, &wait);
 222                 set_current_state(TASK_INTERRUPTIBLE);
 223                 spin_unlock(&cache->uc_lock);
 224
 225                 schedule_timeout(cache->uc_acquire_expire);
 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 = time_before(jiffies, expiry) ? -EINTR : -ETIMEDOUT;
 232                         put_entry(cache, entry);
 233                         CERROR("acquire timeout exceeded for key "LPU64
 234                                "\n", entry->ue_key);
 235                         GOTO(out, entry = ERR_PTR(rc));
 236                 }
 237                 /* fall through */
 238         }
 239
 240         /* invalid means error, don't need to try again */
 241         if (UC_CACHE_IS_INVALID(entry)) {
 242                 put_entry(cache, entry);
 243                 GOTO(out, entry = ERR_PTR(-EIDRM));
 244         }
 245
 246         /* check expired
 247          * We can't refresh the existing one because some
 248          * memory might be shared by multiple processes.
 249          */
 250         if (check_unlink_entry(cache, entry)) {
 251                 /* if expired, try again. but if this entry is
 252                  * created by me but too quickly turn to expired
 253                  * without any error, should at least give a
 254                  * chance to use it once.
 255                  */
 256                 if (entry != new) {
 257                         put_entry(cache, entry);
 258                         spin_unlock(&cache->uc_lock);
 259                         new = NULL;
 260                         goto find_again;
 261                 }
 262         }
 263
 264         /* Now we know it's good */
 265 out:
 266         spin_unlock(&cache->uc_lock);
 267         RETURN(entry);
 268 }
 269 EXPORT_SYMBOL(upcall_cache_get_entry);
 270
 271 void upcall_cache_put_entry(struct upcall_cache *cache,
 272                             struct upcall_cache_entry *entry)
 273 {
 274         ENTRY;
 275
 276         if (!entry) {
 277                 EXIT;
 278                 return;
 279         }
 280
 281         LASSERT(atomic_read(&entry->ue_refcount) > 0);
 282         spin_lock(&cache->uc_lock);
 283         put_entry(cache, entry);
 284         spin_unlock(&cache->uc_lock);
 285         EXIT;
 286 }
 287 EXPORT_SYMBOL(upcall_cache_put_entry);
 288
 289 int upcall_cache_downcall(struct upcall_cache *cache, __u32 err, __u64 key,
 290                           void *args)
 291 {
 292         struct upcall_cache_entry *entry = NULL;
 293         struct list_head *head;
 294         int found = 0, rc = 0;
 295         ENTRY;
 296
 297         LASSERT(cache);
 298
 299         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 300
 301         spin_lock(&cache->uc_lock);
 302         list_for_each_entry(entry, head, ue_hash) {
 303                 if (downcall_compare(cache, entry, key, args) == 0) {
 304                         found = 1;
 305                         get_entry(entry);
 306                         break;
 307                 }
 308         }
 309
 310         if (!found) {
 311                 CDEBUG(D_OTHER, "%s: upcall for key "LPU64" not expected\n",
 312                        cache->uc_name, key);
 313                 /* haven't found, it's possible */
 314                 spin_unlock(&cache->uc_lock);
 315                 RETURN(-EINVAL);
 316         }
 317
 318         if (err) {
 319                 CDEBUG(D_OTHER, "%s: upcall for key "LPU64" returned %d\n",
 320                        cache->uc_name, entry->ue_key, err);
 321                 GOTO(out, rc = -EINVAL);
 322         }
 323
 324         if (!UC_CACHE_IS_ACQUIRING(entry)) {
 325                 CDEBUG(D_HA, "%s: found uptodate entry %p (key "LPU64")\n",
 326                        cache->uc_name, entry, entry->ue_key);
 327                 GOTO(out, rc = 0);
 328         }
 329
 330         if (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry)) {
 331                 CERROR("%s: found a stale entry %p (key "LPU64") in ioctl\n",
 332                        cache->uc_name, entry, entry->ue_key);
 333                 GOTO(out, rc = -EINVAL);
 334         }
 335
 336         spin_unlock(&cache->uc_lock);
 337         if (cache->uc_ops->parse_downcall)
 338                 rc = cache->uc_ops->parse_downcall(cache, entry, args);
 339         spin_lock(&cache->uc_lock);
 340         if (rc)
 341                 GOTO(out, rc);
 342
 343         entry->ue_expire = jiffies + cache->uc_entry_expire;
 344         UC_CACHE_SET_VALID(entry);
 345         CDEBUG(D_OTHER, "%s: created upcall cache entry %p for key "LPU64"\n",
 346                cache->uc_name, entry, entry->ue_key);
 347 out:
 348         if (rc) {
 349                 UC_CACHE_SET_INVALID(entry);
 350                 list_del_init(&entry->ue_hash);
 351         }
 352         UC_CACHE_CLEAR_ACQUIRING(entry);
 353         spin_unlock(&cache->uc_lock);
 354         wake_up_all(&entry->ue_waitq);
 355         put_entry(cache, entry);
 356
 357         RETURN(rc);
 358 }
 359 EXPORT_SYMBOL(upcall_cache_downcall);
 360
 361 static void cache_flush(struct upcall_cache *cache, int force)
 362 {
 363         struct upcall_cache_entry *entry, *next;
 364         int i;
 365         ENTRY;
 366
 367         spin_lock(&cache->uc_lock);
 368         for (i = 0; i < UC_CACHE_HASH_SIZE; i++) {
 369                 list_for_each_entry_safe(entry, next,
 370                                          &cache->uc_hashtable[i], ue_hash) {
 371                         if (!force && atomic_read(&entry->ue_refcount)) {
 372                                 UC_CACHE_SET_EXPIRED(entry);
 373                                 continue;
 374                         }
 375                         LASSERT(!atomic_read(&entry->ue_refcount));
 376                         free_entry(cache, entry);
 377                 }
 378         }
 379         spin_unlock(&cache->uc_lock);
 380         EXIT;
 381 }
 382
 383 void upcall_cache_flush_idle(struct upcall_cache *cache)
 384 {
 385         cache_flush(cache, 0);
 386 }
 387 EXPORT_SYMBOL(upcall_cache_flush_idle);
 388
 389 void upcall_cache_flush_all(struct upcall_cache *cache)
 390 {
 391         cache_flush(cache, 1);
 392 }
 393 EXPORT_SYMBOL(upcall_cache_flush_all);
 394
 395 void upcall_cache_flush_one(struct upcall_cache *cache, __u64 key, void *args)
 396 {
 397         struct list_head *head;
 398         struct upcall_cache_entry *entry;
 399         int found = 0;
 400         ENTRY;
 401
 402         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 403
 404         spin_lock(&cache->uc_lock);
 405         list_for_each_entry(entry, head, ue_hash) {
 406                 if (upcall_compare(cache, entry, key, args) == 0) {
 407                         found = 1;
 408                         break;
 409                 }
 410         }
 411
 412         if (found) {
 413                 CWARN("%s: flush entry %p: key "LPU64", ref %d, fl %x, "
 414                       "cur %lu, ex %ld/%ld\n",
 415                       cache->uc_name, entry, entry->ue_key,
 416                       atomic_read(&entry->ue_refcount), entry->ue_flags,
 417                       get_seconds(), entry->ue_acquire_expire,
 418                       entry->ue_expire);
 419                 UC_CACHE_SET_EXPIRED(entry);
 420                 if (!atomic_read(&entry->ue_refcount))
 421                         free_entry(cache, entry);
 422         }
 423         spin_unlock(&cache->uc_lock);
 424 }
 425 EXPORT_SYMBOL(upcall_cache_flush_one);
 426
 427 struct upcall_cache *upcall_cache_init(const char *name, const char *upcall,
 428                                        struct upcall_cache_ops *ops)
 429 {
 430         struct upcall_cache *cache;
 431         int i;
 432         ENTRY;
 433
 434         OBD_ALLOC(cache, sizeof(*cache));
 435         if (!cache)
 436                 RETURN(ERR_PTR(-ENOMEM));
 437
 438         spin_lock_init(&cache->uc_lock);
 439         for (i = 0; i < UC_CACHE_HASH_SIZE; i++)
 440                 INIT_LIST_HEAD(&cache->uc_hashtable[i]);
 441         strncpy(cache->uc_name, name, sizeof(cache->uc_name) - 1);
 442         /* upcall pathname proc tunable */
 443         strncpy(cache->uc_upcall, upcall, sizeof(cache->uc_upcall) - 1);
 444         cache->uc_entry_expire = 10 * 60 * HZ;
 445         cache->uc_acquire_expire = 15 * HZ;
 446         cache->uc_ops = ops;
 447
 448         RETURN(cache);
 449 }
 450 EXPORT_SYMBOL(upcall_cache_init);
 451
 452 void upcall_cache_cleanup(struct upcall_cache *cache)
 453 {
 454         if (!cache)
 455                 return;
 456         upcall_cache_flush_all(cache);
 457         OBD_FREE(cache, sizeof(*cache));
 458 }
 459 EXPORT_SYMBOL(upcall_cache_cleanup);