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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/upcall_cache.c
  37  *
  38  * Supplementary groups cache.
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_SEC
  42
  43 #ifndef AUTOCONF_INCLUDED
  44 #include <linux/config.h>
  45 #endif
  46 #include <linux/module.h>
  47 #include <linux/kernel.h>
  48 #include <linux/mm.h>
  49 #include <linux/kmod.h>
  50 #include <linux/string.h>
  51 #include <linux/stat.h>
  52 #include <linux/errno.h>
  53 #include <linux/version.h>
  54 #include <linux/unistd.h>
  55
  56 #include <asm/system.h>
  57 #include <asm/uaccess.h>
  58
  59 #include <linux/fs.h>
  60 #include <linux/stat.h>
  61 #include <asm/uaccess.h>
  62 #include <linux/slab.h>
  63
  64 #include <obd_support.h>
  65 #include <lustre_lib.h>
  66
  67 static struct upcall_cache_entry *alloc_entry(struct upcall_cache *cache,
  68                                               __u64 key, void *args)
  69 {
  70         struct upcall_cache_entry *entry;
  71
  72         OBD_ALLOC_PTR(entry);
  73         if (!entry)
  74                 return NULL;
  75
  76         UC_CACHE_SET_NEW(entry);
  77         CFS_INIT_LIST_HEAD(&entry->ue_hash);
  78         entry->ue_key = key;
  79         cfs_atomic_set(&entry->ue_refcount, 0);
  80         cfs_waitq_init(&entry->ue_waitq);
  81         if (cache->uc_ops->init_entry)
  82                 cache->uc_ops->init_entry(entry, args);
  83         return entry;
  84 }
  85
  86 /* protected by cache lock */
  87 static void free_entry(struct upcall_cache *cache,
  88                        struct upcall_cache_entry *entry)
  89 {
  90         if (cache->uc_ops->free_entry)
  91                 cache->uc_ops->free_entry(cache, entry);
  92
  93         cfs_list_del(&entry->ue_hash);
  94         CDEBUG(D_OTHER, "destroy cache entry %p for key "LPU64"\n",
  95                entry, entry->ue_key);
  96         OBD_FREE_PTR(entry);
  97 }
  98
  99 static inline int upcall_compare(struct upcall_cache *cache,
 100                                  struct upcall_cache_entry *entry,
 101                                  __u64 key, void *args)
 102 {
 103         if (entry->ue_key != key)
 104                 return -1;
 105
 106         if (cache->uc_ops->upcall_compare)
 107                 return cache->uc_ops->upcall_compare(cache, entry, key, args);
 108
 109         return 0;
 110 }
 111
 112 static inline int downcall_compare(struct upcall_cache *cache,
 113                                    struct upcall_cache_entry *entry,
 114                                    __u64 key, void *args)
 115 {
 116         if (entry->ue_key != key)
 117                 return -1;
 118
 119         if (cache->uc_ops->downcall_compare)
 120                 return cache->uc_ops->downcall_compare(cache, entry, key, args);
 121
 122         return 0;
 123 }
 124
 125 static inline void get_entry(struct upcall_cache_entry *entry)
 126 {
 127         cfs_atomic_inc(&entry->ue_refcount);
 128 }
 129
 130 static inline void put_entry(struct upcall_cache *cache,
 131                              struct upcall_cache_entry *entry)
 132 {
 133         if (cfs_atomic_dec_and_test(&entry->ue_refcount) &&
 134             (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry))) {
 135                 free_entry(cache, entry);
 136         }
 137 }
 138
 139 static int check_unlink_entry(struct upcall_cache *cache,
 140                               struct upcall_cache_entry *entry)
 141 {
 142         if (UC_CACHE_IS_VALID(entry) &&
 143             cfs_time_before(jiffies, entry->ue_expire))
 144                 return 0;
 145
 146         if (UC_CACHE_IS_ACQUIRING(entry)) {
 147                 if (cfs_time_before(jiffies, entry->ue_acquire_expire))
 148                         return 0;
 149
 150                 UC_CACHE_SET_EXPIRED(entry);
 151                 cfs_waitq_broadcast(&entry->ue_waitq);
 152         } else if (!UC_CACHE_IS_INVALID(entry)) {
 153                 UC_CACHE_SET_EXPIRED(entry);
 154         }
 155
 156         cfs_list_del_init(&entry->ue_hash);
 157         if (!cfs_atomic_read(&entry->ue_refcount))
 158                 free_entry(cache, entry);
 159         return 1;
 160 }
 161
 162 static inline int refresh_entry(struct upcall_cache *cache,
 163                          struct upcall_cache_entry *entry)
 164 {
 165         LASSERT(cache->uc_ops->do_upcall);
 166         return cache->uc_ops->do_upcall(cache, entry);
 167 }
 168
 169 struct upcall_cache_entry *upcall_cache_get_entry(struct upcall_cache *cache,
 170                                                   __u64 key, void *args)
 171 {
 172         struct upcall_cache_entry *entry = NULL, *new = NULL, *next;
 173         cfs_list_t *head;
 174         cfs_waitlink_t wait;
 175         int rc, found;
 176         ENTRY;
 177
 178         LASSERT(cache);
 179
 180         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 181 find_again:
 182         found = 0;
 183         cfs_spin_lock(&cache->uc_lock);
 184         cfs_list_for_each_entry_safe(entry, next, head, ue_hash) {
 185                 /* check invalid & expired items */
 186                 if (check_unlink_entry(cache, entry))
 187                         continue;
 188                 if (upcall_compare(cache, entry, key, args) == 0) {
 189                         found = 1;
 190                         break;
 191                 }
 192         }
 193
 194         if (!found) { /* didn't find it */
 195                 if (!new) {
 196                         cfs_spin_unlock(&cache->uc_lock);
 197                         new = alloc_entry(cache, key, args);
 198                         if (!new) {
 199                                 CERROR("fail to alloc entry\n");
 200                                 RETURN(ERR_PTR(-ENOMEM));
 201                         }
 202                         goto find_again;
 203                 } else {
 204                         cfs_list_add(&new->ue_hash, head);
 205                         entry = new;
 206                 }
 207         } else {
 208                 if (new) {
 209                         free_entry(cache, new);
 210                         new = NULL;
 211                 }
 212                 cfs_list_move(&entry->ue_hash, head);
 213         }
 214         get_entry(entry);
 215
 216         /* acquire for new one */
 217         if (UC_CACHE_IS_NEW(entry)) {
 218                 UC_CACHE_SET_ACQUIRING(entry);
 219                 UC_CACHE_CLEAR_NEW(entry);
 220                 entry->ue_acquire_expire = jiffies + cache->uc_acquire_expire;
 221                 cfs_spin_unlock(&cache->uc_lock);
 222                 rc = refresh_entry(cache, entry);
 223                 cfs_spin_lock(&cache->uc_lock);
 224                 if (rc < 0) {
 225                         UC_CACHE_CLEAR_ACQUIRING(entry);
 226                         UC_CACHE_SET_INVALID(entry);
 227                         if (unlikely(rc == -EREMCHG)) {
 228                                 put_entry(cache, entry);
 229                                 GOTO(out, entry = ERR_PTR(rc));
 230                         }
 231                 }
 232                 /* fall through */
 233         }
 234         /* someone (and only one) is doing upcall upon
 235          * this item, just wait it complete
 236          */
 237         if (UC_CACHE_IS_ACQUIRING(entry)) {
 238                 unsigned long expiry = jiffies + cache->uc_acquire_expire;
 239
 240                 cfs_waitlink_init(&wait);
 241                 cfs_waitq_add(&entry->ue_waitq, &wait);
 242                 cfs_set_current_state(CFS_TASK_INTERRUPTIBLE);
 243                 cfs_spin_unlock(&cache->uc_lock);
 244
 245                 cfs_waitq_timedwait(&wait, CFS_TASK_INTERRUPTIBLE,
 246                                     cache->uc_acquire_expire);
 247
 248                 cfs_spin_lock(&cache->uc_lock);
 249                 cfs_waitq_del(&entry->ue_waitq, &wait);
 250                 if (UC_CACHE_IS_ACQUIRING(entry)) {
 251                         /* we're interrupted or upcall failed in the middle */
 252                         rc = cfs_time_before(jiffies, expiry) ? \
 253                                 -EINTR : -ETIMEDOUT;
 254                         put_entry(cache, entry);
 255                         CERROR("acquire timeout exceeded for key "LPU64
 256                                "\n", entry->ue_key);
 257                         GOTO(out, entry = ERR_PTR(rc));
 258                 }
 259                 /* fall through */
 260         }
 261
 262         /* invalid means error, don't need to try again */
 263         if (UC_CACHE_IS_INVALID(entry)) {
 264                 put_entry(cache, entry);
 265                 GOTO(out, entry = ERR_PTR(-EIDRM));
 266         }
 267
 268         /* check expired
 269          * We can't refresh the existing one because some
 270          * memory might be shared by multiple processes.
 271          */
 272         if (check_unlink_entry(cache, entry)) {
 273                 /* if expired, try again. but if this entry is
 274                  * created by me but too quickly turn to expired
 275                  * without any error, should at least give a
 276                  * chance to use it once.
 277                  */
 278                 if (entry != new) {
 279                         put_entry(cache, entry);
 280                         cfs_spin_unlock(&cache->uc_lock);
 281                         new = NULL;
 282                         goto find_again;
 283                 }
 284         }
 285
 286         /* Now we know it's good */
 287 out:
 288         cfs_spin_unlock(&cache->uc_lock);
 289         RETURN(entry);
 290 }
 291 EXPORT_SYMBOL(upcall_cache_get_entry);
 292
 293 void upcall_cache_put_entry(struct upcall_cache *cache,
 294                             struct upcall_cache_entry *entry)
 295 {
 296         ENTRY;
 297
 298         if (!entry) {
 299                 EXIT;
 300                 return;
 301         }
 302
 303         LASSERT(cfs_atomic_read(&entry->ue_refcount) > 0);
 304         cfs_spin_lock(&cache->uc_lock);
 305         put_entry(cache, entry);
 306         cfs_spin_unlock(&cache->uc_lock);
 307         EXIT;
 308 }
 309 EXPORT_SYMBOL(upcall_cache_put_entry);
 310
 311 int upcall_cache_downcall(struct upcall_cache *cache, __u32 err, __u64 key,
 312                           void *args)
 313 {
 314         struct upcall_cache_entry *entry = NULL;
 315         cfs_list_t *head;
 316         int found = 0, rc = 0;
 317         ENTRY;
 318
 319         LASSERT(cache);
 320
 321         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 322
 323         cfs_spin_lock(&cache->uc_lock);
 324         cfs_list_for_each_entry(entry, head, ue_hash) {
 325                 if (downcall_compare(cache, entry, key, args) == 0) {
 326                         found = 1;
 327                         get_entry(entry);
 328                         break;
 329                 }
 330         }
 331
 332         if (!found) {
 333                 CDEBUG(D_OTHER, "%s: upcall for key "LPU64" not expected\n",
 334                        cache->uc_name, key);
 335                 /* haven't found, it's possible */
 336                 cfs_spin_unlock(&cache->uc_lock);
 337                 RETURN(-EINVAL);
 338         }
 339
 340         if (err) {
 341                 CDEBUG(D_OTHER, "%s: upcall for key "LPU64" returned %d\n",
 342                        cache->uc_name, entry->ue_key, err);
 343                 GOTO(out, rc = -EINVAL);
 344         }
 345
 346         if (!UC_CACHE_IS_ACQUIRING(entry)) {
 347                 CDEBUG(D_RPCTRACE,"%s: found uptodate entry %p (key "LPU64")\n",
 348                        cache->uc_name, entry, entry->ue_key);
 349                 GOTO(out, rc = 0);
 350         }
 351
 352         if (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry)) {
 353                 CERROR("%s: found a stale entry %p (key "LPU64") in ioctl\n",
 354                        cache->uc_name, entry, entry->ue_key);
 355                 GOTO(out, rc = -EINVAL);
 356         }
 357
 358         cfs_spin_unlock(&cache->uc_lock);
 359         if (cache->uc_ops->parse_downcall)
 360                 rc = cache->uc_ops->parse_downcall(cache, entry, args);
 361         cfs_spin_lock(&cache->uc_lock);
 362         if (rc)
 363                 GOTO(out, rc);
 364
 365         entry->ue_expire = jiffies + cache->uc_entry_expire;
 366         UC_CACHE_SET_VALID(entry);
 367         CDEBUG(D_OTHER, "%s: created upcall cache entry %p for key "LPU64"\n",
 368                cache->uc_name, entry, entry->ue_key);
 369 out:
 370         if (rc) {
 371                 UC_CACHE_SET_INVALID(entry);
 372                 cfs_list_del_init(&entry->ue_hash);
 373         }
 374         UC_CACHE_CLEAR_ACQUIRING(entry);
 375         cfs_spin_unlock(&cache->uc_lock);
 376         cfs_waitq_broadcast(&entry->ue_waitq);
 377         put_entry(cache, entry);
 378
 379         RETURN(rc);
 380 }
 381 EXPORT_SYMBOL(upcall_cache_downcall);
 382
 383 static void cache_flush(struct upcall_cache *cache, int force)
 384 {
 385         struct upcall_cache_entry *entry, *next;
 386         int i;
 387         ENTRY;
 388
 389         cfs_spin_lock(&cache->uc_lock);
 390         for (i = 0; i < UC_CACHE_HASH_SIZE; i++) {
 391                 cfs_list_for_each_entry_safe(entry, next,
 392                                          &cache->uc_hashtable[i], ue_hash) {
 393                         if (!force && cfs_atomic_read(&entry->ue_refcount)) {
 394                                 UC_CACHE_SET_EXPIRED(entry);
 395                                 continue;
 396                         }
 397                         LASSERT(!cfs_atomic_read(&entry->ue_refcount));
 398                         free_entry(cache, entry);
 399                 }
 400         }
 401         cfs_spin_unlock(&cache->uc_lock);
 402         EXIT;
 403 }
 404
 405 void upcall_cache_flush_idle(struct upcall_cache *cache)
 406 {
 407         cache_flush(cache, 0);
 408 }
 409 EXPORT_SYMBOL(upcall_cache_flush_idle);
 410
 411 void upcall_cache_flush_all(struct upcall_cache *cache)
 412 {
 413         cache_flush(cache, 1);
 414 }
 415 EXPORT_SYMBOL(upcall_cache_flush_all);
 416
 417 void upcall_cache_flush_one(struct upcall_cache *cache, __u64 key, void *args)
 418 {
 419         cfs_list_t *head;
 420         struct upcall_cache_entry *entry;
 421         int found = 0;
 422         ENTRY;
 423
 424         head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 425
 426         cfs_spin_lock(&cache->uc_lock);
 427         cfs_list_for_each_entry(entry, head, ue_hash) {
 428                 if (upcall_compare(cache, entry, key, args) == 0) {
 429                         found = 1;
 430                         break;
 431                 }
 432         }
 433
 434         if (found) {
 435                 CWARN("%s: flush entry %p: key "LPU64", ref %d, fl %x, "
 436                       "cur %lu, ex %ld/%ld\n",
 437                       cache->uc_name, entry, entry->ue_key,
 438                       cfs_atomic_read(&entry->ue_refcount), entry->ue_flags,
 439                       get_seconds(), entry->ue_acquire_expire,
 440                       entry->ue_expire);
 441                 UC_CACHE_SET_EXPIRED(entry);
 442                 if (!cfs_atomic_read(&entry->ue_refcount))
 443                         free_entry(cache, entry);
 444         }
 445         cfs_spin_unlock(&cache->uc_lock);
 446 }
 447 EXPORT_SYMBOL(upcall_cache_flush_one);
 448
 449 struct upcall_cache *upcall_cache_init(const char *name, const char *upcall,
 450                                        struct upcall_cache_ops *ops)
 451 {
 452         struct upcall_cache *cache;
 453         int i;
 454         ENTRY;
 455
 456         OBD_ALLOC(cache, sizeof(*cache));
 457         if (!cache)
 458                 RETURN(ERR_PTR(-ENOMEM));
 459
 460         cfs_spin_lock_init(&cache->uc_lock);
 461         cfs_rwlock_init(&cache->uc_upcall_rwlock);
 462         for (i = 0; i < UC_CACHE_HASH_SIZE; i++)
 463                 CFS_INIT_LIST_HEAD(&cache->uc_hashtable[i]);
 464         strncpy(cache->uc_name, name, sizeof(cache->uc_name) - 1);
 465         /* upcall pathname proc tunable */
 466         strncpy(cache->uc_upcall, upcall, sizeof(cache->uc_upcall) - 1);
 467         cache->uc_entry_expire = 10 * 60 * CFS_HZ;
 468         cache->uc_acquire_expire = 15 * CFS_HZ;
 469         cache->uc_ops = ops;
 470
 471         RETURN(cache);
 472 }
 473 EXPORT_SYMBOL(upcall_cache_init);
 474
 475 void upcall_cache_cleanup(struct upcall_cache *cache)
 476 {
 477         if (!cache)
 478                 return;
 479         upcall_cache_flush_all(cache);
 480         OBD_FREE(cache, sizeof(*cache));
 481 }
 482 EXPORT_SYMBOL(upcall_cache_cleanup);