Whamcloud - gitweb
- make HEAD from b_post_cmd3
[fs/lustre-release.git] / lustre / lvfs / upcall_cache.c
index 2db5d90..71a0022 100644 (file)
@@ -76,11 +76,12 @@ void groups_free(struct group_info *ginfo)
 }
 #endif
 
-static struct upcall_cache_entry *alloc_entry(__u64 key)
+static struct upcall_cache_entry *alloc_entry(struct upcall_cache *cache,
+                                              __u64 key, void *args)
 {
         struct upcall_cache_entry *entry;
 
-        OBD_ALLOC(entry, sizeof(*entry));
+        OBD_ALLOC_PTR(entry);
         if (!entry)
                 return NULL;
 
@@ -89,34 +90,66 @@ static struct upcall_cache_entry *alloc_entry(__u64 key)
         entry->ue_key = key;
         atomic_set(&entry->ue_refcount, 0);
         init_waitqueue_head(&entry->ue_waitq);
+        if (cache->uc_ops->init_entry)
+                cache->uc_ops->init_entry(entry, args);
         return entry;
 }
 
-/* protected by hash lock */
-static void free_entry(struct upcall_cache_entry *entry)
+/* protected by cache lock */
+static void free_entry(struct upcall_cache *cache,
+                       struct upcall_cache_entry *entry)
 {
-        if (entry->ue_group_info)
-                groups_free(entry->ue_group_info);
+        if (cache->uc_ops->free_entry)
+                cache->uc_ops->free_entry(cache, entry);
+
         list_del(&entry->ue_hash);
         CDEBUG(D_OTHER, "destroy cache entry %p for key "LPU64"\n",
                entry, entry->ue_key);
-        OBD_FREE(entry, sizeof(*entry));
+        OBD_FREE_PTR(entry);
+}
+
+static inline int upcall_compare(struct upcall_cache *cache,
+                                 struct upcall_cache_entry *entry,
+                                 __u64 key, void *args)
+{
+        if (entry->ue_key != key)
+                return -1;
+
+        if (cache->uc_ops->upcall_compare)
+                return cache->uc_ops->upcall_compare(cache, entry, key, args);
+
+        return 0;
+}
+
+static inline int downcall_compare(struct upcall_cache *cache,
+                                   struct upcall_cache_entry *entry,
+                                   __u64 key, void *args)
+{
+        if (entry->ue_key != key)
+                return -1;
+
+        if (cache->uc_ops->downcall_compare)
+                return cache->uc_ops->downcall_compare(cache, entry, key, args);
+
+        return 0;
 }
 
-static void get_entry(struct upcall_cache_entry *entry)
+static inline void get_entry(struct upcall_cache_entry *entry)
 {
         atomic_inc(&entry->ue_refcount);
 }
 
-static void put_entry(struct upcall_cache_entry *entry)
+static inline void put_entry(struct upcall_cache *cache,
+                             struct upcall_cache_entry *entry)
 {
         if (atomic_dec_and_test(&entry->ue_refcount) &&
             (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry))) {
-                free_entry(entry);
+                free_entry(cache, entry);
         }
 }
 
-static int check_unlink_entry(struct upcall_cache_entry *entry)
+static int check_unlink_entry(struct upcall_cache *cache,
+                              struct upcall_cache_entry *entry)
 {
         if (UC_CACHE_IS_VALID(entry) &&
             time_before(jiffies, entry->ue_expire))
@@ -134,86 +167,19 @@ static int check_unlink_entry(struct upcall_cache_entry *entry)
 
         list_del_init(&entry->ue_hash);
         if (!atomic_read(&entry->ue_refcount))
-                free_entry(entry);
+                free_entry(cache, entry);
         return 1;
 }
 
-static int refresh_entry(struct upcall_cache *hash,
+static inline int refresh_entry(struct upcall_cache *cache,
                          struct upcall_cache_entry *entry)
 {
-        char *argv[4];
-        char *envp[3];
-        char keystr[16];
-        int rc;
-        ENTRY;
-
-        snprintf(keystr, 16, LPU64, entry->ue_key);
-
-        CDEBUG(D_INFO, "The groups upcall is: %s \n", hash->uc_upcall);
-        argv[0] = hash->uc_upcall;
-        argv[1] = hash->uc_name;
-        argv[2] = keystr;
-        argv[3] = NULL;
-
-        envp[0] = "HOME=/";
-        envp[1] = "PATH=/sbin:/usr/sbin";
-        envp[2] = NULL;
-
-        rc = USERMODEHELPER(argv[0], argv, envp);
-        if (rc < 0) {
-                CERROR("%s: error invoking getgroups upcall %s %s %s: rc %d; "
-                       "check /proc/fs/lustre/mds/%s/group_upcall\n",
-                       hash->uc_name, argv[0], argv[1], argv[2], rc, argv[1]);
-        } else {
-                CDEBUG(D_HA, "%s: invoked upcall %s %s %s\n", hash->uc_name,
-                       argv[0], argv[1], argv[2]);
-                rc = 0;
-        }
-        RETURN(rc);
+        LASSERT(cache->uc_ops->do_upcall);
+        return cache->uc_ops->do_upcall(cache, entry);
 }
 
-static int entry_set_group_info(struct upcall_cache_entry *entry, __u32 primary,
-                                __u32 ngroups, __u32 *groups)
-{
-        struct group_info *ginfo;
-        int i, j;
-        ENTRY;
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,4)
-        if (ngroups > NGROUPS)
-                ngroups = NGROUPS;
-#endif
-
-        if (ngroups > NGROUPS_MAX) {
-                CERROR("using first %d supplementary groups for uid "LPU64"\n",
-                       NGROUPS_MAX, entry->ue_key);
-                ngroups = NGROUPS_MAX;
-        }
-
-        ginfo = groups_alloc(ngroups);
-        if (!ginfo) {
-                CERROR("uid "LPU64" update can't alloc ginfo for %d groups\n",
-                       entry->ue_key, ngroups);
-                RETURN(-ENOMEM);
-        }
-        entry->ue_group_info = ginfo;
-        entry->ue_primary = primary;
-
-        for (i = 0; i < ginfo->nblocks; i++) {
-                int cp_count = min(NGROUPS_PER_BLOCK, (int)ngroups);
-                int off = i * NGROUPS_PER_BLOCK;
-
-                for (j = 0; j < cp_count; j++)
-                        ginfo->blocks[i][j] = groups[off + j];
-
-                ngroups -= cp_count;
-        }
-        RETURN(0);
-}
-
-struct upcall_cache_entry *upcall_cache_get_entry(struct upcall_cache *hash,
-                                                  __u64 key, __u32 primary,
-                                                  __u32 ngroups, __u32 *groups)
+struct upcall_cache_entry *upcall_cache_get_entry(struct upcall_cache *cache,
+                                                  __u64 key, void *args)
 {
         struct upcall_cache_entry *entry = NULL, *new = NULL, *next;
         struct list_head *head;
@@ -221,49 +187,17 @@ struct upcall_cache_entry *upcall_cache_get_entry(struct upcall_cache *hash,
         int rc, found;
         ENTRY;
 
-        LASSERT(hash);
+        LASSERT(cache);
 
-        if (strcmp(hash->uc_upcall, "NONE") == 0) {
-                new = alloc_entry(key);
-                if (!new) {
-                        CERROR("fail to alloc entry\n");
-                        RETURN(NULL);
-                }
-                get_entry(new);
-
-                /* We have to sort the groups for 2.6 kernels */
-                LASSERT(ngroups <= 2);
-                if (ngroups == 2 && groups[1] == -1)
-                        ngroups--;
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
-                /* 2.6 needs groups array sorted */
-                if (ngroups == 2 && groups[0] > groups[1]) {
-                        __u32 tmp = groups[1];
-                        groups[1] = groups[0];
-                        groups[0] = tmp;
-                }
-#endif
-                if (ngroups > 0 && groups[0] == -1) {
-                        groups[0] = groups[1];
-                        ngroups--;
-                }
-
-                rc = entry_set_group_info(new, primary, ngroups, groups);
-
-                /* We can't cache this entry as it only has a subset of
-                 * the user's groups, as sent in suppgid1, suppgid2. */
-                UC_CACHE_SET_EXPIRED(new);
-                RETURN(new);
-        }
-        head = &hash->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
+        head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 find_again:
         found = 0;
-        spin_lock(&hash->uc_lock);
+        spin_lock(&cache->uc_lock);
         list_for_each_entry_safe(entry, next, head, ue_hash) {
                 /* check invalid & expired items */
-                if (check_unlink_entry(entry))
+                if (check_unlink_entry(cache, entry))
                         continue;
-                if (entry->ue_key == key) {
+                if (upcall_compare(cache, entry, key, args) == 0) {
                         found = 1;
                         break;
                 }
@@ -271,8 +205,8 @@ find_again:
 
         if (!found) { /* didn't find it */
                 if (!new) {
-                        spin_unlock(&hash->uc_lock);
-                        new = alloc_entry(key);
+                        spin_unlock(&cache->uc_lock);
+                        new = alloc_entry(cache, key, args);
                         if (!new) {
                                 CERROR("fail to alloc entry\n");
                                 RETURN(ERR_PTR(-ENOMEM));
@@ -284,7 +218,7 @@ find_again:
                 }
         } else {
                 if (new) {
-                        free_entry(new);
+                        free_entry(cache, new);
                         new = NULL;
                 }
                 list_move(&entry->ue_hash, head);
@@ -295,10 +229,10 @@ find_again:
         if (UC_CACHE_IS_NEW(entry)) {
                 UC_CACHE_SET_ACQUIRING(entry);
                 UC_CACHE_CLEAR_NEW(entry);
-                entry->ue_acquire_expire = jiffies + hash->uc_acquire_expire;
-                spin_unlock(&hash->uc_lock);
-                rc = refresh_entry(hash, entry);
-                spin_lock(&hash->uc_lock);
+                entry->ue_acquire_expire = jiffies + cache->uc_acquire_expire;
+                spin_unlock(&cache->uc_lock);
+                rc = refresh_entry(cache, entry);
+                spin_lock(&cache->uc_lock);
                 if (rc < 0) {
                         UC_CACHE_CLEAR_ACQUIRING(entry);
                         UC_CACHE_SET_INVALID(entry);
@@ -309,32 +243,31 @@ find_again:
          * this item, just wait it complete
          */
         if (UC_CACHE_IS_ACQUIRING(entry)) {
+                unsigned long expiry = jiffies + cache->uc_acquire_expire;
+
                 init_waitqueue_entry(&wait, current);
                 add_wait_queue(&entry->ue_waitq, &wait);
                 set_current_state(TASK_INTERRUPTIBLE);
-                spin_unlock(&hash->uc_lock);
+                spin_unlock(&cache->uc_lock);
 
-                schedule_timeout(hash->uc_acquire_expire);
+                schedule_timeout(cache->uc_acquire_expire);
 
-                spin_lock(&hash->uc_lock);
+                spin_lock(&cache->uc_lock);
                 remove_wait_queue(&entry->ue_waitq, &wait);
                 if (UC_CACHE_IS_ACQUIRING(entry)) {
-                        static unsigned long next;
                         /* we're interrupted or upcall failed in the middle */
-                        if (time_after(jiffies, next)) {
-                                CERROR("acquire timeout exceeded for key "LPU64
-                                       "\n", entry->ue_key);
-                                next = jiffies + 1800;
-                        }
-                        put_entry(entry);
-                        GOTO(out, entry = ERR_PTR(-EIDRM));
+                        rc = time_before(jiffies, expiry) ? -EINTR : -ETIMEDOUT;
+                        put_entry(cache, entry);
+                        CERROR("acquire timeout exceeded for key "LPU64
+                               "\n", entry->ue_key);
+                        GOTO(out, entry = ERR_PTR(rc));
                 }
                 /* fall through */
         }
 
         /* invalid means error, don't need to try again */
         if (UC_CACHE_IS_INVALID(entry)) {
-                put_entry(entry);
+                put_entry(cache, entry);
                 GOTO(out, entry = ERR_PTR(-EIDRM));
         }
 
@@ -342,15 +275,15 @@ find_again:
          * We can't refresh the existing one because some
          * memory might be shared by multiple processes.
          */
-        if (check_unlink_entry(entry)) {
+        if (check_unlink_entry(cache, entry)) {
                 /* if expired, try again. but if this entry is
                  * created by me but too quickly turn to expired
                  * without any error, should at least give a
                  * chance to use it once.
                  */
                 if (entry != new) {
-                        put_entry(entry);
-                        spin_unlock(&hash->uc_lock);
+                        put_entry(cache, entry);
+                        spin_unlock(&cache->uc_lock);
                         new = NULL;
                         goto find_again;
                 }
@@ -358,12 +291,12 @@ find_again:
 
         /* Now we know it's good */
 out:
-        spin_unlock(&hash->uc_lock);
+        spin_unlock(&cache->uc_lock);
         RETURN(entry);
 }
 EXPORT_SYMBOL(upcall_cache_get_entry);
 
-void upcall_cache_put_entry(struct upcall_cache *hash,
+void upcall_cache_put_entry(struct upcall_cache *cache,
                             struct upcall_cache_entry *entry)
 {
         ENTRY;
@@ -374,28 +307,28 @@ void upcall_cache_put_entry(struct upcall_cache *hash,
         }
 
         LASSERT(atomic_read(&entry->ue_refcount) > 0);
-        spin_lock(&hash->uc_lock);
-        put_entry(entry);
-        spin_unlock(&hash->uc_lock);
+        spin_lock(&cache->uc_lock);
+        put_entry(cache, entry);
+        spin_unlock(&cache->uc_lock);
         EXIT;
 }
 EXPORT_SYMBOL(upcall_cache_put_entry);
 
-int upcall_cache_downcall(struct upcall_cache *hash, __u32 err, __u64 key,
-                          __u32 primary, __u32 ngroups, __u32 *groups)
+int upcall_cache_downcall(struct upcall_cache *cache, __u32 err, __u64 key,
+                          void *args)
 {
         struct upcall_cache_entry *entry = NULL;
         struct list_head *head;
         int found = 0, rc = 0;
         ENTRY;
 
-        LASSERT(hash);
+        LASSERT(cache);
 
-        head = &hash->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
+        head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
 
-        spin_lock(&hash->uc_lock);
+        spin_lock(&cache->uc_lock);
         list_for_each_entry(entry, head, ue_hash) {
-                if (entry->ue_key == key) {
+                if (downcall_compare(cache, entry, key, args) == 0) {
                         found = 1;
                         get_entry(entry);
                         break;
@@ -404,73 +337,74 @@ int upcall_cache_downcall(struct upcall_cache *hash, __u32 err, __u64 key,
 
         if (!found) {
                 CDEBUG(D_OTHER, "%s: upcall for key "LPU64" not expected\n",
-                       hash->uc_name, entry->ue_key);
+                       cache->uc_name, key);
                 /* haven't found, it's possible */
-                spin_unlock(&hash->uc_lock);
+                spin_unlock(&cache->uc_lock);
                 RETURN(-EINVAL);
         }
 
         if (err) {
                 CDEBUG(D_OTHER, "%s: upcall for key "LPU64" returned %d\n",
-                       hash->uc_name, entry->ue_key, err);
+                       cache->uc_name, entry->ue_key, err);
                 GOTO(out, rc = -EINVAL);
         }
 
         if (!UC_CACHE_IS_ACQUIRING(entry)) {
                 CDEBUG(D_HA, "%s: found uptodate entry %p (key "LPU64")\n",
-                       hash->uc_name, entry, entry->ue_key);
+                       cache->uc_name, entry, entry->ue_key);
                 GOTO(out, rc = 0);
         }
 
         if (UC_CACHE_IS_INVALID(entry) || UC_CACHE_IS_EXPIRED(entry)) {
                 CERROR("%s: found a stale entry %p (key "LPU64") in ioctl\n",
-                       hash->uc_name, entry, entry->ue_key);
+                       cache->uc_name, entry, entry->ue_key);
                 GOTO(out, rc = -EINVAL);
         }
 
-        spin_unlock(&hash->uc_lock);
-        rc = entry_set_group_info(entry, primary, ngroups, groups);
-        spin_lock(&hash->uc_lock);
+        spin_unlock(&cache->uc_lock);
+        if (cache->uc_ops->parse_downcall)
+                rc = cache->uc_ops->parse_downcall(cache, entry, args);
+        spin_lock(&cache->uc_lock);
         if (rc)
                 GOTO(out, rc);
 
-        entry->ue_expire = jiffies + hash->uc_entry_expire;
+        entry->ue_expire = jiffies + cache->uc_entry_expire;
         UC_CACHE_SET_VALID(entry);
         CDEBUG(D_OTHER, "%s: created upcall cache entry %p for key "LPU64"\n",
-               hash->uc_name, entry, entry->ue_key);
+               cache->uc_name, entry, entry->ue_key);
 out:
         if (rc) {
                 UC_CACHE_SET_INVALID(entry);
                 list_del_init(&entry->ue_hash);
         }
         UC_CACHE_CLEAR_ACQUIRING(entry);
-        spin_unlock(&hash->uc_lock);
+        spin_unlock(&cache->uc_lock);
         wake_up_all(&entry->ue_waitq);
-        put_entry(entry);
+        put_entry(cache, entry);
 
         RETURN(rc);
 }
 EXPORT_SYMBOL(upcall_cache_downcall);
 
-static void cache_flush(struct upcall_cache *hash, int force)
+static void cache_flush(struct upcall_cache *cache, int force)
 {
         struct upcall_cache_entry *entry, *next;
         int i;
         ENTRY;
 
-        spin_lock(&hash->uc_lock);
+        spin_lock(&cache->uc_lock);
         for (i = 0; i < UC_CACHE_HASH_SIZE; i++) {
                 list_for_each_entry_safe(entry, next,
-                                         &hash->uc_hashtable[i], ue_hash) {
+                                         &cache->uc_hashtable[i], ue_hash) {
                         if (!force && atomic_read(&entry->ue_refcount)) {
                                 UC_CACHE_SET_EXPIRED(entry);
                                 continue;
                         }
                         LASSERT(!atomic_read(&entry->ue_refcount));
-                        free_entry(entry);
+                        free_entry(cache, entry);
                 }
         }
-        spin_unlock(&hash->uc_lock);
+        spin_unlock(&cache->uc_lock);
         EXIT;
 }
 
@@ -486,34 +420,68 @@ void upcall_cache_flush_all(struct upcall_cache *cache)
 }
 EXPORT_SYMBOL(upcall_cache_flush_all);
 
-struct upcall_cache *upcall_cache_init(const char *name)
+void upcall_cache_flush_one(struct upcall_cache *cache, __u64 key, void *args)
+{
+        struct list_head *head;
+        struct upcall_cache_entry *entry;
+        int found = 0;
+        ENTRY;
+
+        head = &cache->uc_hashtable[UC_CACHE_HASH_INDEX(key)];
+
+        spin_lock(&cache->uc_lock);
+        list_for_each_entry(entry, head, ue_hash) {
+                if (upcall_compare(cache, entry, key, args) == 0) {
+                        found = 1;
+                        break;
+                }
+        }
+
+        if (found) {
+                CWARN("%s: flush entry %p: key "LPU64", ref %d, fl %x, "
+                      "cur %lu, ex %ld/%ld\n",
+                      cache->uc_name, entry, entry->ue_key,
+                      atomic_read(&entry->ue_refcount), entry->ue_flags,
+                      get_seconds(), entry->ue_acquire_expire,
+                      entry->ue_expire);
+                UC_CACHE_SET_EXPIRED(entry);
+                if (!atomic_read(&entry->ue_refcount))
+                        free_entry(cache, entry);
+        }
+        spin_unlock(&cache->uc_lock);
+}
+EXPORT_SYMBOL(upcall_cache_flush_one);
+
+struct upcall_cache *upcall_cache_init(const char *name, const char *upcall,
+                                       struct upcall_cache_ops *ops)
 {
-        struct upcall_cache *hash;
+        struct upcall_cache *cache;
         int i;
         ENTRY;
 
-        OBD_ALLOC(hash, sizeof(*hash));
-        if (!hash)
+        OBD_ALLOC(cache, sizeof(*cache));
+        if (!cache)
                 RETURN(ERR_PTR(-ENOMEM));
 
-        spin_lock_init(&hash->uc_lock);
+        spin_lock_init(&cache->uc_lock);
         for (i = 0; i < UC_CACHE_HASH_SIZE; i++)
-                INIT_LIST_HEAD(&hash->uc_hashtable[i]);
-        strncpy(hash->uc_name, name, sizeof(hash->uc_name) - 1);
-        /* set default value, proc tunable */
-        strcpy(hash->uc_upcall, "NONE");
-        hash->uc_entry_expire = 10 * 60 * HZ;
-        hash->uc_acquire_expire = 15 * HZ;
-
-        RETURN(hash);
+                INIT_LIST_HEAD(&cache->uc_hashtable[i]);
+        strncpy(cache->uc_name, name, sizeof(cache->uc_name) - 1);
+        /* upcall pathname proc tunable */
+        strncpy(cache->uc_upcall, upcall, sizeof(cache->uc_upcall) - 1);
+        cache->uc_entry_expire = 10 * 60 * HZ;
+        cache->uc_acquire_expire = 15 * HZ;
+        cache->uc_ops = ops;
+
+        RETURN(cache);
 }
 EXPORT_SYMBOL(upcall_cache_init);
 
-void upcall_cache_cleanup(struct upcall_cache *hash)
+void upcall_cache_cleanup(struct upcall_cache *cache)
 {
-        if (!hash)
+        if (!cache)
                 return;
-        upcall_cache_flush_all(hash);
-        OBD_FREE(hash, sizeof(*hash));
+        upcall_cache_flush_all(cache);
+        OBD_FREE(cache, sizeof(*cache));
 }
 EXPORT_SYMBOL(upcall_cache_cleanup);