* vim:expandtab:shiftwidth=8:tabstop=8:
*
* lustre/obdclass/capa.c
- * Lustre Capability Cache Management
+ * Lustre Capability Hash Management
*
- * Copyright (c) 2001-2003 Cluster File Systems, Inc.
+ * Copyright (c) 2005 Cluster File Systems, Inc.
* Author: Lai Siyao<lsy@clusterfs.com>
*
* This file is part of Lustre, http://www.lustre.org.
#define DEBUG_SUBSYSTEM S_SEC
+#ifdef __KERNEL__
#include <linux/version.h>
#include <linux/fs.h>
#include <asm/unistd.h>
#include <linux/module.h>
#include <linux/init.h>
-#include <linux/obd_class.h>
-#include <linux/lustre_debug.h>
-#include <linux/lustre_idl.h>
-#include <linux/lustre_sec.h>
+#include <obd_class.h>
+#include <lustre_debug.h>
+#include <lustre/lustre_idl.h>
+#else
+#include <liblustre.h>
+#endif
+
#include <libcfs/list.h>
+#include <lustre_capa.h>
+
+#define NR_CAPAHASH 32
+#define CAPA_HASH_SIZE 3000 /* for MDS & OSS */
-kmem_cache_t *capa_cachep = NULL;
+cfs_mem_cache_t *capa_cachep = NULL;
-/* capa_lock protect capa hash, list and content. */
+#ifdef __KERNEL__
+/* lock for capa hash/capa_list/fo_capa_keys */
spinlock_t capa_lock = SPIN_LOCK_UNLOCKED;
-struct hlist_head *capa_hash;
-struct list_head capa_list[3];
-static int capa_count[3] = { 0 };
-/* TODO: mdc and llite all need this, so define it here.
- * in the future it will be moved to ll_sb_info to support multi-
- * mount point */
-struct timer_list ll_capa_timer;
+struct list_head capa_list[CAPA_SITE_MAX];
-EXPORT_SYMBOL(capa_lock);
-EXPORT_SYMBOL(capa_hash);
+static struct capa_hmac_alg capa_hmac_algs[] = {
+ DEF_CAPA_HMAC_ALG("sha1", SHA1, 20, 20),
+};
+#endif
+/* capa count */
+int capa_count[CAPA_SITE_MAX] = { 0, };
+
+EXPORT_SYMBOL(capa_cachep);
EXPORT_SYMBOL(capa_list);
-EXPORT_SYMBOL(ll_capa_timer);
+EXPORT_SYMBOL(capa_lock);
+EXPORT_SYMBOL(capa_count);
-static inline int const
-capa_hashfn(unsigned int uid, int capa_op, __u64 mdsid, unsigned long ino)
+struct hlist_head *init_capa_hash(void)
{
- return (ino ^ uid) * (unsigned long)capa_op * (unsigned long)mdsid %
- NR_CAPAHASH;
-}
+ struct hlist_head *hash;
+ int nr_hash, i;
-int capa_op(int flags)
-{
- if (flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
- return MAY_WRITE;
- else if (flags & FMODE_READ)
- return MAY_READ;
+ OBD_ALLOC(hash, PAGE_SIZE);
+ if (!hash)
+ return NULL;
- LBUG(); /* should be either MAY_READ or MAY_WRITE */
- return 0;
+ nr_hash = PAGE_SIZE / sizeof(struct hlist_head);
+ LASSERT(nr_hash > NR_CAPAHASH);
+
+ for (i = 0; i < NR_CAPAHASH; i++)
+ INIT_HLIST_HEAD(hash + i);
+ return hash;
}
-static struct obd_capa *
-find_capa(struct hlist_head *head, uid_t uid, int capa_op, __u64 mdsid,
- unsigned long ino, int type)
+#ifdef __KERNEL__
+static inline int capa_on_server(struct obd_capa *ocapa)
{
- struct hlist_node *pos;
- struct obd_capa *ocapa;
- ENTRY;
-
- CDEBUG(D_CACHE, "find_capa uid %u op %u mdsid "LPU64" ino %lu "
- "type %d\n", uid, capa_op, mdsid, ino, type);
-
- hlist_for_each_entry(ocapa, pos, head, c_hash) {
- if (ocapa->c_capa.lc_uid != uid)
- continue;
- if (ocapa->c_capa.lc_op != capa_op)
- continue;
- if (ocapa->c_capa.lc_mdsid != mdsid)
- continue;
- if (ocapa->c_capa.lc_ino != ino)
- continue;
- if (ocapa->c_type != type)
- continue;
- RETURN(ocapa);
- }
-
- RETURN(NULL);
+ return ocapa->c_site == CAPA_SITE_SERVER;
}
-inline void __capa_get(struct obd_capa *ocapa)
+static inline void capa_delete(struct obd_capa *ocapa)
{
- atomic_inc(&ocapa->c_refc);
+ LASSERT(capa_on_server(ocapa));
+ hlist_del(&ocapa->u.tgt.c_hash);
+ list_del(&ocapa->c_list);
+ capa_count[ocapa->c_site]--;
+ free_capa(ocapa);
}
-static struct obd_capa *
-find_capa_locked(struct hlist_head *head, uid_t uid, int capa_op, __u64 mdsid,
- unsigned long ino, int type)
+void cleanup_capa_hash(struct hlist_head *hash)
{
- struct obd_capa *ocapa;
- ENTRY;
+ int i;
+ struct hlist_node *pos, *next;
+ struct obd_capa *oc;
spin_lock(&capa_lock);
- ocapa = find_capa(head, uid, capa_op, mdsid, ino, type);
- if (ocapa)
- __capa_get(ocapa);
+ for (i = 0; i < NR_CAPAHASH; i++) {
+ hlist_for_each_entry_safe(oc, pos, next, hash + i, u.tgt.c_hash)
+ capa_delete(oc);
+ }
spin_unlock(&capa_lock);
- RETURN(ocapa);
+ OBD_FREE(hash, PAGE_SIZE);
}
-static struct obd_capa *alloc_capa(void)
+static inline int const capa_hashfn(struct lu_fid *fid)
{
- struct obd_capa *ocapa;
- ENTRY;
-
- OBD_SLAB_ALLOC(ocapa, capa_cachep, SLAB_NOFS, sizeof(*ocapa));
- if (ocapa) {
- INIT_HLIST_NODE(&ocapa->c_hash);
- INIT_LIST_HEAD(&ocapa->c_list);
- }
-
- RETURN(ocapa);
+ return (fid_oid(fid) ^ fid_ver(fid)) *
+ (unsigned long)(fid_seq(fid) + 1) % NR_CAPAHASH;
}
-static void destroy_capa(struct obd_capa *ocapa)
+/* capa renewal time check is earlier than that on client, which is to prevent
+ * client renew right after obtaining it. */
+static inline int capa_is_to_expire(struct obd_capa *oc)
{
- OBD_SLAB_FREE(ocapa, capa_cachep, sizeof(*ocapa));
+ return cfs_time_before(cfs_time_sub(oc->c_expiry,
+ cfs_time_seconds(oc->c_capa.lc_timeout)*2/3),
+ cfs_time_current());
}
-int capa_cache_init(void)
+static struct obd_capa *find_capa(struct lustre_capa *capa,
+ struct hlist_head *head, int alive)
{
- int order = 0, nr_hash, i;
-
- capa_hash = (struct hlist_head *)
- __get_free_pages(GFP_ATOMIC, order);
- if (!capa_hash)
- panic("Cannot create capa_hash hash table");
+ struct hlist_node *pos;
+ struct obd_capa *ocapa;
+ int len = alive ? offsetof(struct lustre_capa, lc_keyid):sizeof(*capa);
- nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
- LASSERT(nr_hash > NR_CAPAHASH);
+ hlist_for_each_entry(ocapa, pos, head, u.tgt.c_hash) {
+ if (memcmp(&ocapa->c_capa, capa, len))
+ continue;
+ /* don't return one that will expire soon in this case */
+ if (alive && capa_is_to_expire(ocapa))
+ continue;
- for (i = 0; i < NR_CAPAHASH; i++)
- INIT_HLIST_HEAD(capa_hash + i);
+ LASSERT(capa_on_server(ocapa));
- for (i =0; i < 3; i++)
- INIT_LIST_HEAD(&capa_list[i]);
+ DEBUG_CAPA(D_SEC, &ocapa->c_capa, "found");
+ return ocapa;
+ }
- return 0;
+ return NULL;
}
-void capa_cache_cleanup(void)
+#define LRU_CAPA_DELETE_COUNT 12
+static inline void capa_delete_lru(struct list_head *head)
{
struct obd_capa *ocapa;
- struct hlist_node *pos, *n;
-
- hlist_for_each_entry_safe(ocapa, pos, n, capa_hash, c_hash) {
- hlist_del(&ocapa->c_hash);
- list_del(&ocapa->c_list);
- OBD_FREE(ocapa, sizeof(*ocapa));
- }
-}
-
+ struct list_head *node = head->next;
+ int count = 0;
+
+ /* free LRU_CAPA_DELETE_COUNT unused capa from head */
+ while (count++ < LRU_CAPA_DELETE_COUNT) {
+ ocapa = list_entry(node, struct obd_capa, c_list);
+ node = node->next;
+ if (atomic_read(&ocapa->c_refc))
+ continue;
-static inline void list_add_capa(struct obd_capa *ocapa, struct list_head *head)
-{
- struct obd_capa *tmp;
-
- /* XXX: capa is sorted in client, this could be optimized */
- if (ocapa->c_type == CLIENT_CAPA) {
- list_for_each_entry_reverse(tmp, head, c_list) {
- if (ocapa->c_capa.lc_expiry > tmp->c_capa.lc_expiry) {
- list_add(&ocapa->c_list, &tmp->c_list);
- return;
- }
- }
+ DEBUG_CAPA(D_SEC, &ocapa->c_capa, "free lru");
+ capa_delete(ocapa);
}
-
- list_add_tail(&ocapa->c_list, head);
}
-#define DEBUG_CAPA(level, ocapa, fmt, args...) \
-do { \
-CDEBUG(level, fmt " capa@%p uid %u op %u ino "LPU64" mdsid %d keyid %d " \
- "expiry "LPU64" flags %u type %d\n", \
- ##args, ocapa, ocapa->c_capa.lc_uid, ocapa->c_capa.lc_op, \
- ocapa->c_capa.lc_ino, ocapa->c_capa.lc_mdsid, ocapa->c_capa.lc_keyid, \
- ocapa->c_capa.lc_expiry, ocapa->c_capa.lc_flags, ocapa->c_type); \
-} while (0)
-
-static struct obd_capa *
-get_new_capa_locked(struct hlist_head *head, uid_t uid, int capa_op,__u64 mdsid,
- unsigned long ino, int type, struct lustre_capa *capa,
- struct inode *inode, struct lustre_handle *handle)
+/* add or update */
+struct obd_capa *capa_add(struct hlist_head *hash, struct lustre_capa *capa)
{
- struct obd_capa *ocapa, *old;
- ENTRY;
+ struct hlist_head *head = hash + capa_hashfn(&capa->lc_fid);
+ struct obd_capa *ocapa, *old = NULL;
+ struct list_head *list = &capa_list[CAPA_SITE_SERVER];
- ocapa = alloc_capa();
+ ocapa = alloc_capa(CAPA_SITE_SERVER);
if (!ocapa)
- RETURN(NULL);
+ return NULL;
spin_lock(&capa_lock);
- old = find_capa(head, uid, capa_op, mdsid, ino, type);
+ old = find_capa(capa, head, 0);
if (!old) {
- memcpy(&ocapa->c_capa, capa, sizeof(*capa));
- ocapa->c_type = type;
- if (type == CLIENT_CAPA) {
- LASSERT(inode);
- igrab(inode);
- ocapa->c_inode = inode;
- memcpy(&ocapa->c_handle, handle, sizeof(*handle));
- }
- list_add_capa(ocapa, &capa_list[type]);
- hlist_add_head(&ocapa->c_hash, capa_hash);
- capa_count[type]++;
- DEBUG_CAPA(D_CACHE, ocapa, "get_new_capa_locked");
- __capa_get(ocapa);
- if (type != CLIENT_CAPA && capa_count[type] > CAPA_CACHE_SIZE) {
- struct list_head *node = capa_list[type].next;
- struct obd_capa *tcapa;
- int count = 0;
-
- /* free 12 unused capa from head */
- while (node->next != &capa_list[type] && count < 12) {
- tcapa = list_entry(node, struct obd_capa, c_list);
- node = node->next;
- if (atomic_read(&tcapa->c_refc) > 0)
- continue;
- list_del(&tcapa->c_list);
- destroy_capa(tcapa);
- count++;
- }
- }
+ ocapa->c_capa = *capa;
+ set_capa_expiry(ocapa);
+ hlist_add_head(&ocapa->u.tgt.c_hash, head);
+ list_add_tail(&ocapa->c_list, list);
+ capa_count[CAPA_SITE_SERVER]++;
+ capa_get(ocapa);
+
+ if (capa_count[CAPA_SITE_SERVER] > CAPA_HASH_SIZE)
+ capa_delete_lru(list);
+
+ DEBUG_CAPA(D_SEC, &ocapa->c_capa, "new");
spin_unlock(&capa_lock);
- RETURN(ocapa);
+ return ocapa;
}
- __capa_get(old);
+ capa_get(old);
spin_unlock(&capa_lock);
- destroy_capa(ocapa);
- ocapa = old;
- RETURN(ocapa);
-}
-
-static struct obd_capa *
-capa_get_locked(uid_t uid, int capa_op,__u64 mdsid, unsigned long ino,
- int type, struct lustre_capa *capa, struct inode *inode,
- struct lustre_handle *handle)
-{
- struct hlist_head *head = capa_hash +
- capa_hashfn(uid, capa_op, mdsid, ino);
- struct obd_capa *ocapa;
- ENTRY;
-
- ocapa = find_capa_locked(head, uid, capa_op, mdsid, ino, type);
- if (ocapa)
- RETURN(ocapa);
-
- if (capa)
- ocapa = get_new_capa_locked(head, uid, capa_op, mdsid, ino,
- type, capa, inode, handle);
- RETURN(ocapa);
-}
-
-struct obd_capa *
-capa_get(uid_t uid, int capa_op, __u64 mdsid, unsigned long ino, int type,
- struct lustre_capa *capa, struct inode *inode,
- struct lustre_handle *handle)
-{
- return capa_get_locked(uid, capa_op, mdsid, ino, type, capa, inode,
- handle);
-}
-
-static void __capa_put(struct obd_capa *ocapa, int type)
-{
- hlist_del_init(&ocapa->c_hash);
- list_del_init(&ocapa->c_list);
- capa_count[type]--;
-}
-
-void capa_put(struct obd_capa *ocapa, int type)
-{
- ENTRY;
-
- if (ocapa) {
- if (atomic_dec_and_lock(&ocapa->c_refc, &capa_lock)) {
- if (type == CLIENT_CAPA) {
- iput(ocapa->c_inode);
- __capa_put(ocapa, type);
- destroy_capa(ocapa);
- }
- spin_unlock(&capa_lock);
- }
- }
- EXIT;
-}
+ DEBUG_CAPA(D_SEC, &old->c_capa, "update");
-static inline void __update_capa(struct obd_capa *ocapa, struct lustre_capa *capa)
-{
- memcpy(&ocapa->c_capa, capa, sizeof(*capa));
+ free_capa(ocapa);
+ return old;
}
-static int update_capa_locked(struct lustre_capa *capa, int type)
+struct obd_capa *capa_lookup(struct hlist_head *hash, struct lustre_capa *capa,
+ int alive)
{
- uid_t uid = capa->lc_uid;
- int capa_op = capa->lc_op;
- __u64 mdsid = capa->lc_mdsid;
- unsigned long ino = capa->lc_ino;
- struct hlist_head *head = capa_hash +
- capa_hashfn(uid, capa_op, mdsid, ino);
struct obd_capa *ocapa;
- ENTRY;
spin_lock(&capa_lock);
- ocapa = find_capa(head, uid, capa_op, mdsid, ino, type);
- if (ocapa)
- __update_capa(ocapa, capa);
+ ocapa = find_capa(capa, hash + capa_hashfn(&capa->lc_fid), alive);
+ if (ocapa) {
+ list_move_tail(&ocapa->c_list, &capa_list[CAPA_SITE_SERVER]);
+ capa_get(ocapa);
+ }
spin_unlock(&capa_lock);
- if (ocapa == NULL && type == MDS_CAPA)
- ocapa = get_new_capa_locked(head, uid, capa_op, mdsid, ino, type,
- capa, NULL, NULL);
-
- RETURN(ocapa ? 0 : -ENOENT);
+ return ocapa;
}
-int capa_renew(struct lustre_capa *capa, int type)
+int capa_hmac(__u8 *hmac, struct lustre_capa *capa, __u8 *key)
{
- return update_capa_locked(capa, type);
-}
-
-void capa_hmac(struct crypto_tfm *tfm, u8 *key, struct lustre_capa *capa)
-{
- int keylen = CAPA_KEY_LEN;
+ struct crypto_tfm *tfm;
+ struct capa_hmac_alg *alg;
+ int keylen;
struct scatterlist sl = {
.page = virt_to_page(capa),
.offset = (unsigned long)(capa) % PAGE_SIZE,
- .length = sizeof(struct lustre_capa_data),
+ .length = offsetof(struct lustre_capa, lc_hmac),
};
- ENTRY;
- LASSERT(tfm);
- crypto_hmac(tfm, key, &keylen, &sl, 1, capa->lc_hmac);
- EXIT;
-}
+ if (capa_alg(capa) != CAPA_HMAC_ALG_SHA1) {
+ CERROR("unknown capability hmac algorithm!\n");
+ return -EFAULT;
+ }
-void capa_dup(void *dst, struct obd_capa *ocapa)
-{
- spin_lock(&capa_lock);
- memcpy(dst, &ocapa->c_capa, sizeof(ocapa->c_capa));
- spin_unlock(&capa_lock);
-}
+ alg = &capa_hmac_algs[capa_alg(capa)];
-void capa_dup2(void *dst, struct lustre_capa *capa)
-{
- spin_lock(&capa_lock);
- memcpy(dst, capa, sizeof(*capa));
- spin_unlock(&capa_lock);
-}
+ tfm = crypto_alloc_tfm(alg->ha_name, 0);
+ if (!tfm) {
+ CERROR("crypto_alloc_tfm failed, check whether your kernel"
+ "has crypto support!\n");
+ return -ENOMEM;
+ }
+ keylen = alg->ha_keylen;
-int capa_expired(struct lustre_capa *capa)
-{
- struct timeval tv;
+ crypto_hmac(tfm, key, &keylen, &sl, 1, hmac);
+ crypto_free_tfm(tfm);
- do_gettimeofday(&tv);
- return (capa->lc_expiry < tv.tv_sec) ? 1 : 0;
+ return 0;
}
+#endif
-int __capa_is_to_expire(struct obd_capa *ocapa)
+void capa_cpy(void *capa, struct obd_capa *ocapa)
{
- struct timeval tv;
- int pre_expiry = capa_pre_expiry(&ocapa->c_capa);
-
- do_gettimeofday(&tv);
- return (ocapa->c_capa.lc_expiry - pre_expiry < tv.tv_sec)? 1 : 0;
+ spin_lock(&ocapa->c_lock);
+ *(struct lustre_capa *)capa = ocapa->c_capa;
+ spin_unlock(&ocapa->c_lock);
}
-int capa_is_to_expire(struct obd_capa *ocapa)
+char *dump_capa_content(char *buf, char *key, int len)
{
- int rc;
+ int i, n = 0;
- spin_lock(&capa_lock);
- rc = __capa_is_to_expire(ocapa);
- spin_unlock(&capa_lock);
-
- return rc;
+ for (i = 0; i < len; i++)
+ n += sprintf(buf + n, "%02x", (unsigned char) key[i]);
+ return buf;
}
-EXPORT_SYMBOL(capa_op);
-EXPORT_SYMBOL(capa_get);
-EXPORT_SYMBOL(capa_put);
-EXPORT_SYMBOL(capa_renew);
-EXPORT_SYMBOL(__capa_get);
+EXPORT_SYMBOL(init_capa_hash);
+EXPORT_SYMBOL(cleanup_capa_hash);
+
+EXPORT_SYMBOL(capa_add);
+EXPORT_SYMBOL(capa_lookup);
+
EXPORT_SYMBOL(capa_hmac);
-EXPORT_SYMBOL(capa_dup);
-EXPORT_SYMBOL(capa_dup2);
-EXPORT_SYMBOL(capa_expired);
-EXPORT_SYMBOL(__capa_is_to_expire);
-EXPORT_SYMBOL(capa_is_to_expire);
+EXPORT_SYMBOL(capa_cpy);
+
+EXPORT_SYMBOL(dump_capa_content);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff