Whamcloud - gitweb
b=5498
[fs/lustre-release.git] / lustre / llite / llite_capa.c
index 4bebb59..adf9560 100644 (file)
@@ -1,7 +1,7 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- * Copyright (C) 2004, 2005 Cluster File Systems, Inc.
+ * Copyright (C) 2005 Cluster File Systems, Inc.
  *
  * Author: Lai Siyao <lsy@clusterfs.com>
  *
 #include <linux/file.h>
 #include <linux/kmod.h>
 
-#include <linux/lustre_lite.h>
+#include <lustre_lite.h>
 #include "llite_internal.h"
 
+/* for obd_capa.c_list, client capa might stay in three places:
+ * 1. ll_capa_list.
+ * 2. ll_idle_capas.
+ * 3. stand alone: just allocated.
+ */
+
+/* capas for oss writeback and those failed to renew */
+static LIST_HEAD(ll_idle_capas);
 static struct ptlrpc_thread ll_capa_thread;
-static struct list_head *ll_capa_list = &capa_list[CLIENT_CAPA];
-static struct thread_ctl {
-        struct completion ctl_starting;
-        struct completion ctl_finishing;
-} ll_capa_ctl;
+static struct list_head *ll_capa_list = &capa_list[CAPA_SITE_CLIENT];
+
+/* llite capa renewal timer */
+struct timer_list ll_capa_timer;
+/* for debug: indicate whether capa on llite is enabled or not */
+static atomic_t ll_capa_debug = ATOMIC_INIT(0);
+static unsigned long long ll_capa_renewed = 0;
+static unsigned long long ll_capa_renewal_noent = 0;
+static unsigned long long ll_capa_renewal_failed = 0;
+static unsigned long long ll_capa_renewal_retries = 0;
+
+static inline void update_capa_timer(struct obd_capa *ocapa, cfs_time_t expiry)
+{
+        if (time_before(expiry, ll_capa_timer.expires) ||
+            !timer_pending(&ll_capa_timer)) {
+                mod_timer(&ll_capa_timer, expiry);
+                DEBUG_CAPA(D_SEC, &ocapa->c_capa,
+                           "ll_capa_timer update: %lu/%lu by", expiry, jiffies);
+        }
+}
+
+static inline cfs_time_t capa_renewal_time(struct obd_capa *ocapa)
+{
+        return cfs_time_sub(ocapa->c_expiry,
+                            cfs_time_seconds(ocapa->c_capa.lc_timeout) / 2);
+}
+
+static inline int capa_is_to_expire(struct obd_capa *ocapa)
+{
+        return cfs_time_beforeq(capa_renewal_time(ocapa), cfs_time_current());
+}
 
 static inline int have_expired_capa(void)
 {
-        struct obd_capa *ocapa;
+        struct obd_capa *ocapa = NULL;
         int expired = 0;
-        ENTRY;
 
+        /* if ll_capa_list has client capa to expire or ll_idle_capas has
+         * expired capa, return 1.
+         */
         spin_lock(&capa_lock);
         if (!list_empty(ll_capa_list)) {
                 ocapa = list_entry(ll_capa_list->next, struct obd_capa, c_list);
-
-                expired = __capa_is_to_expire(ocapa);
+                expired = capa_is_to_expire(ocapa);
+                if (!expired)
+                        update_capa_timer(ocapa, capa_renewal_time(ocapa));
+        } else if (!list_empty(&ll_idle_capas)) {
+                ocapa = list_entry(ll_idle_capas.next, struct obd_capa, c_list);
+                expired = capa_is_expired(ocapa);
+                if (!expired)
+                        update_capa_timer(ocapa, ocapa->c_expiry);
         }
         spin_unlock(&capa_lock);
 
-        RETURN(expired);
+        if (expired)
+                DEBUG_CAPA(D_SEC, &ocapa->c_capa, "expired");
+        return expired;
 }
 
-static int inline ll_capa_check_stop(void)
+static inline int ll_capa_check_stop(void)
 {
         return (ll_capa_thread.t_flags & SVC_STOPPING) ? 1: 0;
 }
 
-static int ll_renew_capa(struct obd_capa *ocapa)
+static void sort_add_capa(struct obd_capa *ocapa, struct list_head *head)
 {
-        struct ptlrpc_request *req = NULL;
-        /* no need to lock, no one else will touch it */
-        struct inode *inode = ocapa->c_inode;
-        struct obd_export *md_exp = ll_i2mdexp(inode);
-        struct ll_inode_info *lli = ll_i2info(inode);
-        __u64 valid = 0;
-        int rc;
-        ENTRY;
+        struct obd_capa *tmp;
+        struct list_head *before = NULL;
 
-        valid |= OBD_MD_CAPA;
+        /* TODO: client capa is sorted by expiry, this could be optimized */
+        list_for_each_entry_reverse(tmp, head, c_list) {
+                if (cfs_time_aftereq(ocapa->c_expiry, tmp->c_expiry)) {
+                        before = &tmp->c_list;
+                        break;
+                }
+        }
 
-        rc = md_getattr(md_exp, &lli->lli_id, valid, NULL, NULL, 0,
-                        0, ocapa, &req);
-        if (rc < 0)
-                CDEBUG(D_INFO, "md_getattr failed: rc = %d\n", rc);
-        RETURN(rc);
+        LASSERT(&ocapa->c_list != before);
+        list_add(&ocapa->c_list, before ?: head);
 }
 
-static int ll_capa_thread_main(void *arg)
+static inline int obd_capa_open_count(struct obd_capa *oc)
 {
-        struct thread_ctl *ctl = arg;
-        unsigned long flags;
-        ENTRY;
+        struct ll_inode_info *lli = ll_i2info(oc->u.cli.inode);
+        return atomic_read(&lli->lli_open_count);
+}
+
+static void ll_delete_capa(struct obd_capa *ocapa)
+{
+        struct ll_inode_info *lli = ll_i2info(ocapa->u.cli.inode);
 
-        {
-                char name[sizeof(current->comm)];
-                snprintf(name, sizeof(name) - 1, "ll_capa");
-                kportal_daemonize(name);
+        if (capa_for_mds(&ocapa->c_capa)) {
+                LASSERT(lli->lli_mds_capa == ocapa);
+                lli->lli_mds_capa = NULL;
+        } else if (capa_for_oss(&ocapa->c_capa)) {
+                list_del_init(&ocapa->u.cli.lli_list);
         }
 
-        SIGNAL_MASK_LOCK(current, flags);
-        sigfillset(&current->blocked);
-        RECALC_SIGPENDING;
-        SIGNAL_MASK_UNLOCK(current, flags);
+        DEBUG_CAPA(D_SEC, &ocapa->c_capa, "free client");
+        list_del(&ocapa->c_list);
+        capa_count[CAPA_SITE_CLIENT]--;
+        free_capa(ocapa);
+}
+
+/* three places where client capa is deleted:
+ * 1. capa_thread_main(), main place to delete expired capa.
+ * 2. ll_clear_inode_capas() in ll_clear_inode().
+ * 3. ll_truncate_free_capa() delete truncate capa explicitly in ll_truncate().
+ */
+static int capa_thread_main(void *unused)
+{
+        struct obd_capa *ocapa, *tmp, *next;
+        struct inode *inode = NULL;
+        struct l_wait_info lwi = { 0 };
+        int rc;
+        ENTRY;
+
+        cfs_daemonize("ll_capa");
 
-        /*
-         * letting starting function know, that we are ready and control may be
-         * returned.
-         */
         ll_capa_thread.t_flags = SVC_RUNNING;
-        complete(&ctl->ctl_starting);
+        wake_up(&ll_capa_thread.t_ctl_waitq);
 
         while (1) {
-                struct l_wait_info lwi = { 0 };
-                struct obd_capa *ocapa, *next = NULL;
-                int sleep = CAPA_PRE_EXPIRY;
-
                 l_wait_event(ll_capa_thread.t_ctl_waitq,
-                             (have_expired_capa() || ll_capa_check_stop()),
+                             (ll_capa_check_stop() || have_expired_capa()),
                              &lwi);
 
+                if (ll_capa_check_stop())
+                        break;
+
+                next = NULL;
+
                 spin_lock(&capa_lock);
-                list_for_each_entry(ocapa, ll_capa_list, c_list) {
-                        if (__capa_is_to_expire(ocapa)) {
-                                /* get capa in case it's deleted */
-                                __capa_get(ocapa);
-
-                                spin_unlock(&capa_lock);
-                                ll_renew_capa(ocapa);
-                                capa_put(ocapa, CLIENT_CAPA);
-                                spin_lock(&capa_lock);
-                        } else {
+                list_for_each_entry_safe(ocapa, tmp, ll_capa_list, c_list) {
+                        LASSERT(ocapa->c_capa.lc_opc != CAPA_OPC_OSS_TRUNC);
+
+                        if (!capa_is_to_expire(ocapa)) {
                                 next = ocapa;
                                 break;
                         }
+
+                        list_del_init(&ocapa->c_list);
+
+                        /* for MDS capability, only renew those which belong to
+                         * dir, or its inode is opened, or client holds LOOKUP
+                         * lock.
+                         */
+                        if (capa_for_mds(&ocapa->c_capa) &&
+                            !S_ISDIR(ocapa->u.cli.inode->i_mode) &&
+                            obd_capa_open_count(ocapa) == 0 &&
+                            !ll_have_md_lock(ocapa->u.cli.inode,
+                                             MDS_INODELOCK_LOOKUP)) {
+                                DEBUG_CAPA(D_SEC, &ocapa->c_capa,
+                                           "skip renewal for");
+                                sort_add_capa(ocapa, &ll_idle_capas);
+                                continue;
+                        }
+
+                        /* for OSS capability, only renew those whose inode is
+                         * opened.
+                         */
+                        if (capa_for_oss(&ocapa->c_capa) &&
+                            obd_capa_open_count(ocapa) == 0) {
+                                /* oss capa with open count == 0 won't renew,
+                                 * move to idle list */
+                                sort_add_capa(ocapa, &ll_idle_capas);
+                                continue;
+                        }
+
+                        /* NB iput() is in ll_update_capa() */
+                        inode = igrab(ocapa->u.cli.inode);
+                        if (inode == NULL) {
+                                DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
+                                           "igrab failed for");
+                                continue;
+                        }
+
+                        capa_get(ocapa);
+                        ll_capa_renewed++;
+                        spin_unlock(&capa_lock);
+
+                        rc = md_renew_capa(ll_i2mdexp(inode), ocapa,
+                                           ll_update_capa);
+                        spin_lock(&capa_lock);
+                        if (rc) {
+                                DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
+                                           "renew failed: %d", rc);
+                                ll_capa_renewal_failed++;
+                        }
                 }
-                if (next) {
-                        mod_timer(&ll_capa_timer,
-                                  expiry_to_jiffies(next->c_capa.lc_expiry));
-                        if (next->c_capa.lc_flags & CAPA_FL_NOROUND)
-                                sleep = CAPA_PRE_EXPIRY_NOROUND;
-                }
-                spin_unlock(&capa_lock);
 
-                if (ll_capa_check_stop())
-                        break;
+                if (next)
+                        update_capa_timer(next, capa_renewal_time(next));
+
+                list_for_each_entry_safe(ocapa, tmp, &ll_idle_capas, c_list) {
+                        if (!capa_is_expired(ocapa)) {
+                                if (!next)
+                                        update_capa_timer(ocapa, ocapa->c_expiry);
+                                break;
+                        }
+
+                        if (atomic_read(&ocapa->c_refc)) {
+                                DEBUG_CAPA(D_SEC, &ocapa->c_capa,
+                                           "expired(c_refc %d), don't release",
+                                           atomic_read(&ocapa->c_refc));
+                                /* don't try to renew any more */
+                                list_del_init(&ocapa->c_list);
+                                continue;
+                        }
 
-                /* wait ll_renew_capa finish */
-                set_current_state(TASK_INTERRUPTIBLE);
-                schedule_timeout(sleep * HZ);
+                        /* expired capa is released. */
+                        DEBUG_CAPA(D_SEC, &ocapa->c_capa, "release expired");
+                        ll_delete_capa(ocapa);
+                }
+
+                spin_unlock(&capa_lock);
         }
 
         ll_capa_thread.t_flags = SVC_STOPPED;
-
-        /* this is SMP-safe way to finish thread. */
-        complete_and_exit(&ctl->ctl_finishing, 0);
-        EXIT;
+        wake_up(&ll_capa_thread.t_ctl_waitq);
+        RETURN(0);
 }
 
-/* just wake up, others are handled by ll_capa_thread */
 void ll_capa_timer_callback(unsigned long unused)
 {
-        ENTRY;
         wake_up(&ll_capa_thread.t_ctl_waitq);
-        EXIT;
 }
 
-int ll_capa_start_thread(void)
+int ll_capa_thread_start(void)
 {
         int rc;
         ENTRY;
 
-        LASSERT(ll_capa_thread.t_flags == 0);
-        init_completion(&ll_capa_ctl.ctl_starting);
-        init_completion(&ll_capa_ctl.ctl_finishing);
         init_waitqueue_head(&ll_capa_thread.t_ctl_waitq);
 
-        rc = kernel_thread(ll_capa_thread_main, &ll_capa_ctl,
-                           (CLONE_VM | CLONE_FILES));
+        rc = kernel_thread(capa_thread_main, NULL, 0);
         if (rc < 0) {
-                CERROR("cannot start expired capa thread, "
-                       "err = %d\n", rc);
+                CERROR("cannot start expired capa thread: rc %d\n", rc);
                 RETURN(rc);
         }
-        wait_for_completion(&ll_capa_ctl.ctl_starting);
-        LASSERT(ll_capa_thread.t_flags == SVC_RUNNING);
+        wait_event(ll_capa_thread.t_ctl_waitq,
+                   ll_capa_thread.t_flags & SVC_RUNNING);
+
         RETURN(0);
 }
 
-void ll_capa_stop_thread(void)
+void ll_capa_thread_stop(void)
 {
-        ENTRY;
-
         ll_capa_thread.t_flags = SVC_STOPPING;
         wake_up(&ll_capa_thread.t_ctl_waitq);
-        wait_for_completion(&ll_capa_ctl.ctl_finishing);
-        LASSERT(ll_capa_thread.t_flags == SVC_STOPPED);
-        ll_capa_thread.t_flags = 0;
+        wait_event(ll_capa_thread.t_ctl_waitq,
+                   ll_capa_thread.t_flags & SVC_STOPPED);
+}
 
-        EXIT;
+static struct obd_capa *do_lookup_oss_capa(struct inode *inode, int opc)
+{
+        struct ll_inode_info *lli = ll_i2info(inode);
+        struct obd_capa *ocapa;
+
+        /* inside capa_lock */
+        list_for_each_entry(ocapa, &lli->lli_oss_capas, u.cli.lli_list) {
+                if ((capa_opc(&ocapa->c_capa) & opc) != opc)
+                        continue;
+
+                LASSERT(lu_fid_eq(capa_fid(&ocapa->c_capa),
+                                  ll_inode2fid(inode)));
+                LASSERT(ocapa->c_site == CAPA_SITE_CLIENT);
+
+                DEBUG_CAPA(D_SEC, &ocapa->c_capa, "found client");
+                return ocapa;
+        }
+
+        return NULL;
 }
 
-int ll_set_och_capa(struct inode *inode, struct lookup_intent *it,
-                   struct obd_client_handle *och)
+struct obd_capa *ll_osscapa_get(struct inode *inode, __u64 opc)
 {
-        struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
         struct ll_inode_info *lli = ll_i2info(inode);
-        struct mds_body *body;
-        struct lustre_capa *capa;
-        __u64 mdsid = lli->lli_id.li_fid.lf_group;
-        unsigned long ino = lli->lli_id.li_stc.u.e3s.l3s_ino;
-        int capa_op = (it->it_flags & MAY_WRITE) ? MAY_WRITE : MAY_READ;
-        unsigned long expiry;
+        struct obd_capa *ocapa;
+        int found = 0;
+
+        if ((ll_i2sbi(inode)->ll_flags & LL_SBI_OSS_CAPA) == 0)
+                return NULL;
+        ENTRY;
+
+        LASSERT(opc == CAPA_OPC_OSS_WRITE || opc == CAPA_OPC_OSS_RW ||
+                opc == CAPA_OPC_OSS_TRUNC);
+
+        spin_lock(&capa_lock);
+        list_for_each_entry(ocapa, &lli->lli_oss_capas, u.cli.lli_list) {
+                if (capa_is_expired(ocapa))
+                        continue;
+                if ((opc & CAPA_OPC_OSS_WRITE) &&
+                    capa_opc_supported(&ocapa->c_capa, CAPA_OPC_OSS_WRITE)) {
+                        found = 1; break;
+                } else if ((opc & CAPA_OPC_OSS_READ) &&
+                           capa_opc_supported(&ocapa->c_capa,
+                                              CAPA_OPC_OSS_READ)) {
+                        found = 1; break;
+                } else if ((opc & CAPA_OPC_OSS_TRUNC) &&
+                           capa_opc_supported(&ocapa->c_capa, opc)) {
+                        found = 1; break;
+                }
+        }
+
+        if (found) {
+                LASSERT(lu_fid_eq(capa_fid(&ocapa->c_capa),
+                                  ll_inode2fid(inode)));
+                LASSERT(ocapa->c_site == CAPA_SITE_CLIENT);
+
+                capa_get(ocapa);
+
+                DEBUG_CAPA(D_SEC, &ocapa->c_capa, "found client");
+        } else {
+                ocapa = NULL;
+
+                if (atomic_read(&ll_capa_debug)) {
+                        CERROR("no capability for "DFID" opc "LPX64"\n",
+                               PFID(&lli->lli_fid), opc);
+                        atomic_set(&ll_capa_debug, 0);
+                }
+        }
+        spin_unlock(&capa_lock);
+
+        RETURN(ocapa);
+}
+EXPORT_SYMBOL(ll_osscapa_get);
+
+struct obd_capa *ll_mdscapa_get(struct inode *inode)
+{
+        struct ll_inode_info *lli = ll_i2info(inode);
+        struct obd_capa *ocapa;
+        ENTRY;
+
+        LASSERT(inode != NULL);
+        
+        if ((ll_i2sbi(inode)->ll_flags & LL_SBI_MDS_CAPA) == 0)
+                RETURN(NULL);
+
+        spin_lock(&capa_lock);
+        ocapa = capa_get(lli->lli_mds_capa);
+        spin_unlock(&capa_lock);
+        if (!ocapa && atomic_read(&ll_capa_debug)) {
+                CERROR("no mds capability for "DFID"\n", PFID(&lli->lli_fid));
+                atomic_set(&ll_capa_debug, 0);
+        }
+
+        RETURN(ocapa);
+}
+
+static struct obd_capa *do_add_mds_capa(struct inode *inode,
+                                        struct obd_capa *ocapa)
+{
+        struct ll_inode_info *lli = ll_i2info(inode);
+        struct obd_capa *old = lli->lli_mds_capa;
+        struct lustre_capa *capa = &ocapa->c_capa;
+
+        if (!old) {
+                ocapa->u.cli.inode = inode;
+                lli->lli_mds_capa = ocapa;
+                capa_count[CAPA_SITE_CLIENT]++;
+
+                DEBUG_CAPA(D_SEC, capa, "add MDS");
+        } else {
+                spin_lock(&old->c_lock);
+                old->c_capa = *capa;
+                spin_unlock(&old->c_lock);
+
+                DEBUG_CAPA(D_SEC, capa, "update MDS");
+
+                free_capa(ocapa);
+                ocapa = old;
+        }
+        return ocapa;
+}
+
+static inline void inode_add_oss_capa(struct inode *inode,
+                                      struct obd_capa *ocapa)
+{
+        struct ll_inode_info *lli = ll_i2info(inode);
+        struct obd_capa *tmp;
+        struct list_head *next = NULL;
+
+        /* capa is sorted in lli_oss_capas so lookup can always find the
+         * latest one */
+        list_for_each_entry(tmp, &lli->lli_oss_capas, u.cli.lli_list) {
+                if (cfs_time_after(ocapa->c_expiry, tmp->c_expiry)) {
+                        next = &tmp->u.cli.lli_list;
+                        break;
+                }
+        }
+        LASSERT(&ocapa->u.cli.lli_list != next);
+        list_move_tail(&ocapa->u.cli.lli_list, next ?: &lli->lli_oss_capas);
+}
+
+static struct obd_capa *do_add_oss_capa(struct inode *inode,
+                                        struct obd_capa *ocapa)
+{
+        struct obd_capa *old;
+        struct lustre_capa *capa = &ocapa->c_capa;
+
+        LASSERTF(S_ISREG(inode->i_mode),
+                 "inode has oss capa, but not regular file, mode: %d\n",
+                 inode->i_mode);
+
+        /* FIXME: can't replace it so easily with fine-grained opc */
+        old = do_lookup_oss_capa(inode, capa_opc(capa) & CAPA_OPC_OSS_ONLY);
+        if (!old) {
+                ocapa->u.cli.inode = inode;
+                INIT_LIST_HEAD(&ocapa->u.cli.lli_list);
+                capa_count[CAPA_SITE_CLIENT]++;
+
+                DEBUG_CAPA(D_SEC, capa, "add OSS");
+        } else {
+                spin_lock(&old->c_lock);
+                old->c_capa = *capa;
+                spin_unlock(&old->c_lock);
+
+                DEBUG_CAPA(D_SEC, capa, "update OSS");
+
+                free_capa(ocapa);
+                ocapa = old;
+        }
+
+        inode_add_oss_capa(inode, ocapa);
+        return ocapa;
+}
+
+struct obd_capa *ll_add_capa(struct inode *inode, struct obd_capa *ocapa)
+{
+        spin_lock(&capa_lock);
+        ocapa = capa_for_mds(&ocapa->c_capa) ? do_add_mds_capa(inode, ocapa) :
+                                               do_add_oss_capa(inode, ocapa);
+
+        /* truncate capa won't renew */
+        if (ocapa->c_capa.lc_opc != CAPA_OPC_OSS_TRUNC) {
+                set_capa_expiry(ocapa);
+                list_del(&ocapa->c_list);
+                sort_add_capa(ocapa, ll_capa_list);
+
+                update_capa_timer(ocapa, capa_renewal_time(ocapa));
+        }
+
+        spin_unlock(&capa_lock);
+
+        atomic_set(&ll_capa_debug, 1);
+        return ocapa;
+}
+
+static inline void delay_capa_renew(struct obd_capa *oc, cfs_time_t delay)
+{
+        /* NB: set a fake expiry for this capa to prevent it renew too soon */
+        oc->c_expiry = cfs_time_add(oc->c_expiry, cfs_time_seconds(delay));
+}
+
+int ll_update_capa(struct obd_capa *ocapa, struct lustre_capa *capa)
+{
+        struct inode *inode = ocapa->u.cli.inode;
         int rc = 0;
         ENTRY;
 
-        body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
-        LASSERT(body != NULL);          /* reply already checked out */
-        LASSERT_REPSWABBED(req, 1);     /* and swabbed down */
+        LASSERT(ocapa);
+
+        if (IS_ERR(capa)) {
+                /* set error code */
+                rc = PTR_ERR(capa);
+                spin_lock(&capa_lock);
+                if (rc == -ENOENT) {
+                        DEBUG_CAPA(D_SEC, &ocapa->c_capa,
+                                   "renewal canceled because object removed");
+                        ll_capa_renewal_noent++;
+                } else {
+                        ll_capa_renewal_failed++;
+
+                        /* failed capa won't be renewed any longer, but if -EIO,
+                         * client might be doing recovery, retry in 2 min. */
+                        if (rc == -EIO && !capa_is_expired(ocapa)) {
+                                delay_capa_renew(ocapa, 120);
+                                DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
+                                           "renewal failed: -EIO, retry in 2 mins");
+                                ll_capa_renewal_retries++;
+                                GOTO(retry, rc);
+                        } else {
+                                DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
+                                           "renewal failed(rc: %d) for", rc);
+                        }
+                }
 
-        capa = lustre_msg_buf(req->rq_repmsg, 7, sizeof (*capa));
-        LASSERT(capa != NULL);          /* reply already checked out */
-        LASSERT_REPSWABBED(req, 7);     /* and swabbed down */
+                list_del(&ocapa->c_list);
+                sort_add_capa(ocapa, &ll_idle_capas);
+                spin_unlock(&capa_lock);
 
-        och->och_capa = capa_get(current->uid, capa_op, mdsid, ino,
-                                 CLIENT_CAPA, capa, inode, &body->handle);
-        if (!och->och_capa)
-                rc = -ENOMEM;
+                capa_put(ocapa);
+                iput(inode);
+                return rc;
+        }
 
-        expiry = expiry_to_jiffies(capa->lc_expiry - capa_pre_expiry(capa));
-        if (time_before(expiry, ll_capa_timer.expires) ||
-            !timer_pending(&ll_capa_timer))
-                mod_timer(&ll_capa_timer, expiry);
+        spin_lock(&ocapa->c_lock);
+        LASSERT(!memcmp(&ocapa->c_capa, capa,
+                        offsetof(struct lustre_capa, lc_flags)));
+        ocapa->c_capa = *capa;
+        set_capa_expiry(ocapa);
+        spin_unlock(&ocapa->c_lock);
 
-        RETURN(rc);
+        spin_lock(&capa_lock);
+        if (capa_for_oss(capa))
+                inode_add_oss_capa(inode, ocapa);
+        DEBUG_CAPA(D_SEC, capa, "renew");
+        EXIT;
+retry:
+        list_del_init(&ocapa->c_list);
+        sort_add_capa(ocapa, ll_capa_list);
+        update_capa_timer(ocapa, capa_renewal_time(ocapa));
+        spin_unlock(&capa_lock);
+
+        capa_put(ocapa);
+        iput(inode);
+        return rc;
+}
+
+void ll_capa_open(struct inode *inode)
+{
+        struct ll_inode_info *lli = ll_i2info(inode);
+
+        if ((ll_i2sbi(inode)->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
+            == 0)
+                return;
+
+        if (!S_ISREG(inode->i_mode))
+                return;
+
+        atomic_inc(&lli->lli_open_count);
+}
+
+void ll_capa_close(struct inode *inode)
+{
+        struct ll_inode_info *lli = ll_i2info(inode);
+
+        if ((ll_i2sbi(inode)->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
+            == 0)
+                return;
+
+        if (!S_ISREG(inode->i_mode))
+                return;
+
+        atomic_dec(&lli->lli_open_count);
+}
+
+/* delete CAPA_OPC_OSS_TRUNC only */
+void ll_truncate_free_capa(struct obd_capa *ocapa)
+{
+        if (!ocapa)
+                return;
+
+        LASSERT(ocapa->c_capa.lc_opc & CAPA_OPC_OSS_TRUNC);
+        DEBUG_CAPA(D_SEC, &ocapa->c_capa, "free truncate");
+
+        capa_put(ocapa);
+        spin_lock(&capa_lock);
+        ll_delete_capa(ocapa);
+        spin_unlock(&capa_lock);
+}
+
+void ll_clear_inode_capas(struct inode *inode)
+{
+        struct ll_inode_info *lli = ll_i2info(inode);
+        struct obd_capa *ocapa, *tmp;
+
+        spin_lock(&capa_lock);
+        ocapa = lli->lli_mds_capa;
+        if (ocapa)
+                ll_delete_capa(ocapa);
+                
+        list_for_each_entry_safe(ocapa, tmp, &lli->lli_oss_capas,
+                                 u.cli.lli_list)
+                ll_delete_capa(ocapa);
+        spin_unlock(&capa_lock);
+}
+
+void ll_print_capa_stat(struct ll_sb_info *sbi)
+{
+        if (sbi->ll_flags & (LL_SBI_MDS_CAPA | LL_SBI_OSS_CAPA))
+                LCONSOLE_INFO("Fid capabilities renewed: %llu\n"
+                              "Fid capabilities renewal ENOENT: %llu\n"
+                              "Fid capabilities failed to renew: %llu\n"
+                              "Fid capabilities renewal retries: %llu\n",
+                              ll_capa_renewed, ll_capa_renewal_noent,
+                              ll_capa_renewal_failed, ll_capa_renewal_retries);
 }