Whamcloud - gitweb
LU-2775 osp: enable fid-on-OST only for DNE.
[fs/lustre-release.git] / lustre / llite / statahead.c
index 537b44d..9eea49e 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -29,8 +27,7 @@
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011 Whamcloud, Inc.
- *
+ * Copyright (c) 2011, 2012, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -40,7 +37,6 @@
 #include <linux/fs.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
-#include <linux/smp_lock.h>
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
 
@@ -49,7 +45,6 @@
 #include <obd_support.h>
 #include <lustre_lite.h>
 #include <lustre_dlm.h>
-#include <linux/lustre_version.h>
 #include "llite_internal.h"
 
 #define SA_OMITTED_ENTRY_MAX 8ULL
@@ -88,7 +83,7 @@ struct ll_sa_entry {
 };
 
 static unsigned int sai_generation = 0;
-static cfs_spinlock_t sai_generation_lock = CFS_SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(sai_generation_lock);
 
 static inline int ll_sa_entry_unlinked(struct ll_sa_entry *entry)
 {
@@ -120,11 +115,11 @@ static inline int ll_sa_entry_hash(int val)
 static inline void
 ll_sa_entry_enhash(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
 {
-        int i = ll_sa_entry_hash(entry->se_qstr.hash);
+       int i = ll_sa_entry_hash(entry->se_qstr.hash);
 
-        cfs_spin_lock(&sai->sai_cache_lock[i]);
-        cfs_list_add_tail(&entry->se_hash, &sai->sai_cache[i]);
-        cfs_spin_unlock(&sai->sai_cache_lock[i]);
+       spin_lock(&sai->sai_cache_lock[i]);
+       cfs_list_add_tail(&entry->se_hash, &sai->sai_cache[i]);
+       spin_unlock(&sai->sai_cache_lock[i]);
 }
 
 /*
@@ -133,20 +128,17 @@ ll_sa_entry_enhash(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
 static inline void
 ll_sa_entry_unhash(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
 {
-        int i = ll_sa_entry_hash(entry->se_qstr.hash);
+       int i = ll_sa_entry_hash(entry->se_qstr.hash);
 
-        cfs_spin_lock(&sai->sai_cache_lock[i]);
-        cfs_list_del_init(&entry->se_hash);
-        cfs_spin_unlock(&sai->sai_cache_lock[i]);
+       spin_lock(&sai->sai_cache_lock[i]);
+       cfs_list_del_init(&entry->se_hash);
+       spin_unlock(&sai->sai_cache_lock[i]);
 }
 
 static inline int agl_should_run(struct ll_statahead_info *sai,
                                  struct inode *inode)
 {
-        if (inode != NULL && S_ISREG(inode->i_mode) &&
-            ll_i2info(inode)->lli_smd != NULL && sai->sai_agl_valid)
-                return 1;
-        return 0;
+       return (inode != NULL && S_ISREG(inode->i_mode) && sai->sai_agl_valid);
 }
 
 static inline struct ll_sa_entry *
@@ -257,14 +249,14 @@ ll_sa_entry_alloc(struct ll_statahead_info *sai, __u64 index,
         entry->se_qstr.name = dname;
 
         lli = ll_i2info(sai->sai_inode);
-        cfs_spin_lock(&lli->lli_sa_lock);
-        cfs_list_add_tail(&entry->se_list, &sai->sai_entries_sent);
-        cfs_spin_unlock(&lli->lli_sa_lock);
+       spin_lock(&lli->lli_sa_lock);
+       cfs_list_add_tail(&entry->se_list, &sai->sai_entries_sent);
+       spin_unlock(&lli->lli_sa_lock);
 
-        cfs_atomic_inc(&sai->sai_cache_count);
-        ll_sa_entry_enhash(sai, entry);
+       cfs_atomic_inc(&sai->sai_cache_count);
+       ll_sa_entry_enhash(sai, entry);
 
-        RETURN(entry);
+       RETURN(entry);
 }
 
 /*
@@ -354,17 +346,17 @@ static void ll_sa_entry_put(struct ll_statahead_info *sai,
 static inline void
 do_sai_entry_fini(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
 {
-        struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
+       struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
 
-        ll_sa_entry_unhash(sai, entry);
+       ll_sa_entry_unhash(sai, entry);
 
-        cfs_spin_lock(&lli->lli_sa_lock);
-        entry->se_stat = SA_ENTRY_DEST;
-        if (likely(!ll_sa_entry_unlinked(entry)))
-                cfs_list_del_init(&entry->se_list);
-        cfs_spin_unlock(&lli->lli_sa_lock);
+       spin_lock(&lli->lli_sa_lock);
+       entry->se_stat = SA_ENTRY_DEST;
+       if (likely(!ll_sa_entry_unlinked(entry)))
+               cfs_list_del_init(&entry->se_list);
+       spin_unlock(&lli->lli_sa_lock);
 
-        ll_sa_entry_put(sai, entry);
+       ll_sa_entry_put(sai, entry);
 }
 
 /*
@@ -428,21 +420,21 @@ do_sai_entry_to_stated(struct ll_statahead_info *sai,
  */
 static int
 ll_sa_entry_to_stated(struct ll_statahead_info *sai,
-                       struct ll_sa_entry *entry, int rc)
+                     struct ll_sa_entry *entry, int rc)
 {
-        struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
-        int                   ret = 1;
+       struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
+       int                   ret = 1;
 
-        ll_sa_entry_cleanup(sai, entry);
+       ll_sa_entry_cleanup(sai, entry);
 
-        cfs_spin_lock(&lli->lli_sa_lock);
-        if (likely(entry->se_stat != SA_ENTRY_DEST)) {
-                do_sai_entry_to_stated(sai, entry, rc);
-                ret = 0;
-        }
-        cfs_spin_unlock(&lli->lli_sa_lock);
+       spin_lock(&lli->lli_sa_lock);
+       if (likely(entry->se_stat != SA_ENTRY_DEST)) {
+               do_sai_entry_to_stated(sai, entry, rc);
+               ret = 0;
+       }
+       spin_unlock(&lli->lli_sa_lock);
 
-        return ret;
+       return ret;
 }
 
 /*
@@ -451,29 +443,29 @@ ll_sa_entry_to_stated(struct ll_statahead_info *sai,
 static void ll_agl_add(struct ll_statahead_info *sai,
                        struct inode *inode, int index)
 {
-        struct ll_inode_info *child  = ll_i2info(inode);
-        struct ll_inode_info *parent = ll_i2info(sai->sai_inode);
-        int                   added  = 0;
-
-        cfs_spin_lock(&child->lli_agl_lock);
-        if (child->lli_agl_index == 0) {
-                child->lli_agl_index = index;
-                cfs_spin_unlock(&child->lli_agl_lock);
-
-                LASSERT(cfs_list_empty(&child->lli_agl_list));
-
-                igrab(inode);
-                cfs_spin_lock(&parent->lli_agl_lock);
-                if (agl_list_empty(sai))
-                        added = 1;
-                cfs_list_add_tail(&child->lli_agl_list, &sai->sai_entries_agl);
-                cfs_spin_unlock(&parent->lli_agl_lock);
-        } else {
-                cfs_spin_unlock(&child->lli_agl_lock);
-        }
-
-        if (added > 0)
-                cfs_waitq_signal(&sai->sai_agl_thread.t_ctl_waitq);
+       struct ll_inode_info *child  = ll_i2info(inode);
+       struct ll_inode_info *parent = ll_i2info(sai->sai_inode);
+       int                   added  = 0;
+
+       spin_lock(&child->lli_agl_lock);
+       if (child->lli_agl_index == 0) {
+               child->lli_agl_index = index;
+               spin_unlock(&child->lli_agl_lock);
+
+               LASSERT(cfs_list_empty(&child->lli_agl_list));
+
+               igrab(inode);
+               spin_lock(&parent->lli_agl_lock);
+               if (agl_list_empty(sai))
+                       added = 1;
+               cfs_list_add_tail(&child->lli_agl_list, &sai->sai_entries_agl);
+               spin_unlock(&parent->lli_agl_lock);
+       } else {
+               spin_unlock(&child->lli_agl_lock);
+       }
+
+       if (added > 0)
+               cfs_waitq_signal(&sai->sai_agl_thread.t_ctl_waitq);
 }
 
 static struct ll_statahead_info *ll_sai_alloc(void)
@@ -488,11 +480,11 @@ static struct ll_statahead_info *ll_sai_alloc(void)
 
         cfs_atomic_set(&sai->sai_refcount, 1);
 
-        cfs_spin_lock(&sai_generation_lock);
-        sai->sai_generation = ++sai_generation;
-        if (unlikely(sai_generation == 0))
-                sai->sai_generation = ++sai_generation;
-        cfs_spin_unlock(&sai_generation_lock);
+       spin_lock(&sai_generation_lock);
+       sai->sai_generation = ++sai_generation;
+       if (unlikely(sai_generation == 0))
+               sai->sai_generation = ++sai_generation;
+       spin_unlock(&sai_generation_lock);
 
         sai->sai_max = LL_SA_RPC_MIN;
         sai->sai_index = 1;
@@ -507,7 +499,7 @@ static struct ll_statahead_info *ll_sai_alloc(void)
 
         for (i = 0; i < LL_SA_CACHE_SIZE; i++) {
                 CFS_INIT_LIST_HEAD(&sai->sai_cache[i]);
-                cfs_spin_lock_init(&sai->sai_cache_lock[i]);
+               spin_lock_init(&sai->sai_cache_lock[i]);
         }
         cfs_atomic_set(&sai->sai_cache_count, 0);
 
@@ -533,17 +525,17 @@ static void ll_sai_put(struct ll_statahead_info *sai)
                 if (unlikely(cfs_atomic_read(&sai->sai_refcount) > 0)) {
                         /* It is race case, the interpret callback just hold
                          * a reference count */
-                        cfs_spin_unlock(&lli->lli_sa_lock);
-                        RETURN_EXIT;
-                }
+                       spin_unlock(&lli->lli_sa_lock);
+                       RETURN_EXIT;
+               }
 
-                LASSERT(lli->lli_opendir_key == NULL);
-                LASSERT(thread_is_stopped(&sai->sai_thread));
-                LASSERT(thread_is_stopped(&sai->sai_agl_thread));
+               LASSERT(lli->lli_opendir_key == NULL);
+               LASSERT(thread_is_stopped(&sai->sai_thread));
+               LASSERT(thread_is_stopped(&sai->sai_agl_thread));
 
-                lli->lli_sai = NULL;
-                lli->lli_opendir_pid = 0;
-                cfs_spin_unlock(&lli->lli_sa_lock);
+               lli->lli_sai = NULL;
+               lli->lli_opendir_pid = 0;
+               spin_unlock(&lli->lli_sa_lock);
 
                 if (sai->sai_sent > sai->sai_replied)
                         CDEBUG(D_READA,"statahead for dir "DFID" does not "
@@ -571,10 +563,6 @@ static void ll_sai_put(struct ll_statahead_info *sai)
         EXIT;
 }
 
-#ifndef HAVE_AGL_SUPPORT
-# define cl_agl(inode)  do {} while (0)
-#endif
-
 /* Do NOT forget to drop inode refcount when into sai_entries_agl. */
 static void ll_agl_trigger(struct inode *inode, struct ll_statahead_info *sai)
 {
@@ -593,7 +581,7 @@ static void ll_agl_trigger(struct inode *inode, struct ll_statahead_info *sai)
         }
 
         /* Someone is in glimpse (sync or async), do nothing. */
-        rc = cfs_down_write_trylock(&lli->lli_glimpse_sem);
+       rc = down_write_trylock(&lli->lli_glimpse_sem);
         if (rc == 0) {
                 lli->lli_agl_index = 0;
                 iput(inode);
@@ -615,7 +603,7 @@ static void ll_agl_trigger(struct inode *inode, struct ll_statahead_info *sai)
          */
         if (lli->lli_glimpse_time != 0 &&
             cfs_time_before(cfs_time_shift(-1), lli->lli_glimpse_time)) {
-                cfs_up_write(&lli->lli_glimpse_sem);
+               up_write(&lli->lli_glimpse_sem);
                 lli->lli_agl_index = 0;
                 iput(inode);
                 RETURN_EXIT;
@@ -627,7 +615,7 @@ static void ll_agl_trigger(struct inode *inode, struct ll_statahead_info *sai)
         cl_agl(inode);
         lli->lli_agl_index = 0;
         lli->lli_glimpse_time = cfs_time_current();
-        cfs_up_write(&lli->lli_glimpse_sem);
+       up_write(&lli->lli_glimpse_sem);
 
         CDEBUG(D_READA, "Handled (init) async glimpse: inode= "
                DFID", idx = "LPU64", rc = %d\n",
@@ -652,20 +640,20 @@ static void do_statahead_interpret(struct ll_statahead_info *sai,
         int                     rc    = 0;
         ENTRY;
 
-        cfs_spin_lock(&lli->lli_sa_lock);
-        if (target != NULL && target->se_req != NULL &&
-            !cfs_list_empty(&target->se_list)) {
-                entry = target;
-        } else if (unlikely(sa_received_empty(sai))) {
-                cfs_spin_unlock(&lli->lli_sa_lock);
-                RETURN_EXIT;
-        } else {
-                entry = sa_first_received_entry(sai);
-        }
-
-        cfs_atomic_inc(&entry->se_refcount);
-        cfs_list_del_init(&entry->se_list);
-        cfs_spin_unlock(&lli->lli_sa_lock);
+       spin_lock(&lli->lli_sa_lock);
+       if (target != NULL && target->se_req != NULL &&
+           !cfs_list_empty(&target->se_list)) {
+               entry = target;
+       } else if (unlikely(sa_received_empty(sai))) {
+               spin_unlock(&lli->lli_sa_lock);
+               RETURN_EXIT;
+       } else {
+               entry = sa_first_received_entry(sai);
+       }
+
+       cfs_atomic_inc(&entry->se_refcount);
+       cfs_list_del_init(&entry->se_list);
+       spin_unlock(&lli->lli_sa_lock);
 
         LASSERT(entry->se_handle != 0);
 
@@ -700,11 +688,11 @@ static void do_statahead_interpret(struct ll_statahead_info *sai,
         }
 
         it->d.lustre.it_lock_handle = entry->se_handle;
-        rc = md_revalidate_lock(ll_i2mdexp(dir), it, NULL, NULL);
+       rc = md_revalidate_lock(ll_i2mdexp(dir), it, ll_inode2fid(dir), NULL);
         if (rc != 1)
                 GOTO(out, rc = -EAGAIN);
 
-        rc = ll_prep_inode(&child, req, dir->i_sb);
+        rc = ll_prep_inode(&child, req, dir->i_sb, it);
         if (rc)
                 GOTO(out, rc);
 
@@ -744,32 +732,32 @@ static int ll_statahead_interpret(struct ptlrpc_request *req,
         if (it_disposition(it, DISP_LOOKUP_NEG))
                 rc = -ENOENT;
 
-        cfs_spin_lock(&lli->lli_sa_lock);
-        /* stale entry */
-        if (unlikely(lli->lli_sai == NULL ||
-                     lli->lli_sai->sai_generation != minfo->mi_generation)) {
-                cfs_spin_unlock(&lli->lli_sa_lock);
-                GOTO(out, rc = -ESTALE);
-        } else {
-                sai = ll_sai_get(lli->lli_sai);
-                if (unlikely(!thread_is_running(&sai->sai_thread))) {
-                        sai->sai_replied++;
-                        cfs_spin_unlock(&lli->lli_sa_lock);
-                        GOTO(out, rc = -EBADFD);
-                }
-
-                entry = ll_sa_entry_get_byindex(sai, minfo->mi_cbdata);
-                if (entry == NULL) {
-                        sai->sai_replied++;
-                        cfs_spin_unlock(&lli->lli_sa_lock);
-                        GOTO(out, rc = -EIDRM);
-                }
-
-                cfs_list_del_init(&entry->se_list);
-                if (rc != 0) {
-                        sai->sai_replied++;
-                        do_sai_entry_to_stated(sai, entry, rc);
-                        cfs_spin_unlock(&lli->lli_sa_lock);
+       spin_lock(&lli->lli_sa_lock);
+       /* stale entry */
+       if (unlikely(lli->lli_sai == NULL ||
+                    lli->lli_sai->sai_generation != minfo->mi_generation)) {
+               spin_unlock(&lli->lli_sa_lock);
+               GOTO(out, rc = -ESTALE);
+       } else {
+               sai = ll_sai_get(lli->lli_sai);
+               if (unlikely(!thread_is_running(&sai->sai_thread))) {
+                       sai->sai_replied++;
+                       spin_unlock(&lli->lli_sa_lock);
+                       GOTO(out, rc = -EBADFD);
+               }
+
+               entry = ll_sa_entry_get_byindex(sai, minfo->mi_cbdata);
+               if (entry == NULL) {
+                       sai->sai_replied++;
+                       spin_unlock(&lli->lli_sa_lock);
+                       GOTO(out, rc = -EIDRM);
+               }
+
+               cfs_list_del_init(&entry->se_list);
+               if (rc != 0) {
+                       sai->sai_replied++;
+                       do_sai_entry_to_stated(sai, entry, rc);
+                       spin_unlock(&lli->lli_sa_lock);
                         if (entry->se_index == sai->sai_index_wait)
                                 cfs_waitq_signal(&sai->sai_waitq);
                 } else {
@@ -785,7 +773,7 @@ static int ll_statahead_interpret(struct ptlrpc_request *req,
                         cfs_list_add_tail(&entry->se_list,
                                           &sai->sai_entries_received);
                         sai->sai_replied++;
-                        cfs_spin_unlock(&lli->lli_sa_lock);
+                       spin_unlock(&lli->lli_sa_lock);
                         if (wakeup)
                                 cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
                 }
@@ -1020,10 +1008,10 @@ static int ll_agl_thread(void *arg)
                cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
 
         atomic_inc(&sbi->ll_agl_total);
-        cfs_spin_lock(&plli->lli_agl_lock);
-        sai->sai_agl_valid = 1;
-        thread_set_flags(thread, SVC_RUNNING);
-        cfs_spin_unlock(&plli->lli_agl_lock);
+       spin_lock(&plli->lli_agl_lock);
+       sai->sai_agl_valid = 1;
+       thread_set_flags(thread, SVC_RUNNING);
+       spin_unlock(&plli->lli_agl_lock);
         cfs_waitq_signal(&thread->t_ctl_waitq);
 
         while (1) {
@@ -1035,36 +1023,36 @@ static int ll_agl_thread(void *arg)
                 if (!thread_is_running(thread))
                         break;
 
-                cfs_spin_lock(&plli->lli_agl_lock);
-                /* The statahead thread maybe help to process AGL entries,
-                 * so check whether list empty again. */
-                if (!agl_list_empty(sai)) {
-                        clli = agl_first_entry(sai);
-                        cfs_list_del_init(&clli->lli_agl_list);
-                        cfs_spin_unlock(&plli->lli_agl_lock);
-                        ll_agl_trigger(&clli->lli_vfs_inode, sai);
-                } else {
-                        cfs_spin_unlock(&plli->lli_agl_lock);
-                }
-        }
-
-        cfs_spin_lock(&plli->lli_agl_lock);
-        sai->sai_agl_valid = 0;
-        while (!agl_list_empty(sai)) {
-                clli = agl_first_entry(sai);
-                cfs_list_del_init(&clli->lli_agl_list);
-                cfs_spin_unlock(&plli->lli_agl_lock);
-                clli->lli_agl_index = 0;
-                iput(&clli->lli_vfs_inode);
-                cfs_spin_lock(&plli->lli_agl_lock);
-        }
-        thread_set_flags(thread, SVC_STOPPED);
-        cfs_spin_unlock(&plli->lli_agl_lock);
-        cfs_waitq_signal(&thread->t_ctl_waitq);
-        ll_sai_put(sai);
-        CDEBUG(D_READA, "agl thread stopped: [pid %d] [parent %.*s]\n",
-               cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
-        RETURN(0);
+               spin_lock(&plli->lli_agl_lock);
+               /* The statahead thread maybe help to process AGL entries,
+                * so check whether list empty again. */
+               if (!agl_list_empty(sai)) {
+                       clli = agl_first_entry(sai);
+                       cfs_list_del_init(&clli->lli_agl_list);
+                       spin_unlock(&plli->lli_agl_lock);
+                       ll_agl_trigger(&clli->lli_vfs_inode, sai);
+               } else {
+                       spin_unlock(&plli->lli_agl_lock);
+               }
+       }
+
+       spin_lock(&plli->lli_agl_lock);
+       sai->sai_agl_valid = 0;
+       while (!agl_list_empty(sai)) {
+               clli = agl_first_entry(sai);
+               cfs_list_del_init(&clli->lli_agl_list);
+               spin_unlock(&plli->lli_agl_lock);
+               clli->lli_agl_index = 0;
+               iput(&clli->lli_vfs_inode);
+               spin_lock(&plli->lli_agl_lock);
+       }
+       thread_set_flags(thread, SVC_STOPPED);
+       spin_unlock(&plli->lli_agl_lock);
+       cfs_waitq_signal(&thread->t_ctl_waitq);
+       ll_sai_put(sai);
+       CDEBUG(D_READA, "agl thread stopped: [pid %d] [parent %.*s]\n",
+              cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
+       RETURN(0);
 }
 
 static void ll_start_agl(struct dentry *parent, struct ll_statahead_info *sai)
@@ -1080,6 +1068,7 @@ static void ll_start_agl(struct dentry *parent, struct ll_statahead_info *sai)
         rc = cfs_create_thread(ll_agl_thread, parent, 0);
         if (rc < 0) {
                 CERROR("can't start ll_agl thread, rc: %d\n", rc);
+                thread_set_flags(thread, SVC_STOPPED);
                 RETURN_EXIT;
         }
 
@@ -1098,6 +1087,7 @@ static int ll_statahead_thread(void *arg)
         struct ll_sb_info        *sbi    = ll_i2sbi(dir);
         struct ll_statahead_info *sai    = ll_sai_get(plli->lli_sai);
         struct ptlrpc_thread     *thread = &sai->sai_thread;
+        struct ptlrpc_thread *agl_thread = &sai->sai_agl_thread;
         struct page              *page;
         __u64                     pos    = 0;
         int                       first  = 0;
@@ -1119,14 +1109,13 @@ static int ll_statahead_thread(void *arg)
                 ll_start_agl(parent, sai);
 
         atomic_inc(&sbi->ll_sa_total);
-        cfs_spin_lock(&plli->lli_sa_lock);
-        thread_set_flags(thread, SVC_RUNNING);
-        cfs_spin_unlock(&plli->lli_sa_lock);
-        cfs_waitq_signal(&thread->t_ctl_waitq);
+       spin_lock(&plli->lli_sa_lock);
+       thread_set_flags(thread, SVC_RUNNING);
+       spin_unlock(&plli->lli_sa_lock);
+       cfs_waitq_signal(&thread->t_ctl_waitq);
 
-        plli->lli_sa_pos = 0;
-        ll_dir_chain_init(&chain);
-        page = ll_get_dir_page(NULL, dir, pos, &chain);
+       ll_dir_chain_init(&chain);
+       page = ll_get_dir_page(dir, pos, &chain);
 
         while (1) {
                 struct lu_dirpage *dp;
@@ -1210,11 +1199,11 @@ interpret_it:
                          * some AGL entries to be triggered, then try to help
                          * to process the AGL entries. */
                         if (sa_sent_full(sai)) {
-                                cfs_spin_lock(&plli->lli_agl_lock);
-                                while (!agl_list_empty(sai)) {
-                                        clli = agl_first_entry(sai);
-                                        cfs_list_del_init(&clli->lli_agl_list);
-                                        cfs_spin_unlock(&plli->lli_agl_lock);
+                               spin_lock(&plli->lli_agl_lock);
+                               while (!agl_list_empty(sai)) {
+                                       clli = agl_first_entry(sai);
+                                       cfs_list_del_init(&clli->lli_agl_list);
+                                       spin_unlock(&plli->lli_agl_lock);
                                         ll_agl_trigger(&clli->lli_vfs_inode,
                                                        sai);
 
@@ -1230,9 +1219,9 @@ interpret_it:
                                         if (!sa_sent_full(sai))
                                                 goto do_it;
 
-                                        cfs_spin_lock(&plli->lli_agl_lock);
-                                }
-                                cfs_spin_unlock(&plli->lli_agl_lock);
+                                       spin_lock(&plli->lli_agl_lock);
+                               }
+                               spin_unlock(&plli->lli_agl_lock);
 
                                 goto keep_it;
                         }
@@ -1264,16 +1253,16 @@ do_it:
                                         break;
                         }
 
-                        cfs_spin_lock(&plli->lli_agl_lock);
-                        while (!agl_list_empty(sai) &&
-                               thread_is_running(thread)) {
-                                clli = agl_first_entry(sai);
-                                cfs_list_del_init(&clli->lli_agl_list);
-                                cfs_spin_unlock(&plli->lli_agl_lock);
-                                ll_agl_trigger(&clli->lli_vfs_inode, sai);
-                                cfs_spin_lock(&plli->lli_agl_lock);
-                        }
-                        cfs_spin_unlock(&plli->lli_agl_lock);
+                       spin_lock(&plli->lli_agl_lock);
+                       while (!agl_list_empty(sai) &&
+                              thread_is_running(thread)) {
+                               clli = agl_first_entry(sai);
+                               cfs_list_del_init(&clli->lli_agl_list);
+                               spin_unlock(&plli->lli_agl_lock);
+                               ll_agl_trigger(&clli->lli_vfs_inode, sai);
+                               spin_lock(&plli->lli_agl_lock);
+                       }
+                       spin_unlock(&plli->lli_agl_lock);
 
                         GOTO(out, rc = 0);
                 } else if (1) {
@@ -1283,9 +1272,8 @@ do_it:
                          */
                         ll_release_page(page, le32_to_cpu(dp->ldp_flags) &
                                               LDF_COLLIDE);
-                        plli->lli_sa_pos = pos;
                         sai->sai_in_readpage = 1;
-                        page = ll_get_dir_page(NULL, dir, pos, &chain);
+                       page = ll_get_dir_page(dir, pos, &chain);
                         sai->sai_in_readpage = 0;
                 } else {
                         LASSERT(le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
@@ -1299,11 +1287,9 @@ do_it:
 
 out:
         if (sai->sai_agl_valid) {
-                struct ptlrpc_thread *agl_thread = &sai->sai_agl_thread;
-
-                cfs_spin_lock(&plli->lli_agl_lock);
-                thread_set_flags(agl_thread, SVC_STOPPING);
-                cfs_spin_unlock(&plli->lli_agl_lock);
+               spin_lock(&plli->lli_agl_lock);
+               thread_set_flags(agl_thread, SVC_STOPPING);
+               spin_unlock(&plli->lli_agl_lock);
                 cfs_waitq_signal(&agl_thread->t_ctl_waitq);
 
                 CDEBUG(D_READA, "stop agl thread: [pid %d]\n",
@@ -1311,22 +1297,24 @@ out:
                 l_wait_event(agl_thread->t_ctl_waitq,
                              thread_is_stopped(agl_thread),
                              &lwi);
+        } else {
+                /* Set agl_thread flags anyway. */
+                thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
         }
-
         ll_dir_chain_fini(&chain);
-        cfs_spin_lock(&plli->lli_sa_lock);
-        if (!sa_received_empty(sai)) {
-                thread_set_flags(thread, SVC_STOPPING);
-                cfs_spin_unlock(&plli->lli_sa_lock);
-
-                /* To release the resources held by received entries. */
-                while (!sa_received_empty(sai))
-                        do_statahead_interpret(sai, NULL);
-
-                cfs_spin_lock(&plli->lli_sa_lock);
-        }
-        thread_set_flags(thread, SVC_STOPPED);
-        cfs_spin_unlock(&plli->lli_sa_lock);
+       spin_lock(&plli->lli_sa_lock);
+       if (!sa_received_empty(sai)) {
+               thread_set_flags(thread, SVC_STOPPING);
+               spin_unlock(&plli->lli_sa_lock);
+
+               /* To release the resources held by received entries. */
+               while (!sa_received_empty(sai))
+                       do_statahead_interpret(sai, NULL);
+
+               spin_lock(&plli->lli_sa_lock);
+       }
+       thread_set_flags(thread, SVC_STOPPED);
+       spin_unlock(&plli->lli_sa_lock);
         cfs_waitq_signal(&sai->sai_waitq);
         cfs_waitq_signal(&thread->t_ctl_waitq);
         ll_sai_put(sai);
@@ -1341,14 +1329,14 @@ out:
  */
 void ll_stop_statahead(struct inode *dir, void *key)
 {
-        struct ll_inode_info *lli = ll_i2info(dir);
+       struct ll_inode_info *lli = ll_i2info(dir);
 
-        if (unlikely(key == NULL))
-                return;
+       if (unlikely(key == NULL))
+               return;
 
-        cfs_spin_lock(&lli->lli_sa_lock);
-        if (lli->lli_opendir_key != key || lli->lli_opendir_pid == 0) {
-                cfs_spin_unlock(&lli->lli_sa_lock);
+       spin_lock(&lli->lli_sa_lock);
+       if (lli->lli_opendir_key != key || lli->lli_opendir_pid == 0) {
+               spin_unlock(&lli->lli_sa_lock);
                 return;
         }
 
@@ -1360,28 +1348,28 @@ void ll_stop_statahead(struct inode *dir, void *key)
 
                 if (!thread_is_stopped(thread)) {
                         thread_set_flags(thread, SVC_STOPPING);
-                        cfs_spin_unlock(&lli->lli_sa_lock);
-                        cfs_waitq_signal(&thread->t_ctl_waitq);
-
-                        CDEBUG(D_READA, "stop statahead thread: [pid %d]\n",
-                               cfs_curproc_pid());
-                        l_wait_event(thread->t_ctl_waitq,
-                                     thread_is_stopped(thread),
-                                     &lwi);
-                } else {
-                        cfs_spin_unlock(&lli->lli_sa_lock);
-                }
-
-                /*
-                 * Put the ref which was held when first statahead_enter.
-                 * It maybe not the last ref for some statahead requests
-                 * maybe inflight.
-                 */
-                ll_sai_put(lli->lli_sai);
-        } else {
-                lli->lli_opendir_pid = 0;
-                cfs_spin_unlock(&lli->lli_sa_lock);
-        }
+                       spin_unlock(&lli->lli_sa_lock);
+                       cfs_waitq_signal(&thread->t_ctl_waitq);
+
+                       CDEBUG(D_READA, "stop statahead thread: [pid %d]\n",
+                              cfs_curproc_pid());
+                       l_wait_event(thread->t_ctl_waitq,
+                                    thread_is_stopped(thread),
+                                    &lwi);
+               } else {
+                       spin_unlock(&lli->lli_sa_lock);
+               }
+
+               /*
+                * Put the ref which was held when first statahead_enter.
+                * It maybe not the last ref for some statahead requests
+                * maybe inflight.
+                */
+               ll_sai_put(lli->lli_sai);
+       } else {
+               lli->lli_opendir_pid = 0;
+               spin_unlock(&lli->lli_sa_lock);
+       }
 }
 
 enum {
@@ -1401,7 +1389,6 @@ enum {
 
 static int is_first_dirent(struct inode *dir, struct dentry *dentry)
 {
-        struct ll_inode_info *lli    = ll_i2info(dir);
         struct ll_dir_chain   chain;
         struct qstr          *target = &dentry->d_name;
         struct page          *page;
@@ -1410,9 +1397,8 @@ static int is_first_dirent(struct inode *dir, struct dentry *dentry)
         int                   rc     = LS_NONE_FIRST_DE;
         ENTRY;
 
-        lli->lli_sa_pos = 0;
         ll_dir_chain_init(&chain);
-        page = ll_get_dir_page(NULL, dir, pos, &chain);
+       page = ll_get_dir_page(dir, pos, &chain);
 
         while (1) {
                 struct lu_dirpage *dp;
@@ -1499,8 +1485,7 @@ static int is_first_dirent(struct inode *dir, struct dentry *dentry)
                          */
                         ll_release_page(page, le32_to_cpu(dp->ldp_flags) &
                                               LDF_COLLIDE);
-                        lli->lli_sa_pos = pos;
-                        page = ll_get_dir_page(NULL, dir, pos, &chain);
+                       page = ll_get_dir_page(dir, pos, &chain);
                 } else {
                         /*
                          * go into overflow page.
@@ -1548,17 +1533,17 @@ ll_sai_unplug(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
                                PFID(&lli->lli_fid), sai->sai_hit,
                                sai->sai_miss, sai->sai_sent,
                                sai->sai_replied, cfs_curproc_pid());
-                        cfs_spin_lock(&lli->lli_sa_lock);
-                        if (!thread_is_stopped(thread))
-                                thread_set_flags(thread, SVC_STOPPING);
-                        cfs_spin_unlock(&lli->lli_sa_lock);
-                }
-        }
+                       spin_lock(&lli->lli_sa_lock);
+                       if (!thread_is_stopped(thread))
+                               thread_set_flags(thread, SVC_STOPPING);
+                       spin_unlock(&lli->lli_sa_lock);
+               }
+       }
 
-        if (!thread_is_stopped(thread))
-                cfs_waitq_signal(&thread->t_ctl_waitq);
+       if (!thread_is_stopped(thread))
+               cfs_waitq_signal(&thread->t_ctl_waitq);
 
-        EXIT;
+       EXIT;
 }
 
 /**
@@ -1651,27 +1636,35 @@ int do_statahead_enter(struct inode *dir, struct dentry **dentryp,
                         struct lookup_intent it = { .it_op = IT_GETATTR,
                                                     .d.lustre.it_lock_handle =
                                                      entry->se_handle };
-                        struct ll_dentry_data *lld;
-                        __u64 bits;
-
-                        rc = md_revalidate_lock(ll_i2mdexp(dir), &it,
-                                                ll_inode2fid(inode), &bits);
-                        if (rc == 1) {
-                                if ((*dentryp)->d_inode == NULL) {
-                                        *dentryp = ll_find_alias(inode,
-                                                                 *dentryp);
-                                        lld = ll_d2d(*dentryp);
-                                        if (unlikely(lld == NULL))
-                                                ll_dops_init(*dentryp, 1, 1);
+                       __u64 bits;
+
+                       rc = md_revalidate_lock(ll_i2mdexp(dir), &it,
+                                               ll_inode2fid(inode), &bits);
+                       if (rc == 1) {
+                               if ((*dentryp)->d_inode == NULL) {
+                                       *dentryp = ll_splice_alias(inode,
+                                                                  *dentryp);
+                                } else if ((*dentryp)->d_inode != inode) {
+                                        /* revalidate, but inode is recreated */
+                                        CDEBUG(D_READA,
+                                              "stale dentry %.*s inode %lu/%u, "
+                                              "statahead inode %lu/%u\n",
+                                              (*dentryp)->d_name.len,
+                                              (*dentryp)->d_name.name,
+                                              (*dentryp)->d_inode->i_ino,
+                                              (*dentryp)->d_inode->i_generation,
+                                              inode->i_ino,
+                                              inode->i_generation);
+                                        ll_sai_unplug(sai, entry);
+                                        RETURN(-ESTALE);
                                 } else {
-                                        LASSERT((*dentryp)->d_inode == inode);
+                                       iput(inode);
+                               }
+                               entry->se_inode = NULL;
 
-                                        ll_dentry_rehash(*dentryp, 0);
-                                        iput(inode);
-                                }
-                                entry->se_inode = NULL;
-
-                                ll_dentry_reset_flags(*dentryp, bits);
+                               if ((bits & MDS_INODELOCK_LOOKUP) &&
+                                   d_lustre_invalid(*dentryp))
+                                       d_lustre_revalidate(*dentryp);
                                 ll_intent_release(&it);
                         }
                 }
@@ -1723,6 +1716,7 @@ int do_statahead_enter(struct inode *dir, struct dentry **dentryp,
                 dput(parent);
                 lli->lli_opendir_key = NULL;
                 thread_set_flags(thread, SVC_STOPPED);
+                thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
                 ll_sai_put(sai);
                 LASSERT(lli->lli_sai == NULL);
                 RETURN(-EAGAIN);
@@ -1734,16 +1728,16 @@ int do_statahead_enter(struct inode *dir, struct dentry **dentryp,
 
         /*
          * We don't stat-ahead for the first dirent since we are already in
-         * lookup, and -EEXIST also indicates that this is the first dirent.
+         * lookup.
          */
-        RETURN(-EEXIST);
+        RETURN(-EAGAIN);
 
 out:
         if (sai != NULL)
                 OBD_FREE_PTR(sai);
-        cfs_spin_lock(&lli->lli_sa_lock);
-        lli->lli_opendir_key = NULL;
-        lli->lli_opendir_pid = 0;
-        cfs_spin_unlock(&lli->lli_sa_lock);
-        return rc;
+       spin_lock(&lli->lli_sa_lock);
+       lli->lli_opendir_key = NULL;
+       lli->lli_opendir_pid = 0;
+       spin_unlock(&lli->lli_sa_lock);
+       return rc;
 }