-/* -*- 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.
* 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, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#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>
#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
} se_stat_t;
struct ll_sa_entry {
- /* link into sai->sai_entries_{sent,received,stated} */
- cfs_list_t se_list;
- /* link into sai hash table locally */
- cfs_list_t se_hash;
- /* entry reference count */
- cfs_atomic_t se_refcount;
- /* entry index in the sai */
- __u64 se_index;
- /* low layer ldlm lock handle */
- __u64 se_handle;
- /* entry status */
- se_stat_t se_stat;
- /* entry size, contains name */
- int se_size;
- /* pointer to async getattr enqueue info */
- struct md_enqueue_info *se_minfo;
- /* pointer to the async getattr request */
- struct ptlrpc_request *se_req;
- /* pointer to the target inode */
- struct inode *se_inode;
- /* entry name */
- struct qstr se_qstr;
+ /* link into sai->sai_entries */
+ cfs_list_t se_link;
+ /* link into sai->sai_entries_{received,stated} */
+ cfs_list_t se_list;
+ /* link into sai hash table locally */
+ cfs_list_t se_hash;
+ /* entry reference count */
+ cfs_atomic_t se_refcount;
+ /* entry index in the sai */
+ __u64 se_index;
+ /* low layer ldlm lock handle */
+ __u64 se_handle;
+ /* entry status */
+ se_stat_t se_stat;
+ /* entry size, contains name */
+ int se_size;
+ /* pointer to async getattr enqueue info */
+ struct md_enqueue_info *se_minfo;
+ /* pointer to the async getattr request */
+ struct ptlrpc_request *se_req;
+ /* pointer to the target inode */
+ struct inode *se_inode;
+ /* entry name */
+ struct qstr se_qstr;
};
static unsigned int sai_generation = 0;
-static cfs_spinlock_t sai_generation_lock = CFS_SPIN_LOCK_UNLOCKED;
-
-static inline int ll_sa_entry_unlinked(struct ll_sa_entry *entry)
-{
- return cfs_list_empty(&entry->se_list);
-}
+static DEFINE_SPINLOCK(sai_generation_lock);
static inline int ll_sa_entry_unhashed(struct ll_sa_entry *entry)
{
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]);
}
/*
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 *
}
/*
- * Insert it into sai_entries_sent tail when init.
+ * Insert it into sai_entries tail when init.
*/
static struct ll_sa_entry *
ll_sa_entry_alloc(struct ll_statahead_info *sai, __u64 index,
if (unlikely(entry == NULL))
RETURN(ERR_PTR(-ENOMEM));
- CDEBUG(D_READA, "alloc sai entry %.*s(%p) index "LPU64"\n",
- len, name, entry, index);
+ CDEBUG(D_READA, "alloc sa entry %.*s(%p) index "LPU64"\n",
+ len, name, entry, index);
entry->se_index = 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_link, &sai->sai_entries);
+ CFS_INIT_LIST_HEAD(&entry->se_list);
+ ll_sa_entry_enhash(sai, entry);
+ 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);
- RETURN(entry);
+ RETURN(entry);
}
/*
static struct ll_sa_entry *
ll_sa_entry_get_byindex(struct ll_statahead_info *sai, __u64 index)
{
- struct ll_sa_entry *entry;
+ struct ll_sa_entry *entry;
- cfs_list_for_each_entry(entry, &sai->sai_entries_sent, se_list) {
- if (entry->se_index == index) {
- cfs_atomic_inc(&entry->se_refcount);
- return entry;
+ cfs_list_for_each_entry(entry, &sai->sai_entries, se_link) {
+ if (entry->se_index == index) {
+ LASSERT(atomic_read(&entry->se_refcount) > 0);
+ cfs_atomic_inc(&entry->se_refcount);
+ return entry;
}
- if (entry->se_index > index)
- break;
- }
- return NULL;
+ if (entry->se_index > index)
+ break;
+ }
+ return NULL;
}
static void ll_sa_entry_cleanup(struct ll_statahead_info *sai,
static void ll_sa_entry_put(struct ll_statahead_info *sai,
struct ll_sa_entry *entry)
{
- if (cfs_atomic_dec_and_test(&entry->se_refcount)) {
- CDEBUG(D_READA, "free sai entry %.*s(%p) index "LPU64"\n",
- entry->se_qstr.len, entry->se_qstr.name, entry,
- entry->se_index);
-
- LASSERT(ll_sa_entry_unhashed(entry));
- LASSERT(ll_sa_entry_unlinked(entry));
-
- ll_sa_entry_cleanup(sai, entry);
- if (entry->se_inode)
- iput(entry->se_inode);
-
- OBD_FREE(entry, entry->se_size);
- cfs_atomic_dec(&sai->sai_cache_count);
- }
+ if (cfs_atomic_dec_and_test(&entry->se_refcount)) {
+ CDEBUG(D_READA, "free sa entry %.*s(%p) index "LPU64"\n",
+ entry->se_qstr.len, entry->se_qstr.name, entry,
+ entry->se_index);
+
+ LASSERT(cfs_list_empty(&entry->se_link));
+ LASSERT(cfs_list_empty(&entry->se_list));
+ LASSERT(ll_sa_entry_unhashed(entry));
+
+ ll_sa_entry_cleanup(sai, entry);
+ if (entry->se_inode)
+ iput(entry->se_inode);
+
+ OBD_FREE(entry, entry->se_size);
+ cfs_atomic_dec(&sai->sai_cache_count);
+ }
}
static inline void
-do_sai_entry_fini(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
+do_sa_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);
+ LASSERT(!ll_sa_entry_unhashed(entry));
+ LASSERT(!cfs_list_empty(&entry->se_link));
- 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);
+ ll_sa_entry_unhash(sai, entry);
- ll_sa_entry_put(sai, entry);
+ spin_lock(&lli->lli_sa_lock);
+ entry->se_stat = SA_ENTRY_DEST;
+ cfs_list_del_init(&entry->se_link);
+ if (likely(!cfs_list_empty(&entry->se_list)))
+ cfs_list_del_init(&entry->se_list);
+ spin_unlock(&lli->lli_sa_lock);
+
+ ll_sa_entry_put(sai, entry);
}
/*
static void
ll_sa_entry_fini(struct ll_statahead_info *sai, struct ll_sa_entry *entry)
{
- struct ll_sa_entry *pos, *next;
-
- if (entry)
- do_sai_entry_fini(sai, entry);
+ struct ll_sa_entry *pos, *next;
- /* drop old entry from sent list */
- cfs_list_for_each_entry_safe(pos, next, &sai->sai_entries_sent,
- se_list) {
- if (is_omitted_entry(sai, pos->se_index))
- do_sai_entry_fini(sai, pos);
- else
- break;
- }
+ if (entry)
+ do_sa_entry_fini(sai, entry);
- /* drop old entry from stated list */
- cfs_list_for_each_entry_safe(pos, next, &sai->sai_entries_stated,
- se_list) {
- if (is_omitted_entry(sai, pos->se_index))
- do_sai_entry_fini(sai, pos);
- else
- break;
- }
+ /* drop old entry, only 'scanner' process does this, no need to lock */
+ cfs_list_for_each_entry_safe(pos, next, &sai->sai_entries, se_link) {
+ if (!is_omitted_entry(sai, pos->se_index))
+ break;
+ do_sa_entry_fini(sai, pos);
+ }
}
/*
* Inside lli_sa_lock.
*/
static void
-do_sai_entry_to_stated(struct ll_statahead_info *sai,
- struct ll_sa_entry *entry, int rc)
+do_sa_entry_to_stated(struct ll_statahead_info *sai,
+ struct ll_sa_entry *entry, se_stat_t stat)
{
- struct ll_sa_entry *se;
- cfs_list_t *pos = &sai->sai_entries_stated;
+ struct ll_sa_entry *se;
+ cfs_list_t *pos = &sai->sai_entries_stated;
- if (!ll_sa_entry_unlinked(entry))
- cfs_list_del_init(&entry->se_list);
+ if (!cfs_list_empty(&entry->se_list))
+ cfs_list_del_init(&entry->se_list);
- cfs_list_for_each_entry_reverse(se, &sai->sai_entries_stated, se_list) {
- if (se->se_index < entry->se_index) {
- pos = &se->se_list;
- break;
- }
- }
+ cfs_list_for_each_entry_reverse(se, &sai->sai_entries_stated, se_list) {
+ if (se->se_index < entry->se_index) {
+ pos = &se->se_list;
+ break;
+ }
+ }
- cfs_list_add(&entry->se_list, pos);
- entry->se_stat = rc;
+ cfs_list_add(&entry->se_list, pos);
+ entry->se_stat = stat;
}
/*
*/
static int
ll_sa_entry_to_stated(struct ll_statahead_info *sai,
- struct ll_sa_entry *entry, int rc)
+ struct ll_sa_entry *entry, se_stat_t stat)
{
- 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_sa_entry_to_stated(sai, entry, stat);
+ ret = 0;
+ }
+ spin_unlock(&lli->lli_sa_lock);
- return ret;
+ return ret;
}
/*
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)
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;
cfs_waitq_init(&sai->sai_thread.t_ctl_waitq);
cfs_waitq_init(&sai->sai_agl_thread.t_ctl_waitq);
- CFS_INIT_LIST_HEAD(&sai->sai_entries_sent);
+ CFS_INIT_LIST_HEAD(&sai->sai_entries);
CFS_INIT_LIST_HEAD(&sai->sai_entries_received);
CFS_INIT_LIST_HEAD(&sai->sai_entries_stated);
CFS_INIT_LIST_HEAD(&sai->sai_entries_agl);
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);
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 "
PFID(&lli->lli_fid),
sai->sai_sent, sai->sai_replied);
- cfs_list_for_each_entry_safe(entry, next,
- &sai->sai_entries_sent, se_list)
- do_sai_entry_fini(sai, entry);
-
- LASSERT(sa_received_empty(sai));
+ cfs_list_for_each_entry_safe(entry, next,
+ &sai->sai_entries, se_link)
+ do_sa_entry_fini(sai, entry);
- cfs_list_for_each_entry_safe(entry, next,
- &sai->sai_entries_stated, se_list)
- do_sai_entry_fini(sai, entry);
+ LASSERT(list_empty(&sai->sai_entries));
+ LASSERT(sa_received_empty(sai));
+ LASSERT(list_empty(&sai->sai_entries_stated));
LASSERT(cfs_atomic_read(&sai->sai_cache_count) == 0);
LASSERT(agl_list_empty(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)
{
}
/* 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);
*/
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;
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",
EXIT;
}
-static void do_statahead_interpret(struct ll_statahead_info *sai,
- struct ll_sa_entry *target)
+static void ll_post_statahead(struct ll_statahead_info *sai)
{
struct inode *dir = sai->sai_inode;
struct inode *child;
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 (unlikely(sa_received_empty(sai))) {
+ spin_unlock(&lli->lli_sa_lock);
+ RETURN_EXIT;
+ }
+ 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);
}
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);
* reference count by calling "ll_intent_drop_lock()" in spite of the
* above operations failed or not. Do not worry about calling
* "ll_intent_drop_lock()" more than once. */
- rc = ll_sa_entry_to_stated(sai, entry, rc < 0 ? rc : SA_ENTRY_SUCC);
- if (rc == 0 && entry->se_index == sai->sai_index_wait && target == NULL)
+ rc = ll_sa_entry_to_stated(sai, entry,
+ rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
+ if (rc == 0 && entry->se_index == sai->sai_index_wait)
cfs_waitq_signal(&sai->sai_waitq);
ll_sa_entry_put(sai, entry);
}
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);
- if (entry->se_index == sai->sai_index_wait)
- cfs_waitq_signal(&sai->sai_waitq);
+ 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);
+ }
+
+ if (rc != 0) {
+ do_sa_entry_to_stated(sai, entry, SA_ENTRY_INVA);
+ wakeup = (entry->se_index == sai->sai_index_wait);
} else {
- entry->se_minfo = minfo;
- entry->se_req = ptlrpc_request_addref(req);
- /* Release the async ibits lock ASAP to avoid deadlock
- * when statahead thread tries to enqueue lock on parent
- * for readpage and other tries to enqueue lock on child
- * with parent's lock held, for example: unlink. */
- entry->se_handle = it->d.lustre.it_lock_handle;
- ll_intent_drop_lock(it);
- wakeup = sa_received_empty(sai);
- cfs_list_add_tail(&entry->se_list,
- &sai->sai_entries_received);
- sai->sai_replied++;
- cfs_spin_unlock(&lli->lli_sa_lock);
- if (wakeup)
- cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
+ entry->se_minfo = minfo;
+ entry->se_req = ptlrpc_request_addref(req);
+ /* Release the async ibits lock ASAP to avoid deadlock
+ * when statahead thread tries to enqueue lock on parent
+ * for readpage and other tries to enqueue lock on child
+ * with parent's lock held, for example: unlink. */
+ entry->se_handle = it->d.lustre.it_lock_handle;
+ ll_intent_drop_lock(it);
+ wakeup = sa_received_empty(sai);
+ cfs_list_add_tail(&entry->se_list,
+ &sai->sai_entries_received);
}
- ll_sa_entry_put(sai, entry);
+ sai->sai_replied++;
+ spin_unlock(&lli->lli_sa_lock);
+
+ ll_sa_entry_put(sai, entry);
+ if (wakeup)
+ cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
}
EXIT;
if (d_mountpoint(dentry))
RETURN(1);
- if (unlikely(dentry == dentry->d_sb->s_root))
- RETURN(1);
-
entry->se_inode = igrab(inode);
rc = md_revalidate_lock(ll_i2mdexp(dir), &it, ll_inode2fid(inode),NULL);
if (rc == 1) {
struct l_wait_info lwi = { 0 };
ENTRY;
- {
- char pname[16];
- snprintf(pname, 15, "ll_agl_%u", plli->lli_opendir_pid);
- cfs_daemonize(pname);
- }
-
CDEBUG(D_READA, "agl thread started: [pid %d] [parent %.*s]\n",
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) {
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)
{
- struct ptlrpc_thread *thread = &sai->sai_agl_thread;
- struct l_wait_info lwi = { 0 };
- int rc;
- ENTRY;
-
- CDEBUG(D_READA, "start agl thread: [pid %d] [parent %.*s]\n",
- cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
-
- rc = cfs_create_thread(ll_agl_thread, parent, 0);
- if (rc < 0) {
- CERROR("can't start ll_agl thread, rc: %d\n", rc);
- RETURN_EXIT;
- }
-
- l_wait_event(thread->t_ctl_waitq,
- thread_is_running(thread) || thread_is_stopped(thread),
- &lwi);
- EXIT;
+ struct ptlrpc_thread *thread = &sai->sai_agl_thread;
+ struct l_wait_info lwi = { 0 };
+ struct ll_inode_info *plli;
+ cfs_task_t *task;
+ ENTRY;
+
+ CDEBUG(D_READA, "start agl thread: [pid %d] [parent %.*s]\n",
+ cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
+
+ plli = ll_i2info(parent->d_inode);
+ task = kthread_run(ll_agl_thread, parent,
+ "ll_agl_%u", plli->lli_opendir_pid);
+ if (IS_ERR(task)) {
+ CERROR("can't start ll_agl thread, rc: %ld\n", PTR_ERR(task));
+ thread_set_flags(thread, SVC_STOPPED);
+ RETURN_EXIT;
+ }
+
+ l_wait_event(thread->t_ctl_waitq,
+ thread_is_running(thread) || thread_is_stopped(thread),
+ &lwi);
+ EXIT;
}
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;
struct l_wait_info lwi = { 0 };
ENTRY;
- {
- char pname[16];
- snprintf(pname, 15, "ll_sa_%u", plli->lli_opendir_pid);
- cfs_daemonize(pname);
- }
-
CDEBUG(D_READA, "statahead thread started: [pid %d] [parent %.*s]\n",
cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
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;
interpret_it:
while (!sa_received_empty(sai))
- do_statahead_interpret(sai, NULL);
+ ll_post_statahead(sai);
if (unlikely(!thread_is_running(thread))) {
ll_release_page(page, 0);
* 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);
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;
}
&lwi);
while (!sa_received_empty(sai))
- do_statahead_interpret(sai, NULL);
+ ll_post_statahead(sai);
if (unlikely(!thread_is_running(thread)))
GOTO(out, rc = 0);
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) {
*/
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);
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",
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))
+ ll_post_statahead(sai);
+
+ 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);
*/
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;
}
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 {
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;
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;
*/
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.
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;
}
/**
struct ptlrpc_thread *thread;
struct l_wait_info lwi = { 0 };
int rc = 0;
+ struct ll_inode_info *plli;
ENTRY;
LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
RETURN(entry ? 1 : -EAGAIN);
}
+ /* if statahead is busy in readdir, help it do post-work */
while (!ll_sa_entry_stated(entry) &&
sai->sai_in_readpage &&
!sa_received_empty(sai))
- do_statahead_interpret(sai, entry);
+ ll_post_statahead(sai);
if (!ll_sa_entry_stated(entry)) {
sai->sai_index_wait = entry->se_index;
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);
+ if (IS_ERR(*dentryp)) {
+ ll_sai_unplug(sai, entry);
+ RETURN(PTR_ERR(*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);
}
}
cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
lli->lli_sai = sai;
- rc = cfs_create_thread(ll_statahead_thread, parent, 0);
- thread = &sai->sai_thread;
- if (rc < 0) {
- CERROR("can't start ll_sa thread, rc: %d\n", rc);
- dput(parent);
+
+ plli = ll_i2info(parent->d_inode);
+ rc = PTR_ERR(kthread_run(ll_statahead_thread, parent,
+ "ll_sa_%u", plli->lli_opendir_pid));
+ thread = &sai->sai_thread;
+ if (IS_ERR_VALUE(rc)) {
+ CERROR("can't start ll_sa thread, rc: %d\n", rc);
+ 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);
/*
* 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;
}