* in the LICENSE file that accompanied this code).
*
* You should have received a copy of the GNU General Public License
- * version 2 along with this program; If not, see [sun.com URL with a
- * copy of GPLv2].
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
*
* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
* CA 95054 USA or visit www.sun.com if you need additional information or
SA_ENTRY_STATED
};
-struct dentry_operations ll_sai_d_ops = {
- .d_release = ll_release,
-};
-
static unsigned int sai_generation = 0;
static spinlock_t sai_generation_lock = SPIN_LOCK_UNLOCKED;
static inline int sa_not_full(struct ll_statahead_info *sai)
{
- return sai->sai_index < sai->sai_hit + sai->sai_miss + sai->sai_max;
+ return (sai->sai_index < sai->sai_hit + sai->sai_miss + sai->sai_max);
}
static inline int sa_is_running(struct ll_statahead_info *sai)
*/
static inline int sa_low_hit(struct ll_statahead_info *sai)
{
- return ((sai->sai_hit < 4 * sai->sai_miss && sai->sai_hit > 7) ||
+ return ((sai->sai_hit > 7 && sai->sai_hit < 4 * sai->sai_miss) ||
(sai->sai_consecutive_miss > 8));
}
* (3) drop dentry's ref count
* (4) release request's ref count
*/
-static void ll_sai_entry_cleanup(struct ll_sai_entry *entry)
+static void ll_sai_entry_cleanup(struct ll_sai_entry *entry, int free)
{
- struct ptlrpc_request *req = entry->se_req;
struct md_enqueue_info *minfo = entry->se_minfo;
+ struct ptlrpc_request *req = entry->se_req;
ENTRY;
if (minfo) {
- struct dentry *dentry = minfo->mi_dentry;
- struct lookup_intent *it = &minfo->mi_it;
-
entry->se_minfo = NULL;
- ll_intent_release(it);
+ ll_intent_release(&minfo->mi_it);
+ dput(minfo->mi_dentry);
+ iput(minfo->mi_dir);
OBD_FREE_PTR(minfo);
- dput(dentry);
}
if (req) {
entry->se_req = NULL;
ptlrpc_req_finished(req);
}
- OBD_FREE_PTR(entry);
+ if (free) {
+ LASSERT(list_empty(&entry->se_list));
+ OBD_FREE_PTR(entry);
+ }
EXIT;
}
static void ll_sai_put(struct ll_statahead_info *sai)
{
struct inode *inode = sai->sai_inode;
- struct ll_inode_info *lli = ll_i2info(inode);
+ struct ll_inode_info *lli;
ENTRY;
- if (atomic_dec_and_lock(&sai->sai_refcount, &lli->lli_lock)) {
+ LASSERT(inode != NULL);
+ lli = ll_i2info(inode);
+ LASSERT(lli->lli_sai == sai);
+
+ if (atomic_dec_and_test(&sai->sai_refcount)) {
struct ll_sai_entry *entry, *next;
+ spin_lock(&lli->lli_lock);
+ if (unlikely(atomic_read(&sai->sai_refcount) > 0)) {
+ /* It is race case, the interpret callback just hold
+ * a reference count */
+ spin_unlock(&lli->lli_lock);
+ EXIT;
+ return;
+ }
+
+ LASSERT(lli->lli_opendir_key == NULL);
lli->lli_sai = NULL;
+ lli->lli_opendir_pid = 0;
spin_unlock(&lli->lli_lock);
LASSERT(sa_is_stopped(sai));
list_for_each_entry_safe(entry, next, &sai->sai_entries_sent,
se_list) {
- list_del(&entry->se_list);
- ll_sai_entry_cleanup(entry);
+ list_del_init(&entry->se_list);
+ ll_sai_entry_cleanup(entry, 1);
}
list_for_each_entry_safe(entry, next, &sai->sai_entries_received,
se_list) {
- list_del(&entry->se_list);
- ll_sai_entry_cleanup(entry);
+ list_del_init(&entry->se_list);
+ ll_sai_entry_cleanup(entry, 1);
}
list_for_each_entry_safe(entry, next, &sai->sai_entries_stated,
se_list) {
- list_del(&entry->se_list);
- ll_sai_entry_cleanup(entry);
+ list_del_init(&entry->se_list);
+ ll_sai_entry_cleanup(entry, 1);
}
- dput(sai->sai_first);
- OBD_FREE_PTR(sai);
iput(inode);
+ OBD_FREE_PTR(sai);
}
EXIT;
}
ENTRY;
if (!list_empty(&sai->sai_entries_sent)) {
- list_for_each_entry(entry, &sai->sai_entries_sent,
- se_list) {
+ list_for_each_entry(entry, &sai->sai_entries_sent, se_list) {
if (entry->se_index == index) {
entry->se_stat = stat;
entry->se_req = ptlrpc_request_addref(req);
struct ll_sai_entry *se;
ENTRY;
+ ll_sai_entry_cleanup(entry, 0);
+
spin_lock(&lli->lli_lock);
if (!list_empty(&entry->se_list))
list_del_init(&entry->se_list);
if (unlikely(entry->se_index < sai->sai_index_next)) {
spin_unlock(&lli->lli_lock);
- ll_sai_entry_cleanup(entry);
+ OBD_FREE_PTR(entry);
RETURN(0);
}
struct ll_sai_entry *entry;
struct ptlrpc_request *req;
struct md_enqueue_info *minfo;
- struct dentry *dentry;
struct lookup_intent *it;
+ struct dentry *dentry;
int rc = 0;
+ struct mdt_body *body;
ENTRY;
spin_lock(&lli->lli_lock);
spin_unlock(&lli->lli_lock);
if (unlikely(entry->se_index < sai->sai_index_next)) {
- ll_sai_entry_cleanup(entry);
+ CWARN("Found stale entry: [index %u] [next %u]\n",
+ entry->se_index, sai->sai_index_next);
+ ll_sai_entry_cleanup(entry, 1);
RETURN(0);
}
+ if (entry->se_stat != SA_ENTRY_STATED)
+ GOTO(out, rc = entry->se_stat);
+
req = entry->se_req;
minfo = entry->se_minfo;
- dentry = minfo->mi_dentry;
it = &minfo->mi_it;
+ dentry = minfo->mi_dentry;
- if (entry->se_stat != SA_ENTRY_STATED)
- GOTO(out, rc = entry->se_stat);
+ body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+ if (body == NULL)
+ GOTO(out, rc = -EFAULT);
if (dentry->d_inode == NULL) {
/*
*/
struct dentry *save = dentry;
struct it_cb_data icbd = {
- .icbd_parent = dentry->d_parent->d_inode,
+ .icbd_parent = minfo->mi_dir,
.icbd_childp = &dentry
};
LASSERT(fid_is_zero(&minfo->mi_data.op_fid2));
+ /*
+ * XXX: No fid in reply, this is probaly cross-ref case.
+ * SA can't handle it yet.
+ */
+ if (body->valid & OBD_MD_MDS)
+ GOTO(out, rc = -EAGAIN);
+
rc = ll_lookup_it_finish(req, it, &icbd);
if (!rc)
/*
/*
* revalidate.
*/
- struct mdt_body *body;
-
- body = lustre_msg_buf(req->rq_repmsg, DLM_REPLY_REC_OFF,
- sizeof(*body));
if (!lu_fid_eq(&minfo->mi_data.op_fid2, &body->fid1)) {
ll_unhash_aliases(dentry->d_inode);
GOTO(out, rc = -EAGAIN);
GOTO(out, rc);
}
+ spin_lock(&ll_lookup_lock);
spin_lock(&dcache_lock);
lock_dentry(dentry);
__d_drop(dentry);
unlock_dentry(dentry);
d_rehash_cond(dentry, 0);
spin_unlock(&dcache_lock);
+ spin_unlock(&ll_lookup_lock);
ll_lookup_finish_locks(it, dentry);
}
EXIT;
out:
- if (likely(ll_sai_entry_to_stated(sai, entry))) {
- entry->se_minfo = NULL;
- entry->se_req = NULL;
+ if (likely(ll_sai_entry_to_stated(sai, entry)))
cfs_waitq_signal(&sai->sai_waitq);
- ll_intent_release(it);
- OBD_FREE_PTR(minfo);
- dput(dentry);
- ptlrpc_req_finished(req);
- }
return rc;
}
struct md_enqueue_info *minfo,
int rc)
{
- struct dentry *dentry = minfo->mi_dentry;
struct lookup_intent *it = &minfo->mi_it;
- struct inode *dir = dentry->d_parent->d_inode;
+ struct dentry *dentry = minfo->mi_dentry;
+ struct inode *dir = minfo->mi_dir;
struct ll_inode_info *lli = ll_i2info(dir);
struct ll_statahead_info *sai;
struct ll_sai_entry *entry;
spin_unlock(&lli->lli_lock);
ll_intent_release(it);
dput(dentry);
+ iput(dir);
OBD_FREE_PTR(minfo);
RETURN(-ESTALE);
} else {
- sai = lli->lli_sai;
- if (rc || dir == NULL)
- rc = -ESTALE;
-
+ sai = ll_sai_get(lli->lli_sai);
entry = ll_sai_entry_set(sai,
(unsigned int)(long)minfo->mi_cbdata,
rc ? SA_ENTRY_UNSTATED :
list_del_init(&entry->se_list);
sai->sai_replied++;
spin_unlock(&lli->lli_lock);
- ll_sai_entry_cleanup(entry);
+ ll_sai_entry_cleanup(entry, 1);
}
+ ll_sai_put(sai);
RETURN(rc);
}
}
struct ldlm_enqueue_info *einfo)
{
LASSERT(minfo && einfo);
+ iput(minfo->mi_dir);
capa_put(minfo->mi_data.op_capa1);
capa_put(minfo->mi_data.op_capa2);
OBD_FREE_PTR(minfo);
minfo->mi_it.it_op = IT_GETATTR;
minfo->mi_dentry = dentry;
+ minfo->mi_dir = igrab(dir);
minfo->mi_cb = ll_statahead_interpret;
minfo->mi_generation = lli->lli_sai->sai_generation;
minfo->mi_cbdata = (void *)(long)lli->lli_sai->sai_index;
* \retval 0 -- will send stat-ahead request
* \retval others -- prepare stat-ahead request failed
*/
-static int do_sa_revalidate(struct dentry *dentry)
+static int do_sa_revalidate(struct inode *dir, struct dentry *dentry)
{
struct inode *inode = dentry->d_inode;
- struct inode *dir = dentry->d_parent->d_inode;
struct lookup_intent it = { .it_op = IT_GETATTR };
struct md_enqueue_info *minfo;
struct ldlm_enqueue_info *einfo;
RETURN(rc);
}
-static inline void ll_name2qstr(struct qstr *this, const char *name, int namelen)
+static inline void ll_name2qstr(struct qstr *q, const char *name, int namelen)
{
- unsigned long hash = init_name_hash();
- unsigned int c;
-
- this->name = name;
- this->len = namelen;
- for (; namelen > 0; namelen--, name++) {
- c = *(const unsigned char *)name;
- hash = partial_name_hash(c, hash);
- }
- this->hash = end_name_hash(hash);
+ q->name = name;
+ q->len = namelen;
+ q->hash = full_name_hash(name, namelen);
}
static int ll_statahead_one(struct dentry *parent, const char* entry_name,
GOTO(out, rc = -ENOMEM);
}
} else {
- rc = do_sa_revalidate(dentry);
+ rc = do_sa_revalidate(dir, dentry);
if (rc)
dput(dentry);
}
return rc;
}
-struct ll_sa_thread_args {
- struct dentry *sta_parent;
- pid_t sta_pid;
-};
-
static int ll_statahead_thread(void *arg)
{
- struct ll_sa_thread_args *sta = arg;
- struct dentry *parent = dget(sta->sta_parent);
+ struct dentry *parent = (struct dentry *)arg;
struct inode *dir = parent->d_inode;
struct ll_inode_info *lli = ll_i2info(dir);
struct ll_sb_info *sbi = ll_i2sbi(dir);
{
char pname[16];
- snprintf(pname, 15, "ll_sa_%u", sta->sta_pid);
+ snprintf(pname, 15, "ll_sa_%u", lli->lli_opendir_pid);
cfs_daemonize(pname);
}
if (IS_ERR(page)) {
rc = PTR_ERR(page);
- CERROR("error reading dir "DFID" at %llu/%u: rc %d\n",
+ CERROR("error reading dir "DFID" at "LPU64"/%u: rc %d\n",
PFID(ll_inode2fid(dir)), pos,
sai->sai_index, rc);
break;
void ll_stop_statahead(struct inode *inode, void *key)
{
struct ll_inode_info *lli = ll_i2info(inode);
- struct ptlrpc_thread *thread;
+
+ if (unlikely(key == NULL))
+ return;
spin_lock(&lli->lli_lock);
- if (lli->lli_opendir_pid == 0 ||
- unlikely(lli->lli_opendir_key != key)) {
+ if (lli->lli_opendir_key != key || lli->lli_opendir_pid == 0) {
spin_unlock(&lli->lli_lock);
return;
}
lli->lli_opendir_key = NULL;
- lli->lli_opendir_pid = 0;
if (lli->lli_sai) {
struct l_wait_info lwi = { 0 };
+ struct ptlrpc_thread *thread = &lli->lli_sai->sai_thread;
- thread = &lli->lli_sai->sai_thread;
if (!sa_is_stopped(lli->lli_sai)) {
thread->t_flags = SVC_STOPPING;
spin_unlock(&lli->lli_lock);
* maybe inflight.
*/
ll_sai_put(lli->lli_sai);
- return;
+ } else {
+ lli->lli_opendir_pid = 0;
+ spin_unlock(&lli->lli_lock);
}
- spin_unlock(&lli->lli_lock);
}
enum {
if (IS_ERR(page)) {
rc = PTR_ERR(page);
- CERROR("error reading dir "DFID" at %llu: rc %d\n",
+ CERROR("error reading dir "DFID" at "LPU64": rc %d\n",
PFID(ll_inode2fid(dir)), pos, rc);
break;
}
}
if (target->len == namelen &&
- !strncmp(target->name, name, target->len))
+ memcmp(target->name, name, namelen) == 0)
rc = LS_FIRST_DE + dot_de;
else
rc = LS_NONE_FIRST_DE;
* \retval 1 -- stat ahead thread process such dentry, for lookup, it hit
* \retval -EEXIST -- stat ahead thread started, and this is the first dentry
* \retval -EBADFD -- statahead thread exit and not dentry available
+ * \retval -EAGAIN -- try to stat by caller
* \retval others -- error
*/
int do_statahead_enter(struct inode *dir, struct dentry **dentryp, int lookup)
{
- struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct ll_inode_info *lli = ll_i2info(dir);
- struct ll_statahead_info *sai = lli->lli_sai;
- struct ll_sa_thread_args sta;
- struct l_wait_info lwi = { 0 };
- int rc;
+ struct ll_inode_info *lli;
+ struct ll_statahead_info *sai;
+ struct dentry *parent;
+ struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
+ int rc = 0;
ENTRY;
+ LASSERT(dir != NULL);
+ lli = ll_i2info(dir);
LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
+ sai = lli->lli_sai;
if (sai) {
+ struct ll_sb_info *sbi;
+
if (unlikely(sa_is_stopped(sai) &&
list_empty(&sai->sai_entries_stated)))
RETURN(-EBADFD);
- /*
- * skip the first dentry.
- */
- if (unlikely((*dentryp)->d_name.len ==
- sai->sai_first->d_name.len &&
- !strncmp((*dentryp)->d_name.name,
- sai->sai_first->d_name.name,
- sai->sai_first->d_name.len)))
- RETURN(-EEXIST);
-
if ((*dentryp)->d_name.name[0] == '.') {
if (likely(sai->sai_ls_all ||
sai->sai_miss_hidden >= sai->sai_skip_hidden)) {
}
}
+ sbi = ll_i2sbi(dir);
if (ll_sai_entry_stated(sai)) {
sbi->ll_sa_cached++;
} else {
/*
* thread started already, avoid double-stat.
*/
- l_wait_event(sai->sai_waitq,
- ll_sai_entry_stated(sai) || sa_is_stopped(sai),
- &lwi);
+ rc = l_wait_event(sai->sai_waitq,
+ ll_sai_entry_stated(sai) ||
+ sa_is_stopped(sai),
+ &lwi);
}
if (lookup) {
&(*dentryp)->d_name);
if (result) {
LASSERT(result != *dentryp);
- dput(*dentryp);
+ /* BUG 16303: do not drop reference count for
+ * "*dentryp", VFS will do that by itself. */
*dentryp = result;
RETURN(1);
}
/*
* do nothing for revalidate.
*/
- RETURN(0);
+ RETURN(rc);
}
- /*
- * I am the "lli_opendir_pid" owner, only me can set "lli_sai".
- */
- LASSERT(lli->lli_sai == NULL);
-
+ /* I am the "lli_opendir_pid" owner, only me can set "lli_sai". */
rc = is_first_dirent(dir, *dentryp);
- if (rc == LS_NONE_FIRST_DE) {
- /*
- * It is not "ls -{a}l" operation, no need statahead for it.
- */
- spin_lock(&lli->lli_lock);
- lli->lli_opendir_key = NULL;
- lli->lli_opendir_pid = 0;
- spin_unlock(&lli->lli_lock);
- RETURN(-EBADF);
- }
+ if (rc == LS_NONE_FIRST_DE)
+ /* It is not "ls -{a}l" operation, no need statahead for it. */
+ GOTO(out, rc = -EAGAIN);
sai = ll_sai_alloc();
if (sai == NULL)
- RETURN(-ENOMEM);
+ GOTO(out, rc = -ENOMEM);
- sai->sai_inode = igrab(dir);
- sai->sai_first = dget(*dentryp);
sai->sai_ls_all = (rc == LS_FIRST_DOT_DE);
+ sai->sai_inode = igrab(dir);
+ if (unlikely(sai->sai_inode == NULL)) {
+ CWARN("Do not start stat ahead on dying inode "DFID" .\n",
+ PFID(&lli->lli_fid));
+ OBD_FREE_PTR(sai);
+ GOTO(out, rc = -ESTALE);
+ }
- sta.sta_parent = (*dentryp)->d_parent;
- sta.sta_pid = cfs_curproc_pid();
+ /* get parent reference count here, and put it in ll_statahead_thread */
+ parent = dget((*dentryp)->d_parent);
+ if (unlikely(sai->sai_inode != parent->d_inode)) {
+ struct ll_inode_info *nlli = ll_i2info(parent->d_inode);
+
+ CWARN("Race condition, someone changed %.*s just now: "
+ "old parent "DFID", new parent "DFID" .\n",
+ (*dentryp)->d_name.len, (*dentryp)->d_name.name,
+ PFID(&lli->lli_fid), PFID(&nlli->lli_fid));
+ dput(parent);
+ iput(sai->sai_inode);
+ OBD_FREE_PTR(sai);
+ RETURN(-EAGAIN);
+ }
lli->lli_sai = sai;
- rc = cfs_kernel_thread(ll_statahead_thread, &sta, 0);
+ rc = cfs_kernel_thread(ll_statahead_thread, parent, 0);
if (rc < 0) {
CERROR("can't start ll_sa thread, rc: %d\n", rc);
+ dput(parent);
+ lli->lli_opendir_key = NULL;
sai->sai_thread.t_flags = SVC_STOPPED;
ll_sai_put(sai);
LASSERT(lli->lli_sai == NULL);
- RETURN(rc);
+ RETURN(-EAGAIN);
}
l_wait_event(sai->sai_thread.t_ctl_waitq,
* lookup, and -EEXIST also indicates that this is the first dirent.
*/
RETURN(-EEXIST);
+
+out:
+ spin_lock(&lli->lli_lock);
+ lli->lli_opendir_key = NULL;
+ lli->lli_opendir_pid = 0;
+ spin_unlock(&lli->lli_lock);
+ return rc;
}
/**
* update hit/miss count.
*/
-int ll_statahead_exit(struct dentry *dentry, int result)
+void ll_statahead_exit(struct inode *dir, struct dentry *dentry, int result)
{
- struct dentry *parent = dentry->d_parent;
- struct ll_inode_info *lli = ll_i2info(parent->d_inode);
- struct ll_sb_info *sbi = ll_i2sbi(parent->d_inode);
- struct ll_dentry_data *ldd = ll_d2d(dentry);
+ struct ll_inode_info *lli;
+ struct ll_statahead_info *sai;
+ struct ll_sb_info *sbi;
+ struct ll_dentry_data *ldd = ll_d2d(dentry);
ENTRY;
- if (lli->lli_opendir_pid != cfs_curproc_pid())
- RETURN(-EBADFD);
-
- if (lli->lli_sai) {
- struct ll_statahead_info *sai = lli->lli_sai;
-
- if (result >= 1) {
- sbi->ll_sa_hit++;
- sai->sai_hit++;
- sai->sai_consecutive_miss = 0;
- sai->sai_max = min(2 * sai->sai_max, sbi->ll_sa_max);
- } else {
- sbi->ll_sa_miss++;
- sai->sai_miss++;
- sai->sai_consecutive_miss++;
- if (sa_low_hit(sai) && sa_is_running(sai)) {
- sbi->ll_sa_wrong++;
- CDEBUG(D_READA, "statahead for dir %.*s hit "
- "ratio too low: hit/miss %u/%u, "
- "sent/replied %u/%u. stopping statahead "
- "thread: pid %d\n",
- parent->d_name.len, parent->d_name.name,
- sai->sai_hit, sai->sai_miss,
- sai->sai_sent, sai->sai_replied,
- cfs_curproc_pid());
- spin_lock(&lli->lli_lock);
- if (!sa_is_stopped(sai))
- sai->sai_thread.t_flags = SVC_STOPPING;
- spin_unlock(&lli->lli_lock);
- }
+ LASSERT(dir != NULL);
+ lli = ll_i2info(dir);
+ LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
+ sai = lli->lli_sai;
+ LASSERT(sai != NULL);
+ sbi = ll_i2sbi(dir);
+
+ if (result >= 1) {
+ sbi->ll_sa_hit++;
+ sai->sai_hit++;
+ sai->sai_consecutive_miss = 0;
+ sai->sai_max = min(2 * sai->sai_max, sbi->ll_sa_max);
+ } else {
+ sbi->ll_sa_miss++;
+ sai->sai_miss++;
+ sai->sai_consecutive_miss++;
+ if (sa_low_hit(sai) && sa_is_running(sai)) {
+ sbi->ll_sa_wrong++;
+ CDEBUG(D_READA, "Statahead for dir "DFID" hit ratio "
+ "too low: hit/miss %u/%u, sent/replied %u/%u, "
+ "stopping statahead thread: pid %d\n",
+ PFID(&lli->lli_fid), sai->sai_hit,
+ sai->sai_miss, sai->sai_sent,
+ sai->sai_replied, cfs_curproc_pid());
+ spin_lock(&lli->lli_lock);
+ if (!sa_is_stopped(sai))
+ sai->sai_thread.t_flags = SVC_STOPPING;
+ spin_unlock(&lli->lli_lock);
}
+ }
- if (!sa_is_stopped(sai))
- cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
- ll_sai_entry_fini(sai);
-
- if (unlikely(ldd == NULL)) {
- ll_set_dd(dentry);
- ldd = ll_d2d(dentry);
- if (ldd != NULL && dentry->d_op == NULL) {
- lock_dentry(dentry);
- dentry->d_op = dentry->d_op ? : &ll_sai_d_ops;
- unlock_dentry(dentry);
- }
- }
+ if (!sa_is_stopped(sai))
+ cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
+ ll_sai_entry_fini(sai);
+ if (likely(ldd != NULL))
+ ldd->lld_sa_generation = sai->sai_generation;
- if (likely(ldd != NULL))
- ldd->lld_sa_generation = sai->sai_generation;
- else
- RETURN(-ENOMEM);
- }
- RETURN(0);
+ EXIT;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff