*
* You should have received a copy of the GNU General Public License
* 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
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <linux/fs.h>
#include <linux/sched.h>
+#include <linux/kthread.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>
+#include <linux/delay.h>
#define DEBUG_SUBSYSTEM S_LLITE
SA_ENTRY_INVA = 2, /** invalid entry */
} se_state_t;
-/* sa_entry is not refcounted: statahead thread allocates it and do async stat,
+/*
+ * sa_entry is not refcounted: statahead thread allocates it and do async stat,
* and in async stat callback ll_statahead_interpret() will add it into
- * sai_cb_entries, later statahead thread will call sa_handle_callback() to
+ * sai_interim_entries, later statahead thread will call sa_handle_callback() to
* instantiate entry and move it into sai_entries, and then only scanner process
- * can access and free it. */
+ * can access and free it.
+ */
struct sa_entry {
- /* link into sai_cb_entries or sai_entries */
+ /* link into sai_interim_entries or sai_entries */
struct list_head se_list;
/* link into sai hash table locally */
struct list_head se_hash;
struct inode *se_inode;
/* entry name */
struct qstr se_qstr;
+ /* entry fid */
+ struct lu_fid se_fid;
};
-static unsigned int sai_generation = 0;
+static unsigned int sai_generation;
static DEFINE_SPINLOCK(sai_generation_lock);
static inline int sa_unhashed(struct sa_entry *entry)
return list_empty(&entry->se_hash);
}
-/*
- * The entry only can be released by the caller, it is necessary to hold lock.
- */
+/* sa_entry is ready to use */
static inline int sa_ready(struct sa_entry *entry)
{
+ /* Make sure sa_entry is updated and ready to use */
smp_rmb();
return (entry->se_state != SA_ENTRY_INIT);
}
+/* hash value to put in sai_cache */
static inline int sa_hash(int val)
{
return val & LL_SA_CACHE_MASK;
}
-/*
- * Insert entry to hash SA table.
- */
+/* hash entry into sai_cache */
static inline void
sa_rehash(struct ll_statahead_info *sai, struct sa_entry *entry)
{
spin_unlock(&sai->sai_cache_lock[i]);
}
-/*
- * Remove entry from SA table.
- */
+/* unhash entry from sai_cache */
static inline void
sa_unhash(struct ll_statahead_info *sai, struct sa_entry *entry)
{
}
static inline int agl_should_run(struct ll_statahead_info *sai,
- struct inode *inode)
+ struct inode *inode)
{
- return (inode != NULL && S_ISREG(inode->i_mode) && sai->sai_agl_valid);
+ return inode && S_ISREG(inode->i_mode) && sai->sai_agl_task;
}
static inline struct ll_inode_info *
lli_agl_list);
}
+/* statahead window is full */
static inline int sa_sent_full(struct ll_statahead_info *sai)
{
return atomic_read(&sai->sai_cache_count) >= sai->sai_max;
}
+/* got async stat replies */
static inline int sa_has_callback(struct ll_statahead_info *sai)
{
return !list_empty(&sai->sai_interim_entries);
*/
static inline int sa_low_hit(struct ll_statahead_info *sai)
{
- return ((sai->sai_hit > 7 && sai->sai_hit < 4 * sai->sai_miss) ||
- (sai->sai_consecutive_miss > 8));
+ return ((sai->sai_hit > 7 && sai->sai_hit < 4 * sai->sai_miss) ||
+ (sai->sai_consecutive_miss > 8));
}
/*
- * If the given index is behind of statahead window more than
+ * if the given index is behind of statahead window more than
* SA_OMITTED_ENTRY_MAX, then it is old.
*/
static inline int is_omitted_entry(struct ll_statahead_info *sai, __u64 index)
{
- return ((__u64)sai->sai_max + index + SA_OMITTED_ENTRY_MAX <
- sai->sai_index);
+ return ((__u64)sai->sai_max + index + SA_OMITTED_ENTRY_MAX <
+ sai->sai_index);
}
-/* allocate sa_entry and add it into hash to let scanner process to find it */
+/* allocate sa_entry and hash it to allow scanner process to find it */
static struct sa_entry *
-sa_alloc(struct ll_statahead_info *sai, __u64 index, const char *name, int len)
+sa_alloc(struct dentry *parent, struct ll_statahead_info *sai, __u64 index,
+ const char *name, int len, const struct lu_fid *fid)
{
struct ll_inode_info *lli;
struct sa_entry *entry;
int entry_size;
char *dname;
+
ENTRY;
entry_size = sizeof(struct sa_entry) + (len & ~3) + 4;
OBD_ALLOC(entry, entry_size);
- if (unlikely(entry == NULL))
+ if (unlikely(!entry))
RETURN(ERR_PTR(-ENOMEM));
- CDEBUG(D_READA, "alloc sa entry %.*s(%p) index "LPU64"\n",
+ CDEBUG(D_READA, "alloc sa entry %.*s(%p) index %llu\n",
len, name, entry, index);
entry->se_index = index;
dname = (char *)entry + sizeof(struct sa_entry);
memcpy(dname, name, len);
dname[len] = 0;
- entry->se_qstr.hash = full_name_hash(name, len);
+ entry->se_qstr.hash = ll_full_name_hash(parent, name, len);
entry->se_qstr.len = len;
entry->se_qstr.name = dname;
+ entry->se_fid = *fid;
+
+ lli = ll_i2info(sai->sai_dentry->d_inode);
- lli = ll_i2info(sai->sai_inode);
spin_lock(&lli->lli_sa_lock);
INIT_LIST_HEAD(&entry->se_list);
sa_rehash(sai, entry);
RETURN(entry);
}
-/* free sa_entry which should have been unhashed and not in any list */
+/* free sa_entry, which should have been unhashed and not in any list */
static void sa_free(struct ll_statahead_info *sai, struct sa_entry *entry)
{
- CDEBUG(D_READA, "free sa entry %.*s(%p) index "LPU64"\n",
+ CDEBUG(D_READA, "free sa entry %.*s(%p) index %llu\n",
entry->se_qstr.len, entry->se_qstr.name, entry,
entry->se_index);
atomic_dec(&sai->sai_cache_count);
}
-/* find sa_entry by name, used by directory scanner, lock is not needed because
- * only scanner can remove the entry from hash.
+/*
+ * find sa_entry by name, used by directory scanner, lock is not needed because
+ * only scanner can remove the entry from cache.
*/
static struct sa_entry *
sa_get(struct ll_statahead_info *sai, const struct qstr *qstr)
return NULL;
}
+/* unhash and unlink sa_entry, and then free it */
static inline void
sa_kill(struct ll_statahead_info *sai, struct sa_entry *entry)
{
- struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
+ struct ll_inode_info *lli = ll_i2info(sai->sai_dentry->d_inode);
LASSERT(!sa_unhashed(entry));
LASSERT(!list_empty(&entry->se_list));
list_del_init(&entry->se_list);
spin_unlock(&lli->lli_sa_lock);
- if (entry->se_inode != NULL)
+ if (entry->se_inode)
iput(entry->se_inode);
sa_free(sai, entry);
{
struct sa_entry *tmp, *next;
- if (entry != NULL && entry->se_state == SA_ENTRY_SUCC) {
- struct ll_sb_info *sbi = ll_i2sbi(sai->sai_inode);
+ if (entry && entry->se_state == SA_ENTRY_SUCC) {
+ struct ll_sb_info *sbi = ll_i2sbi(sai->sai_dentry->d_inode);
sai->sai_hit++;
sai->sai_consecutive_miss = 0;
sai->sai_consecutive_miss++;
}
- if (entry != NULL)
+ if (entry)
sa_kill(sai, entry);
- /* kill old completed entries, only scanner process does this, no need
- * to lock */
+ /*
+ * kill old completed entries, only scanner process does this, no need
+ * to lock
+ */
list_for_each_entry_safe(tmp, next, &sai->sai_entries, se_list) {
if (!is_omitted_entry(sai, tmp->se_index))
break;
sa_kill(sai, tmp);
}
-
- wake_up(&sai->sai_thread.t_ctl_waitq);
}
-/* update state and sort add entry to sai_entries by index, return true if
- * scanner is waiting on this entry. */
+/*
+ * update state and sort add entry to sai_entries by index, return true if
+ * scanner is waiting on this entry.
+ */
static bool
__sa_make_ready(struct ll_statahead_info *sai, struct sa_entry *entry, int ret)
{
struct sa_entry *se;
struct list_head *pos = &sai->sai_entries;
+ __u64 index = entry->se_index;
LASSERT(!sa_ready(entry));
LASSERT(list_empty(&entry->se_list));
- entry->se_state = ret < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC;
-
list_for_each_entry_reverse(se, &sai->sai_entries, se_list) {
if (se->se_index < entry->se_index) {
pos = &se->se_list;
}
}
list_add(&entry->se_list, pos);
+ /*
+ * LU-9210: ll_statahead_interpet must be able to see this before
+ * we wake it up
+ */
+ smp_store_release(&entry->se_state,
+ ret < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
+
+ return (index == sai->sai_index_wait);
+}
- return (entry->se_index == sai->sai_index_wait);
+/* finish async stat RPC arguments */
+static void sa_fini_data(struct md_enqueue_info *minfo)
+{
+ ll_unlock_md_op_lsm(&minfo->mi_data);
+ iput(minfo->mi_dir);
+ OBD_FREE_PTR(minfo);
}
-/* release resources used in async stat RPC, complete entry information and
- * wakeup if necessary */
+static int ll_statahead_interpret(struct ptlrpc_request *req,
+ struct md_enqueue_info *minfo, int rc);
+
+/*
+ * prepare arguments for async stat RPC.
+ */
+static struct md_enqueue_info *
+sa_prep_data(struct inode *dir, struct inode *child, struct sa_entry *entry)
+{
+ struct md_enqueue_info *minfo;
+ struct ldlm_enqueue_info *einfo;
+ struct md_op_data *op_data;
+
+ OBD_ALLOC_PTR(minfo);
+ if (!minfo)
+ return ERR_PTR(-ENOMEM);
+
+ op_data = ll_prep_md_op_data(&minfo->mi_data, dir, child,
+ entry->se_qstr.name, entry->se_qstr.len, 0,
+ LUSTRE_OPC_ANY, NULL);
+ if (IS_ERR(op_data)) {
+ OBD_FREE_PTR(minfo);
+ return (struct md_enqueue_info *)op_data;
+ }
+
+ if (!child)
+ op_data->op_fid2 = entry->se_fid;
+
+ minfo->mi_it.it_op = IT_GETATTR;
+ minfo->mi_dir = igrab(dir);
+ minfo->mi_cb = ll_statahead_interpret;
+ minfo->mi_cbdata = entry;
+
+ einfo = &minfo->mi_einfo;
+ einfo->ei_type = LDLM_IBITS;
+ einfo->ei_mode = it_to_lock_mode(&minfo->mi_it);
+ einfo->ei_cb_bl = ll_md_blocking_ast;
+ einfo->ei_cb_cp = ldlm_completion_ast;
+ einfo->ei_cb_gl = NULL;
+ einfo->ei_cbdata = NULL;
+
+ return minfo;
+}
+
+/*
+ * release resources used in async stat RPC, update entry state and wakeup if
+ * scanner process it waiting on this entry.
+ */
static void
sa_make_ready(struct ll_statahead_info *sai, struct sa_entry *entry, int ret)
{
- struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
+ struct ll_inode_info *lli = ll_i2info(sai->sai_dentry->d_inode);
struct md_enqueue_info *minfo = entry->se_minfo;
struct ptlrpc_request *req = entry->se_req;
bool wakeup;
if (minfo) {
entry->se_minfo = NULL;
ll_intent_release(&minfo->mi_it);
- iput(minfo->mi_dir);
- OBD_FREE_PTR(minfo);
+ sa_fini_data(minfo);
}
if (req) {
wake_up(&sai->sai_waitq);
}
-/*
- * Insert inode into the list of sai_agls.
- */
+/* insert inode into the list of sai_agls */
static void ll_agl_add(struct ll_statahead_info *sai,
- struct inode *inode, int index)
+ 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;
+ struct ll_inode_info *parent = ll_i2info(sai->sai_dentry->d_inode);
spin_lock(&child->lli_agl_lock);
if (child->lli_agl_index == 0) {
LASSERT(list_empty(&child->lli_agl_list));
- igrab(inode);
spin_lock(&parent->lli_agl_lock);
- if (agl_list_empty(sai))
- added = 1;
- list_add_tail(&child->lli_agl_list, &sai->sai_agls);
+ /* Re-check under the lock */
+ if (agl_should_run(sai, inode)) {
+ if (agl_list_empty(sai))
+ wake_up_process(sai->sai_agl_task);
+ igrab(inode);
+ list_add_tail(&child->lli_agl_list, &sai->sai_agls);
+ } else
+ child->lli_agl_index = 0;
spin_unlock(&parent->lli_agl_lock);
} else {
spin_unlock(&child->lli_agl_lock);
}
-
- if (added > 0)
- wake_up(&sai->sai_agl_thread.t_ctl_waitq);
}
-static struct ll_statahead_info *ll_sai_alloc(void)
+/* allocate sai */
+static struct ll_statahead_info *ll_sai_alloc(struct dentry *dentry)
{
struct ll_statahead_info *sai;
- int i;
+ struct ll_inode_info *lli = ll_i2info(dentry->d_inode);
+ int i;
+
ENTRY;
OBD_ALLOC_PTR(sai);
if (!sai)
RETURN(NULL);
+ sai->sai_dentry = dget(dentry);
atomic_set(&sai->sai_refcount, 1);
-
- 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;
init_waitqueue_head(&sai->sai_waitq);
- init_waitqueue_head(&sai->sai_thread.t_ctl_waitq);
- init_waitqueue_head(&sai->sai_agl_thread.t_ctl_waitq);
INIT_LIST_HEAD(&sai->sai_interim_entries);
INIT_LIST_HEAD(&sai->sai_entries);
}
atomic_set(&sai->sai_cache_count, 0);
+ spin_lock(&sai_generation_lock);
+ lli->lli_sa_generation = ++sai_generation;
+ if (unlikely(sai_generation == 0))
+ lli->lli_sa_generation = ++sai_generation;
+ spin_unlock(&sai_generation_lock);
+
RETURN(sai);
}
+/* free sai */
+static inline void ll_sai_free(struct ll_statahead_info *sai)
+{
+ LASSERT(sai->sai_dentry != NULL);
+ dput(sai->sai_dentry);
+ OBD_FREE_PTR(sai);
+}
+
+/*
+ * take refcount of sai if sai for @dir exists, which means statahead is on for
+ * this directory.
+ */
static inline struct ll_statahead_info *ll_sai_get(struct inode *dir)
{
struct ll_inode_info *lli = ll_i2info(dir);
spin_lock(&lli->lli_sa_lock);
sai = lli->lli_sai;
- if (sai != NULL)
+ if (sai)
atomic_inc(&sai->sai_refcount);
spin_unlock(&lli->lli_sa_lock);
return sai;
}
+/*
+ * put sai refcount after use, if refcount reaches zero, free sai and sa_entries
+ * attached to it.
+ */
static void ll_sai_put(struct ll_statahead_info *sai)
{
- struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
+ struct ll_inode_info *lli = ll_i2info(sai->sai_dentry->d_inode);
if (atomic_dec_and_lock(&sai->sai_refcount, &lli->lli_sa_lock)) {
struct sa_entry *entry, *next;
- struct ll_sb_info *sbi = ll_i2sbi(sai->sai_inode);
+ struct ll_sb_info *sbi = ll_i2sbi(sai->sai_dentry->d_inode);
lli->lli_sai = NULL;
spin_unlock(&lli->lli_sa_lock);
- LASSERT(thread_is_stopped(&sai->sai_thread));
- LASSERT(thread_is_stopped(&sai->sai_agl_thread));
+ LASSERT(!sai->sai_task);
+ LASSERT(!sai->sai_agl_task);
LASSERT(sai->sai_sent == sai->sai_replied);
LASSERT(!sa_has_callback(sai));
LASSERT(atomic_read(&sai->sai_cache_count) == 0);
LASSERT(agl_list_empty(sai));
- iput(sai->sai_inode);
- OBD_FREE_PTR(sai);
+ ll_sai_free(sai);
atomic_dec(&sbi->ll_sa_running);
}
}
static void ll_agl_trigger(struct inode *inode, struct ll_statahead_info *sai)
{
struct ll_inode_info *lli = ll_i2info(inode);
- __u64 index = lli->lli_agl_index;
+ u64 index = lli->lli_agl_index;
+ ktime_t expire;
int rc;
+
ENTRY;
LASSERT(list_empty(&lli->lli_agl_list));
- /* AGL maybe fall behind statahead with one entry */
- if (is_omitted_entry(sai, index + 1)) {
- lli->lli_agl_index = 0;
- iput(inode);
- RETURN_EXIT;
- }
+ /* AGL maybe fall behind statahead with one entry */
+ if (is_omitted_entry(sai, index + 1)) {
+ lli->lli_agl_index = 0;
+ iput(inode);
+ RETURN_EXIT;
+ }
+
+ /*
+ * In case of restore, the MDT has the right size and has already
+ * sent it back without granting the layout lock, inode is up-to-date.
+ * Then AGL (async glimpse lock) is useless.
+ * Also to glimpse we need the layout, in case of a runninh restore
+ * the MDT holds the layout lock so the glimpse will block up to the
+ * end of restore (statahead/agl will block)
+ */
+ if (ll_file_test_flag(lli, LLIF_FILE_RESTORING)) {
+ lli->lli_agl_index = 0;
+ iput(inode);
+ RETURN_EXIT;
+ }
- /* Someone is in glimpse (sync or async), do nothing. */
+ /* Someone is in glimpse (sync or async), do nothing. */
rc = down_write_trylock(&lli->lli_glimpse_sem);
- if (rc == 0) {
- lli->lli_agl_index = 0;
- iput(inode);
- RETURN_EXIT;
- }
-
- /*
- * Someone triggered glimpse within 1 sec before.
- * 1) The former glimpse succeeded with glimpse lock granted by OST, and
- * if the lock is still cached on client, AGL needs to do nothing. If
- * it is cancelled by other client, AGL maybe cannot obtaion new lock
- * for no glimpse callback triggered by AGL.
- * 2) The former glimpse succeeded, but OST did not grant glimpse lock.
- * Under such case, it is quite possible that the OST will not grant
- * glimpse lock for AGL also.
- * 3) The former glimpse failed, compared with other two cases, it is
- * relative rare. AGL can ignore such case, and it will not muchly
- * affect the performance.
- */
- if (lli->lli_glimpse_time != 0 &&
- cfs_time_before(cfs_time_shift(-1), lli->lli_glimpse_time)) {
+ if (rc == 0) {
+ lli->lli_agl_index = 0;
+ iput(inode);
+ RETURN_EXIT;
+ }
+
+ /*
+ * Someone triggered glimpse within 1 sec before.
+ * 1) The former glimpse succeeded with glimpse lock granted by OST, and
+ * if the lock is still cached on client, AGL needs to do nothing. If
+ * it is cancelled by other client, AGL maybe cannot obtaion new lock
+ * for no glimpse callback triggered by AGL.
+ * 2) The former glimpse succeeded, but OST did not grant glimpse lock.
+ * Under such case, it is quite possible that the OST will not grant
+ * glimpse lock for AGL also.
+ * 3) The former glimpse failed, compared with other two cases, it is
+ * relative rare. AGL can ignore such case, and it will not muchly
+ * affect the performance.
+ */
+ expire = ktime_sub_ns(ktime_get(), NSEC_PER_SEC);
+ if (ktime_to_ns(lli->lli_glimpse_time) &&
+ ktime_before(expire, lli->lli_glimpse_time)) {
up_write(&lli->lli_glimpse_sem);
- lli->lli_agl_index = 0;
- iput(inode);
- RETURN_EXIT;
- }
+ lli->lli_agl_index = 0;
+ iput(inode);
+ RETURN_EXIT;
+ }
- CDEBUG(D_READA, "Handling (init) async glimpse: inode = "
- DFID", idx = "LPU64"\n", PFID(&lli->lli_fid), index);
+ CDEBUG(D_READA,
+ "Handling (init) async glimpse: inode = " DFID", idx = %llu\n",
+ PFID(&lli->lli_fid), index);
- cl_agl(inode);
- lli->lli_agl_index = 0;
- lli->lli_glimpse_time = cfs_time_current();
+ cl_agl(inode);
+ lli->lli_agl_index = 0;
+ lli->lli_glimpse_time = ktime_get();
up_write(&lli->lli_glimpse_sem);
- CDEBUG(D_READA, "Handled (init) async glimpse: inode= "
- DFID", idx = "LPU64", rc = %d\n",
- PFID(&lli->lli_fid), index, rc);
+ CDEBUG(D_READA,
+ "Handled (init) async glimpse: inode= " DFID", idx = %llu, rc = %d\n",
+ PFID(&lli->lli_fid), index, rc);
- iput(inode);
+ iput(inode);
- EXIT;
+ EXIT;
}
-/* prepare inode for sa entry, add it into agl list, now sa_entry is ready
- * to be used by scanner process. */
+/*
+ * prepare inode for sa entry, add it into agl list, now sa_entry is ready
+ * to be used by scanner process.
+ */
static void sa_instantiate(struct ll_statahead_info *sai,
- struct sa_entry *entry)
+ struct sa_entry *entry)
{
- struct inode *dir = sai->sai_inode;
+ struct inode *dir = sai->sai_dentry->d_inode;
struct inode *child;
struct md_enqueue_info *minfo;
struct lookup_intent *it;
struct ptlrpc_request *req;
struct mdt_body *body;
int rc = 0;
+
ENTRY;
- LASSERT(entry->se_handle != 0);
-
- minfo = entry->se_minfo;
- it = &minfo->mi_it;
- req = entry->se_req;
- body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
- if (body == NULL)
- GOTO(out, rc = -EFAULT);
-
- child = entry->se_inode;
- if (child == NULL) {
- /*
- * lookup.
- */
- 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->mbo_valid & OBD_MD_MDS)
- GOTO(out, rc = -EAGAIN);
- } else {
- /*
- * revalidate.
- */
- /* unlinked and re-created with the same name */
- if (unlikely(!lu_fid_eq(&minfo->mi_data.op_fid2,
- &body->mbo_fid1))) {
- entry->se_inode = NULL;
- iput(child);
- child = NULL;
- }
- }
-
- it->d.lustre.it_lock_handle = entry->se_handle;
+ LASSERT(entry->se_handle != 0);
+
+ minfo = entry->se_minfo;
+ it = &minfo->mi_it;
+ req = entry->se_req;
+ body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+ if (!body)
+ GOTO(out, rc = -EFAULT);
+
+ child = entry->se_inode;
+ /* revalidate; unlinked and re-created with the same name */
+ if (unlikely(!lu_fid_eq(&minfo->mi_data.op_fid2, &body->mbo_fid1))) {
+ if (child) {
+ entry->se_inode = NULL;
+ iput(child);
+ }
+ /* The mdt_body is invalid. Skip this entry */
+ GOTO(out, rc = -EAGAIN);
+ }
+
+ it->it_lock_handle = entry->se_handle;
rc = md_revalidate_lock(ll_i2mdexp(dir), it, ll_inode2fid(dir), NULL);
- if (rc != 1)
- GOTO(out, rc = -EAGAIN);
+ if (rc != 1)
+ GOTO(out, rc = -EAGAIN);
- rc = ll_prep_inode(&child, req, dir->i_sb, it);
- if (rc)
- GOTO(out, rc);
+ rc = ll_prep_inode(&child, req, dir->i_sb, it);
+ if (rc)
+ GOTO(out, rc);
CDEBUG(D_READA, "%s: setting %.*s"DFID" l_data to inode %p\n",
- ll_get_fsname(child->i_sb, NULL, 0),
- entry->se_qstr.len, entry->se_qstr.name,
- PFID(ll_inode2fid(child)), child);
- ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, child, it, NULL);
+ ll_i2sbi(dir)->ll_fsname, entry->se_qstr.len,
+ entry->se_qstr.name, PFID(ll_inode2fid(child)), child);
+ ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, child, it, NULL);
- entry->se_inode = child;
+ entry->se_inode = child;
- if (agl_should_run(sai, child))
- ll_agl_add(sai, child, entry->se_index);
+ if (agl_should_run(sai, child))
+ ll_agl_add(sai, child, entry->se_index);
- EXIT;
+ EXIT;
out:
- /* sa_make_ready() will drop ldlm ibits lock refcount by calling
+ /*
+ * sa_make_ready() will drop ldlm ibits lock refcount by calling
* ll_intent_drop_lock() in spite of failures. Do not worry about
- * calling ll_intent_drop_lock() more than once. */
+ * calling ll_intent_drop_lock() more than once.
+ */
sa_make_ready(sai, entry, rc);
}
-/* once there are async stat replies, instantiate sa_entry */
+/* once there are async stat replies, instantiate sa_entry from replies */
static void sa_handle_callback(struct ll_statahead_info *sai)
{
struct ll_inode_info *lli;
- lli = ll_i2info(sai->sai_inode);
+ lli = ll_i2info(sai->sai_dentry->d_inode);
+ spin_lock(&lli->lli_sa_lock);
while (sa_has_callback(sai)) {
struct sa_entry *entry;
- spin_lock(&lli->lli_sa_lock);
- if (unlikely(!sa_has_callback(sai))) {
- spin_unlock(&lli->lli_sa_lock);
- break;
- }
entry = list_entry(sai->sai_interim_entries.next,
struct sa_entry, se_list);
list_del_init(&entry->se_list);
spin_unlock(&lli->lli_sa_lock);
sa_instantiate(sai, entry);
+ spin_lock(&lli->lli_sa_lock);
}
-
- spin_lock(&lli->lli_agl_lock);
- while (!agl_list_empty(sai)) {
- struct ll_inode_info *clli;
-
- clli = agl_first_entry(sai);
- list_del_init(&clli->lli_agl_list);
- spin_unlock(&lli->lli_agl_lock);
-
- ll_agl_trigger(&clli->lli_vfs_inode, sai);
-
- spin_lock(&lli->lli_agl_lock);
- }
- spin_unlock(&lli->lli_agl_lock);
+ spin_unlock(&lli->lli_sa_lock);
}
-/* callback for async stat, because this is called in ptlrpcd context, we only
- * put sa_entry in sai_cb_entries list, and let sa_handle_callback() to really
- * prepare inode and instantiate sa_entry later. */
+/*
+ * callback for async stat RPC, because this is called in ptlrpcd context, we
+ * only put sa_entry in sai_interim_entries, and wake up statahead thread to
+ * really prepare inode and instantiate sa_entry later.
+ */
static int ll_statahead_interpret(struct ptlrpc_request *req,
struct md_enqueue_info *minfo, int rc)
{
struct ll_statahead_info *sai = lli->lli_sai;
struct sa_entry *entry = (struct sa_entry *)minfo->mi_cbdata;
__u64 handle = 0;
- bool wakeup;
+
ENTRY;
if (it_disposition(it, DISP_LOOKUP_NEG))
rc = -ENOENT;
- /* because statahead thread will wait for all inflight RPC to finish,
- * sai should be always valid, no need to refcount */
+ /*
+ * because statahead thread will wait for all inflight RPC to finish,
+ * sai should be always valid, no need to refcount
+ */
LASSERT(sai != NULL);
- LASSERT(!thread_is_stopped(&sai->sai_thread));
LASSERT(entry != NULL);
CDEBUG(D_READA, "sa_entry %.*s rc %d\n",
if (rc != 0) {
ll_intent_release(it);
- iput(dir);
- OBD_FREE_PTR(minfo);
+ sa_fini_data(minfo);
} else {
- /* release ibits lock ASAP to avoid deadlock when statahead
+ /*
+ * release ibits lock ASAP to avoid deadlock when statahead
* thread enqueues lock on parent in readdir and another
* process enqueues lock on child with parent lock held, eg.
- * unlink. */
- handle = it->d.lustre.it_lock_handle;
+ * unlink.
+ */
+ handle = it->it_lock_handle;
ll_intent_drop_lock(it);
+ ll_unlock_md_op_lsm(&minfo->mi_data);
}
spin_lock(&lli->lli_sa_lock);
if (rc != 0) {
- wakeup = __sa_make_ready(sai, entry, rc);
+ if (__sa_make_ready(sai, entry, rc))
+ wake_up(&sai->sai_waitq);
} else {
+ int first = 0;
+
entry->se_minfo = minfo;
entry->se_req = ptlrpc_request_addref(req);
- /* Release the async ibits lock ASAP to avoid deadlock
+ /*
+ * 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. */
+ * with parent's lock held, for example: unlink.
+ */
entry->se_handle = handle;
- wakeup = !sa_has_callback(sai);
+ if (!sa_has_callback(sai))
+ first = 1;
+
list_add_tail(&entry->se_list, &sai->sai_interim_entries);
+ if (first && sai->sai_task)
+ wake_up_process(sai->sai_task);
}
sai->sai_replied++;
- if (wakeup)
- wake_up(&sai->sai_thread.t_ctl_waitq);
+
spin_unlock(&lli->lli_sa_lock);
RETURN(rc);
}
-static void sa_fini_data(struct md_enqueue_info *minfo,
- 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);
- OBD_FREE_PTR(einfo);
-}
-
-/**
- * There is race condition between "capa_put" and "ll_statahead_interpret" for
- * accessing "op_data.op_capa[1,2]" as following:
- * "capa_put" releases "op_data.op_capa[1,2]"'s reference count after calling
- * "md_intent_getattr_async". But "ll_statahead_interpret" maybe run first, and
- * fill "op_data.op_capa[1,2]" as POISON, then cause "capa_put" access invalid
- * "ocapa". So here reserve "op_data.op_capa[1,2]" in "pcapa" before calling
- * "md_intent_getattr_async".
- */
-static int sa_prep_data(struct inode *dir, struct inode *child,
- struct sa_entry *entry, struct md_enqueue_info **pmi,
- struct ldlm_enqueue_info **pei,
- struct obd_capa **pcapa)
-{
- struct qstr *qstr = &entry->se_qstr;
- struct md_enqueue_info *minfo;
- struct ldlm_enqueue_info *einfo;
- struct md_op_data *op_data;
-
- OBD_ALLOC_PTR(einfo);
- if (einfo == NULL)
- return -ENOMEM;
-
- OBD_ALLOC_PTR(minfo);
- if (minfo == NULL) {
- OBD_FREE_PTR(einfo);
- return -ENOMEM;
- }
-
- op_data = ll_prep_md_op_data(&minfo->mi_data, dir, child, qstr->name,
- qstr->len, 0, LUSTRE_OPC_ANY, NULL);
- if (IS_ERR(op_data)) {
- OBD_FREE_PTR(einfo);
- OBD_FREE_PTR(minfo);
- return PTR_ERR(op_data);
- }
-
- minfo->mi_it.it_op = IT_GETATTR;
- minfo->mi_dir = igrab(dir);
- minfo->mi_cb = ll_statahead_interpret;
- minfo->mi_cbdata = entry;
-
- einfo->ei_type = LDLM_IBITS;
- einfo->ei_mode = it_to_lock_mode(&minfo->mi_it);
- einfo->ei_cb_bl = ll_md_blocking_ast;
- einfo->ei_cb_cp = ldlm_completion_ast;
- einfo->ei_cb_gl = NULL;
- einfo->ei_cbdata = NULL;
-
- *pmi = minfo;
- *pei = einfo;
- pcapa[0] = op_data->op_capa1;
- pcapa[1] = op_data->op_capa2;
-
- return 0;
-}
-
+/* async stat for file not found in dcache */
static int sa_lookup(struct inode *dir, struct sa_entry *entry)
{
struct md_enqueue_info *minfo;
- struct ldlm_enqueue_info *einfo;
- struct obd_capa *capas[2];
int rc;
+
ENTRY;
- rc = sa_prep_data(dir, NULL, entry, &minfo, &einfo, capas);
- if (rc)
- RETURN(rc);
+ minfo = sa_prep_data(dir, NULL, entry);
+ if (IS_ERR(minfo))
+ RETURN(PTR_ERR(minfo));
- rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
- if (!rc) {
- capa_put(capas[0]);
- capa_put(capas[1]);
- } else {
- sa_fini_data(minfo, einfo);
- }
+ rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo);
+ if (rc < 0)
+ sa_fini_data(minfo);
RETURN(rc);
}
/**
- * similar to ll_revalidate_it().
- * \retval 1 -- dentry valid
- * \retval 0 -- will send stat-ahead request
- * \retval others -- prepare stat-ahead request failed
+ * async stat for file found in dcache, similar to .revalidate
+ *
+ * \retval 1 dentry valid, no RPC sent
+ * \retval 0 dentry invalid, will send async stat RPC
+ * \retval negative number upon error
*/
static int sa_revalidate(struct inode *dir, struct sa_entry *entry,
struct dentry *dentry)
{
struct inode *inode = dentry->d_inode;
struct lookup_intent it = { .it_op = IT_GETATTR,
- .d.lustre.it_lock_handle = 0 };
+ .it_lock_handle = 0 };
struct md_enqueue_info *minfo;
- struct ldlm_enqueue_info *einfo;
- struct obd_capa *capas[2];
int rc;
+
ENTRY;
- if (unlikely(inode == NULL))
+ if (unlikely(!inode))
RETURN(1);
if (d_mountpoint(dentry))
RETURN(1);
+ minfo = sa_prep_data(dir, inode, entry);
+ if (IS_ERR(minfo))
+ RETURN(PTR_ERR(minfo));
+
entry->se_inode = igrab(inode);
rc = md_revalidate_lock(ll_i2mdexp(dir), &it, ll_inode2fid(inode),
NULL);
if (rc == 1) {
- entry->se_handle = it.d.lustre.it_lock_handle;
+ entry->se_handle = it.it_lock_handle;
ll_intent_release(&it);
+ sa_fini_data(minfo);
RETURN(1);
}
- rc = sa_prep_data(dir, inode, entry, &minfo, &einfo, capas);
- if (rc) {
- entry->se_inode = NULL;
- iput(inode);
- RETURN(rc);
- }
-
- rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo, einfo);
- if (!rc) {
- capa_put(capas[0]);
- capa_put(capas[1]);
- } else {
+ rc = md_intent_getattr_async(ll_i2mdexp(dir), minfo);
+ if (rc < 0) {
entry->se_inode = NULL;
iput(inode);
- sa_fini_data(minfo, einfo);
+ sa_fini_data(minfo);
}
RETURN(rc);
}
-static void sa_statahead(struct dentry *parent, const char *name, int len)
+/* async stat for file with @name */
+static void sa_statahead(struct dentry *parent, const char *name, int len,
+ const struct lu_fid *fid)
{
struct inode *dir = parent->d_inode;
struct ll_inode_info *lli = ll_i2info(dir);
struct dentry *dentry = NULL;
struct sa_entry *entry;
int rc;
+
ENTRY;
- entry = sa_alloc(sai, sai->sai_index, name, len);
+ entry = sa_alloc(parent, sai, sai->sai_index, name, len, fid);
if (IS_ERR(entry))
RETURN_EXIT;
ll_agl_add(sai, dentry->d_inode, entry->se_index);
}
- if (dentry != NULL)
+ if (dentry)
dput(dentry);
if (rc != 0)
EXIT;
}
+/* async glimpse (agl) thread main function */
static int ll_agl_thread(void *arg)
{
struct dentry *parent = (struct dentry *)arg;
struct inode *dir = parent->d_inode;
struct ll_inode_info *plli = ll_i2info(dir);
struct ll_inode_info *clli;
- struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct ll_statahead_info *sai;
- struct ptlrpc_thread *thread;
- struct l_wait_info lwi = { 0 };
- ENTRY;
-
-
- sai = ll_sai_get(dir);
- thread = &sai->sai_agl_thread;
- thread->t_pid = current_pid();
- CDEBUG(D_READA, "agl thread started: sai %p, parent %.*s\n",
- sai, parent->d_name.len, parent->d_name.name);
-
- atomic_inc(&sbi->ll_agl_total);
- spin_lock(&plli->lli_agl_lock);
- sai->sai_agl_valid = 1;
- if (thread_is_init(thread))
- /* If someone else has changed the thread state
- * (e.g. already changed to SVC_STOPPING), we can't just
- * blindly overwrite that setting. */
- thread_set_flags(thread, SVC_RUNNING);
- spin_unlock(&plli->lli_agl_lock);
- wake_up(&thread->t_ctl_waitq);
+ /*
+ * We already own this reference, so it is safe to take it
+ * without a lock.
+ */
+ struct ll_statahead_info *sai = plli->lli_sai;
- while (1) {
- l_wait_event(thread->t_ctl_waitq,
- !agl_list_empty(sai) ||
- !thread_is_running(thread),
- &lwi);
+ ENTRY;
- if (!thread_is_running(thread))
- break;
+ CDEBUG(D_READA, "agl thread started: sai %p, parent %pd\n",
+ sai, parent);
+ while (({set_current_state(TASK_IDLE);
+ !kthread_should_stop(); })) {
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)) {
+ __set_current_state(TASK_RUNNING);
clli = agl_first_entry(sai);
list_del_init(&clli->lli_agl_list);
spin_unlock(&plli->lli_agl_lock);
ll_agl_trigger(&clli->lli_vfs_inode, sai);
+ cond_resched();
} else {
spin_unlock(&plli->lli_agl_lock);
+ schedule();
}
}
+ __set_current_state(TASK_RUNNING);
+ RETURN(0);
+}
+
+static void ll_stop_agl(struct ll_statahead_info *sai)
+{
+ struct dentry *parent = sai->sai_dentry;
+ struct ll_inode_info *plli = ll_i2info(parent->d_inode);
+ struct ll_inode_info *clli;
+ struct task_struct *agl_task;
+
+ spin_lock(&plli->lli_agl_lock);
+ agl_task = sai->sai_agl_task;
+ sai->sai_agl_task = NULL;
+ spin_unlock(&plli->lli_agl_lock);
+ if (!agl_task)
+ return;
+
+ CDEBUG(D_READA, "stop agl thread: sai %p pid %u\n",
+ sai, (unsigned int)agl_task->pid);
+ kthread_stop(agl_task);
spin_lock(&plli->lli_agl_lock);
- sai->sai_agl_valid = 0;
while (!agl_list_empty(sai)) {
clli = agl_first_entry(sai);
list_del_init(&clli->lli_agl_list);
iput(&clli->lli_vfs_inode);
spin_lock(&plli->lli_agl_lock);
}
- thread_set_flags(thread, SVC_STOPPED);
spin_unlock(&plli->lli_agl_lock);
- wake_up(&thread->t_ctl_waitq);
+ CDEBUG(D_READA, "agl thread stopped: sai %p, parent %pd\n",
+ sai, parent);
ll_sai_put(sai);
- CDEBUG(D_READA, "agl thread stopped: sai %p, parent %.*s\n",
- sai, parent->d_name.len, parent->d_name.name);
- RETURN(0);
}
+/* start agl thread */
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 };
- struct ll_inode_info *plli;
- struct task_struct *task;
+ int node = cfs_cpt_spread_node(cfs_cpt_tab, CFS_CPT_ANY);
+ struct ll_inode_info *plli;
+ struct task_struct *task;
+
ENTRY;
- CDEBUG(D_READA, "start agl thread: sai %p, parent %.*s\n",
- sai, parent->d_name.len, parent->d_name.name);
+ CDEBUG(D_READA, "start agl thread: sai %p, parent %pd\n",
+ sai, parent);
plli = ll_i2info(parent->d_inode);
- task = kthread_run(ll_agl_thread, parent,
- "ll_agl_%u", plli->lli_opendir_pid);
+ task = kthread_create_on_node(ll_agl_thread, parent, node, "ll_agl_%d",
+ 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;
}
+ sai->sai_agl_task = task;
+ atomic_inc(&ll_i2sbi(d_inode(parent))->ll_agl_total);
+ /* Get an extra reference that the thread holds */
+ ll_sai_get(d_inode(parent));
+
+ wake_up_process(task);
- l_wait_event(thread->t_ctl_waitq,
- thread_is_running(thread) || thread_is_stopped(thread),
- &lwi);
EXIT;
}
+/* statahead thread main function */
static int ll_statahead_thread(void *arg)
{
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);
- struct ll_statahead_info *sai;
- struct ptlrpc_thread *thread;
- struct ptlrpc_thread *agl_thread;
+ struct ll_statahead_info *sai = lli->lli_sai;
int first = 0;
struct md_op_data *op_data;
- struct ll_dir_chain chain;
- struct l_wait_info lwi = { 0 };
struct page *page = NULL;
__u64 pos = 0;
int rc = 0;
- ENTRY;
-
- sai = ll_sai_get(dir);
- thread = &sai->sai_thread;
- agl_thread = &sai->sai_agl_thread;
- thread->t_pid = current_pid();
- CDEBUG(D_READA, "statahead thread starting: sai %p, parent %.*s\n",
- sai, parent->d_name.len, parent->d_name.name);
- op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
- LUSTRE_OPC_ANY, dir);
- if (IS_ERR(op_data))
- GOTO(out, rc = PTR_ERR(op_data));
-
- op_data->op_max_pages = ll_i2sbi(dir)->ll_md_brw_pages;
+ ENTRY;
- if (sbi->ll_flags & LL_SBI_AGL_ENABLED)
- ll_start_agl(parent, sai);
+ CDEBUG(D_READA, "statahead thread starting: sai %p, parent %pd\n",
+ sai, parent);
- atomic_inc(&sbi->ll_sa_total);
- spin_lock(&lli->lli_sa_lock);
- if (thread_is_init(thread))
- /* If someone else has changed the thread state
- * (e.g. already changed to SVC_STOPPING), we can't just
- * blindly overwrite that setting. */
- thread_set_flags(thread, SVC_RUNNING);
- spin_unlock(&lli->lli_sa_lock);
- wake_up(&thread->t_ctl_waitq);
+ OBD_ALLOC_PTR(op_data);
+ if (!op_data)
+ GOTO(out, rc = -ENOMEM);
- ll_dir_chain_init(&chain);
- while (pos != MDS_DIR_END_OFF && thread_is_running(thread)) {
+ while (pos != MDS_DIR_END_OFF && sai->sai_task) {
struct lu_dirpage *dp;
struct lu_dirent *ent;
+ op_data = ll_prep_md_op_data(op_data, dir, dir, NULL, 0, 0,
+ LUSTRE_OPC_ANY, dir);
+ if (IS_ERR(op_data)) {
+ rc = PTR_ERR(op_data);
+ break;
+ }
+
sai->sai_in_readpage = 1;
- page = ll_get_dir_page(dir, op_data, pos, &chain);
+ page = ll_get_dir_page(dir, op_data, pos);
+ ll_unlock_md_op_lsm(op_data);
sai->sai_in_readpage = 0;
if (IS_ERR(page)) {
rc = PTR_ERR(page);
- CDEBUG(D_READA, "error reading dir "DFID" at "LPU64
- "/"LPU64" opendir_pid = %u: rc = %d\n",
+ CDEBUG(D_READA,
+ "error reading dir "DFID" at %llu /%llu opendir_pid = %u: rc = %d\n",
PFID(ll_inode2fid(dir)), pos, sai->sai_index,
lli->lli_opendir_pid, rc);
break;
dp = page_address(page);
for (ent = lu_dirent_start(dp);
- ent != NULL && thread_is_running(thread) &&
+ ent != NULL && sai->sai_task &&
!sa_low_hit(sai);
ent = lu_dirent_next(ent)) {
__u64 hash;
int namelen;
char *name;
+ struct lu_fid fid;
hash = le64_to_cpu(ent->lde_hash);
if (unlikely(hash < pos))
if (unlikely(++first == 1))
continue;
- /* wait for spare statahead window */
- do {
- l_wait_event(thread->t_ctl_waitq,
- !sa_sent_full(sai) ||
- sa_has_callback(sai) ||
- !agl_list_empty(sai) ||
- !thread_is_running(thread),
- &lwi);
+ fid_le_to_cpu(&fid, &ent->lde_fid);
+
+ while (({set_current_state(TASK_IDLE);
+ sai->sai_task; })) {
+ if (sa_has_callback(sai)) {
+ __set_current_state(TASK_RUNNING);
+ sa_handle_callback(sai);
+ }
+
+ spin_lock(&lli->lli_agl_lock);
+ while (sa_sent_full(sai) &&
+ !agl_list_empty(sai)) {
+ struct ll_inode_info *clli;
- sa_handle_callback(sai);
- } while (sa_sent_full(sai) &&
- thread_is_running(thread));
+ __set_current_state(TASK_RUNNING);
+ clli = agl_first_entry(sai);
+ list_del_init(&clli->lli_agl_list);
+ spin_unlock(&lli->lli_agl_lock);
- sa_statahead(parent, name, namelen);
+ ll_agl_trigger(&clli->lli_vfs_inode,
+ sai);
+ cond_resched();
+ spin_lock(&lli->lli_agl_lock);
+ }
+ spin_unlock(&lli->lli_agl_lock);
+
+ if (!sa_sent_full(sai))
+ break;
+ schedule();
+ }
+ __set_current_state(TASK_RUNNING);
+
+ sa_statahead(parent, name, namelen, &fid);
}
pos = le64_to_cpu(dp->ldp_hash_end);
if (sa_low_hit(sai)) {
rc = -EFAULT;
atomic_inc(&sbi->ll_sa_wrong);
- CDEBUG(D_READA, "Statahead for dir "DFID" hit "
- "ratio too low: hit/miss "LPU64"/"LPU64
- ", sent/replied "LPU64"/"LPU64", stopping "
- "statahead thread: pid %d\n",
+ CDEBUG(D_READA,
+ "Statahead for dir "DFID" hit ratio too low: hit/miss %llu/%llu, sent/replied %llu/%llu, stoppingstatahead thread: pid %d\n",
PFID(&lli->lli_fid), sai->sai_hit,
sai->sai_miss, sai->sai_sent,
- sai->sai_replied, current_pid());
+ sai->sai_replied, current->pid);
break;
}
}
- ll_dir_chain_fini(&chain);
ll_finish_md_op_data(op_data);
if (rc < 0) {
spin_lock(&lli->lli_sa_lock);
- thread_set_flags(thread, SVC_STOPPING);
+ sai->sai_task = NULL;
lli->lli_sa_enabled = 0;
spin_unlock(&lli->lli_sa_lock);
}
- /* statahead is finished, but statahead entries need to be cached, wait
- * for file release to stop me. */
- while (thread_is_running(thread)) {
- l_wait_event(thread->t_ctl_waitq,
- sa_has_callback(sai) ||
- !agl_list_empty(sai) ||
- !thread_is_running(thread),
- &lwi);
-
- sa_handle_callback(sai);
+ /*
+ * statahead is finished, but statahead entries need to be cached, wait
+ * for file release to stop me.
+ */
+ while (({set_current_state(TASK_IDLE);
+ sai->sai_task; })) {
+ if (sa_has_callback(sai)) {
+ __set_current_state(TASK_RUNNING);
+ sa_handle_callback(sai);
+ } else {
+ schedule();
+ }
}
+ __set_current_state(TASK_RUNNING);
EXIT;
out:
- if (sai->sai_agl_valid) {
- spin_lock(&lli->lli_agl_lock);
- thread_set_flags(agl_thread, SVC_STOPPING);
- spin_unlock(&lli->lli_agl_lock);
- wake_up(&agl_thread->t_ctl_waitq);
-
- CDEBUG(D_READA, "stop agl thread: sai %p pid %u\n",
- sai, (unsigned int)agl_thread->t_pid);
- l_wait_event(agl_thread->t_ctl_waitq,
- thread_is_stopped(agl_thread),
- &lwi);
- } else {
- /* Set agl_thread flags anyway. */
- thread_set_flags(agl_thread, SVC_STOPPED);
- }
+ ll_stop_agl(sai);
- /* wait for inflight statahead RPCs to finish, and then we can free sai
- * safely because statahead RPC will access sai data */
- while (sai->sai_sent != sai->sai_replied) {
+ /*
+ * wait for inflight statahead RPCs to finish, and then we can free sai
+ * safely because statahead RPC will access sai data
+ */
+ while (sai->sai_sent != sai->sai_replied)
/* in case we're not woken up, timeout wait */
- lwi = LWI_TIMEOUT(HZ >> 3, NULL, NULL);
- l_wait_event(thread->t_ctl_waitq,
- sai->sai_sent == sai->sai_replied, &lwi);
- }
+ msleep(125);
/* release resources held by statahead RPCs */
sa_handle_callback(sai);
+ CDEBUG(D_READA, "%s: statahead thread stopped: sai %p, parent %pd\n",
+ sbi->ll_fsname, sai, parent);
+
spin_lock(&lli->lli_sa_lock);
- thread_set_flags(thread, SVC_STOPPED);
+ sai->sai_task = NULL;
spin_unlock(&lli->lli_sa_lock);
-
wake_up(&sai->sai_waitq);
- wake_up(&thread->t_ctl_waitq);
- ll_sai_put(sai);
- CDEBUG(D_READA, "statahead thread stopped: sai %p, parent %.*s\n",
- sai, parent->d_name.len, parent->d_name.name);
- dput(parent);
+
+ ll_sai_put(sai);
+
return rc;
}
-/* authorize opened dir handle @key to statahead later */
+/* authorize opened dir handle @key to statahead */
void ll_authorize_statahead(struct inode *dir, void *key)
{
struct ll_inode_info *lli = ll_i2info(dir);
spin_lock(&lli->lli_sa_lock);
- if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL) {
+ if (!lli->lli_opendir_key && !lli->lli_sai) {
/*
* if lli_sai is not NULL, it means previous statahead is not
* finished yet, we'd better not start a new statahead for now.
*/
LASSERT(lli->lli_opendir_pid == 0);
lli->lli_opendir_key = key;
- lli->lli_opendir_pid = current_pid();
+ lli->lli_opendir_pid = current->pid;
lli->lli_sa_enabled = 1;
}
spin_unlock(&lli->lli_sa_lock);
}
-/* deauthorize opened dir handle @key to statahead, but statahead thread may
- * still be running, notify it to quit. */
+/*
+ * deauthorize opened dir handle @key to statahead, and notify statahead thread
+ * to quit if it's running.
+ */
void ll_deauthorize_statahead(struct inode *dir, void *key)
{
struct ll_inode_info *lli = ll_i2info(dir);
LASSERT(lli->lli_opendir_pid != 0);
CDEBUG(D_READA, "deauthorize statahead for "DFID"\n",
- PFID(&lli->lli_fid));
+ PFID(&lli->lli_fid));
spin_lock(&lli->lli_sa_lock);
lli->lli_opendir_key = NULL;
lli->lli_opendir_pid = 0;
lli->lli_sa_enabled = 0;
sai = lli->lli_sai;
- if (sai != NULL && thread_is_running(&sai->sai_thread)) {
+ if (sai && sai->sai_task) {
/*
- * statahead thread may not quit yet because it needs to cache
- * entries, now it's time to tell it to quit.
+ * statahead thread may not have quit yet because it needs to
+ * cache entries, now it's time to tell it to quit.
+ *
+ * wake_up_process() provides the necessary barriers
+ * to pair with set_current_state().
*/
- thread_set_flags(&sai->sai_thread, SVC_STOPPING);
- wake_up(&sai->sai_thread.t_ctl_waitq);
+ struct task_struct *task = sai->sai_task;
+
+ sai->sai_task = NULL;
+ wake_up_process(task);
}
spin_unlock(&lli->lli_sa_lock);
}
enum {
- /**
- * not first dirent, or is "."
- */
- LS_NONE_FIRST_DE = 0,
- /**
- * the first non-hidden dirent
- */
- LS_FIRST_DE,
- /**
- * the first hidden dirent, that is "."
- */
- LS_FIRST_DOT_DE
+ /**
+ * not first dirent, or is "."
+ */
+ LS_NOT_FIRST_DE = 0,
+ /**
+ * the first non-hidden dirent
+ */
+ LS_FIRST_DE,
+ /**
+ * the first hidden dirent, that is "."
+ */
+ LS_FIRST_DOT_DE
};
+/* file is first dirent under @dir */
static int is_first_dirent(struct inode *dir, struct dentry *dentry)
{
- struct ll_dir_chain chain;
struct qstr *target = &dentry->d_name;
struct md_op_data *op_data;
int dot_de;
struct page *page = NULL;
- int rc = LS_NONE_FIRST_DE;
+ int rc = LS_NOT_FIRST_DE;
__u64 pos = 0;
+
ENTRY;
op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
/**
*FIXME choose the start offset of the readdir
*/
- op_data->op_max_pages = ll_i2sbi(dir)->ll_md_brw_pages;
- ll_dir_chain_init(&chain);
- page = ll_get_dir_page(dir, op_data, 0, &chain);
+ page = ll_get_dir_page(dir, op_data, 0);
while (1) {
struct lu_dirpage *dp;
struct ll_inode_info *lli = ll_i2info(dir);
rc = PTR_ERR(page);
- CERROR("%s: reading dir "DFID" at "LPU64
- "opendir_pid = %u : rc = %d\n",
- ll_get_fsname(dir->i_sb, NULL, 0),
+ CERROR("%s: reading dir "DFID" at %llu opendir_pid = %u : rc = %d\n",
+ ll_i2sbi(dir)->ll_fsname,
PFID(ll_inode2fid(dir)), pos,
lli->lli_opendir_pid, rc);
break;
char *name;
hash = le64_to_cpu(ent->lde_hash);
- /* The ll_get_dir_page() can return any page containing
- * the given hash which may be not the start hash. */
+ /*
+ * The ll_get_dir_page() can return any page containing
+ * the given hash which may be not the start hash.
+ */
if (unlikely(hash < pos))
continue;
if (target->len != namelen ||
memcmp(target->name, name, namelen) != 0)
- rc = LS_NONE_FIRST_DE;
+ rc = LS_NOT_FIRST_DE;
else if (!dot_de)
rc = LS_FIRST_DE;
else
*/
ll_release_page(dir, page, le32_to_cpu(dp->ldp_flags) &
LDF_COLLIDE);
- page = ll_get_dir_page(dir, op_data, pos, &chain);
+ page = ll_get_dir_page(dir, op_data, pos);
}
}
EXIT;
out:
- ll_dir_chain_fini(&chain);
ll_finish_md_op_data(op_data);
- return rc;
+
+ return rc;
}
+/**
+ * revalidate @dentryp from statahead cache
+ *
+ * \param[in] dir parent directory
+ * \param[in] sai sai structure
+ * \param[out] dentryp pointer to dentry which will be revalidated
+ * \param[in] unplug unplug statahead window only (normally for negative
+ * dentry)
+ * \retval 1 on success, dentry is saved in @dentryp
+ * \retval 0 if revalidation failed (no proper lock on client)
+ * \retval negative number upon error
+ */
static int revalidate_statahead_dentry(struct inode *dir,
- struct ll_statahead_info *sai,
- struct dentry **dentryp,
- int only_unplug)
+ struct ll_statahead_info *sai,
+ struct dentry **dentryp,
+ bool unplug)
{
struct sa_entry *entry = NULL;
- struct l_wait_info lwi = { 0 };
+ struct ll_dentry_data *ldd;
+ struct ll_inode_info *lli = ll_i2info(dir);
int rc = 0;
+
ENTRY;
if ((*dentryp)->d_name.name[0] == '.') {
}
}
+ if (unplug)
+ GOTO(out, rc = 1);
+
entry = sa_get(sai, &(*dentryp)->d_name);
- if (entry == NULL || only_unplug) {
- sa_put(sai, entry);
- RETURN(entry ? 1 : -EAGAIN);
- }
+ if (!entry)
+ GOTO(out, rc = -EAGAIN);
/* if statahead is busy in readdir, help it do post-work */
if (!sa_ready(entry) && sai->sai_in_readpage)
sa_handle_callback(sai);
if (!sa_ready(entry)) {
+ spin_lock(&lli->lli_sa_lock);
sai->sai_index_wait = entry->se_index;
- lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(30), NULL,
- LWI_ON_SIGNAL_NOOP, NULL);
- rc = l_wait_event(sai->sai_waitq,
- sa_ready(entry) ||
- thread_is_stopped(&sai->sai_thread),
- &lwi);
- if (rc < 0) {
- sa_put(sai, entry);
- RETURN(-EAGAIN);
+ spin_unlock(&lli->lli_sa_lock);
+ rc = wait_event_idle_timeout(sai->sai_waitq, sa_ready(entry),
+ cfs_time_seconds(30));
+ if (rc == 0) {
+ /*
+ * entry may not be ready, so it may be used by inflight
+ * statahead RPC, don't free it.
+ */
+ entry = NULL;
+ GOTO(out, rc = -EAGAIN);
}
}
- if (entry->se_state == SA_ENTRY_SUCC && entry->se_inode != NULL) {
+ /*
+ * We need to see the value that was set immediately before we
+ * were woken up.
+ */
+ if (smp_load_acquire(&entry->se_state) == SA_ENTRY_SUCC &&
+ entry->se_inode) {
struct inode *inode = entry->se_inode;
struct lookup_intent it = { .it_op = IT_GETATTR,
- .d.lustre.it_lock_handle =
+ .it_lock_handle =
entry->se_handle };
__u64 bits;
rc = md_revalidate_lock(ll_i2mdexp(dir), &it,
ll_inode2fid(inode), &bits);
if (rc == 1) {
- if ((*dentryp)->d_inode == NULL) {
+ if (!(*dentryp)->d_inode) {
struct dentry *alias;
alias = ll_splice_alias(inode, *dentryp);
if (IS_ERR(alias)) {
- sa_put(sai, entry);
- RETURN(PTR_ERR(alias));
+ ll_intent_release(&it);
+ GOTO(out, rc = PTR_ERR(alias));
}
*dentryp = alias;
+ /*
+ * statahead prepared this inode, transfer inode
+ * refcount from sa_entry to dentry
+ */
+ entry->se_inode = NULL;
} else if ((*dentryp)->d_inode != inode) {
/* revalidate, but inode is recreated */
CDEBUG(D_READA,
- "%s: stale dentry %.*s inode "
- DFID", statahead inode "DFID
- "\n",
- ll_get_fsname((*dentryp)->d_inode->i_sb,
- NULL, 0),
- (*dentryp)->d_name.len,
- (*dentryp)->d_name.name,
- PFID(ll_inode2fid((*dentryp)->d_inode)),
- PFID(ll_inode2fid(inode)));
- sa_put(sai, entry);
- RETURN(-ESTALE);
- } else {
- iput(inode);
+ "%s: stale dentry %pd inode " DFID", statahead inode "DFID "\n",
+ ll_i2sbi(inode)->ll_fsname, *dentryp,
+ PFID(ll_inode2fid((*dentryp)->d_inode)),
+ PFID(ll_inode2fid(inode)));
+ ll_intent_release(&it);
+ GOTO(out, rc = -ESTALE);
}
- entry->se_inode = NULL;
if ((bits & MDS_INODELOCK_LOOKUP) &&
d_lustre_invalid(*dentryp))
ll_intent_release(&it);
}
}
-
+out:
+ /*
+ * statahead cached sa_entry can be used only once, and will be killed
+ * right after use, so if lookup/revalidate accessed statahead cache,
+ * set dentry ldd_sa_generation to parent lli_sa_generation, later if we
+ * stat this file again, we know we've done statahead before, see
+ * dentry_may_statahead().
+ */
+ ldd = ll_d2d(*dentryp);
+ /* ldd can be NULL if llite lookup failed. */
+ if (ldd)
+ ldd->lld_sa_generation = lli->lli_sa_generation;
sa_put(sai, entry);
+ spin_lock(&lli->lli_sa_lock);
+ if (sai->sai_task)
+ wake_up_process(sai->sai_task);
+ spin_unlock(&lli->lli_sa_lock);
+
RETURN(rc);
}
-static int start_statahead_thread(struct inode *dir, struct dentry *dentry)
+/**
+ * start statahead thread
+ *
+ * \param[in] dir parent directory
+ * \param[in] dentry dentry that triggers statahead, normally the first
+ * dirent under @dir
+ * \param[in] agl indicate whether AGL is needed
+ * \retval -EAGAIN on success, because when this function is
+ * called, it's already in lookup call, so client should
+ * do it itself instead of waiting for statahead thread
+ * to do it asynchronously.
+ * \retval negative number upon error
+ */
+static int start_statahead_thread(struct inode *dir, struct dentry *dentry,
+ bool agl)
{
+ int node = cfs_cpt_spread_node(cfs_cpt_tab, CFS_CPT_ANY);
struct ll_inode_info *lli = ll_i2info(dir);
struct ll_statahead_info *sai = NULL;
- struct dentry *parent;
- struct ptlrpc_thread *thread;
- struct l_wait_info lwi = { 0 };
+ struct dentry *parent = dentry->d_parent;
struct task_struct *task;
- int rc;
+ struct ll_sb_info *sbi = ll_i2sbi(parent->d_inode);
+ int first = LS_FIRST_DE;
+ int rc = 0;
+
ENTRY;
/* I am the "lli_opendir_pid" owner, only me can set "lli_sai". */
- rc = is_first_dirent(dir, dentry);
- if (rc == LS_NONE_FIRST_DE)
+ first = is_first_dirent(dir, dentry);
+ if (first == LS_NOT_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)
- GOTO(out, rc = -ENOMEM);
+ GOTO(out, rc = -EFAULT);
- 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));
- GOTO(out, rc = -ESTALE);
+ if (unlikely(atomic_inc_return(&sbi->ll_sa_running) >
+ sbi->ll_sa_running_max)) {
+ CDEBUG(D_READA,
+ "Too many concurrent statahead instances, avoid new statahead instance temporarily.\n");
+ GOTO(out, rc = -EMFILE);
}
- /* get parent reference count here, and put it in ll_statahead_thread */
- parent = dget(dentry->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",
- dentry->d_name.len, dentry->d_name.name,
- PFID(&lli->lli_fid), PFID(&nlli->lli_fid));
- dput(parent);
- iput(sai->sai_inode);
- GOTO(out, rc = -EAGAIN);
- }
+ sai = ll_sai_alloc(parent);
+ if (!sai)
+ GOTO(out, rc = -ENOMEM);
- CDEBUG(D_READA, "start statahead thread: sai %p, parent %.*s\n",
- sai, parent->d_name.len, parent->d_name.name);
+ sai->sai_ls_all = (first == LS_FIRST_DOT_DE);
- /* if another process started statahead thread, or deauthorized current
- * lli_opendir_key, don't start statahead. */
+ /*
+ * if current lli_opendir_key was deauthorized, or dir re-opened by
+ * another process, don't start statahead, otherwise the newly spawned
+ * statahead thread won't be notified to quit.
+ */
spin_lock(&lli->lli_sa_lock);
- if (unlikely(lli->lli_sai != NULL ||
- lli->lli_opendir_key == NULL ||
+ if (unlikely(lli->lli_sai || !lli->lli_opendir_key ||
lli->lli_opendir_pid != current->pid)) {
spin_unlock(&lli->lli_sa_lock);
-
- dput(parent);
- iput(sai->sai_inode);
- GOTO(out, rc = -EAGAIN);
+ GOTO(out, rc = -EPERM);
}
lli->lli_sai = sai;
spin_unlock(&lli->lli_sa_lock);
- atomic_inc(&ll_i2sbi(parent->d_inode)->ll_sa_running);
+ CDEBUG(D_READA, "start statahead thread: [pid %d] [parent %pd]\n",
+ current->pid, parent);
- task = kthread_run(ll_statahead_thread, parent, "ll_sa_%u",
- lli->lli_opendir_pid);
- thread = &sai->sai_thread;
+ task = kthread_create_on_node(ll_statahead_thread, parent, node,
+ "ll_sa_%u", lli->lli_opendir_pid);
if (IS_ERR(task)) {
- rc = PTR_ERR(task);
- CERROR("cannot start ll_sa thread: rc = %d\n", rc);
- dput(parent);
-
spin_lock(&lli->lli_sa_lock);
- thread_set_flags(thread, SVC_STOPPED);
- thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
+ lli->lli_sai = NULL;
spin_unlock(&lli->lli_sa_lock);
-
- ll_sai_put(sai);
- LASSERT(lli->lli_sai == NULL);
- RETURN(-EAGAIN);
+ rc = PTR_ERR(task);
+ CERROR("can't start ll_sa thread, rc: %d\n", rc);
+ GOTO(out, rc);
}
- l_wait_event(thread->t_ctl_waitq,
- thread_is_running(thread) || thread_is_stopped(thread),
- &lwi);
- ll_sai_put(sai);
+ if (ll_i2sbi(parent->d_inode)->ll_flags & LL_SBI_AGL_ENABLED && agl)
+ ll_start_agl(parent, sai);
+ atomic_inc(&ll_i2sbi(parent->d_inode)->ll_sa_total);
+ sai->sai_task = task;
+
+ wake_up_process(task);
/*
* We don't stat-ahead for the first dirent since we are already in
* lookup.
RETURN(-EAGAIN);
out:
- if (sai != NULL)
- OBD_FREE_PTR(sai);
-
- /* once we start statahead thread failed, disable statahead so
- * subsequent won't waste time to try it. */
+ /*
+ * once we start statahead thread failed, disable statahead so that
+ * subsequent stat won't waste time to try it.
+ */
spin_lock(&lli->lli_sa_lock);
- lli->lli_sa_enabled = 0;
+ if (lli->lli_opendir_pid == current->pid)
+ lli->lli_sa_enabled = 0;
spin_unlock(&lli->lli_sa_lock);
+ if (sai)
+ ll_sai_free(sai);
+ if (first != LS_NOT_FIRST_DE)
+ atomic_dec(&sbi->ll_sa_running);
+
RETURN(rc);
}
+/*
+ * Check whether statahead for @dir was started.
+ */
+static inline bool ll_statahead_started(struct inode *dir, bool agl)
+{
+ struct ll_inode_info *lli = ll_i2info(dir);
+ struct ll_statahead_info *sai;
+
+ spin_lock(&lli->lli_sa_lock);
+ sai = lli->lli_sai;
+ if (sai && (sai->sai_agl_task != NULL) != agl)
+ CDEBUG(D_READA,
+ "%s: Statahead AGL hint changed from %d to %d\n",
+ ll_i2sbi(dir)->ll_fsname,
+ sai->sai_agl_task != NULL, agl);
+ spin_unlock(&lli->lli_sa_lock);
+
+ return !!sai;
+}
+
+/**
+ * statahead entry function, this is called when client getattr on a file, it
+ * will start statahead thread if this is the first dir entry, else revalidate
+ * dentry from statahead cache.
+ *
+ * \param[in] dir parent directory
+ * \param[out] dentryp dentry to getattr
+ * \param[in] agl whether start the agl thread
+ *
+ * \retval 1 on success
+ * \retval 0 revalidation from statahead cache failed, caller needs
+ * to getattr from server directly
+ * \retval negative number on error, caller often ignores this and
+ * then getattr from server
+ */
+int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl)
+{
+ if (!ll_statahead_started(dir, agl))
+ return start_statahead_thread(dir, dentry, agl);
+ return 0;
+}
+
/**
- * Start statahead thread if this is the first dir entry.
- * Otherwise if a thread is started already, wait it until it is ahead of me.
- * \retval 1 -- find entry with lock in cache, the caller needs to do
- * nothing.
- * \retval 0 -- find entry in cache, but without lock, the caller needs
- * refresh from MDS.
- * \retval others -- the caller need to process as non-statahead.
+ * revalidate dentry from statahead cache.
+ *
+ * \param[in] dir parent directory
+ * \param[out] dentryp dentry to getattr
+ * \param[in] unplug unplug statahead window only (normally for negative
+ * dentry)
+ * \retval 1 on success
+ * \retval 0 revalidation from statahead cache failed, caller needs
+ * to getattr from server directly
+ * \retval negative number on error, caller often ignores this and
+ * then getattr from server
*/
-int do_statahead_enter(struct inode *dir, struct dentry **dentryp,
- int only_unplug)
+int ll_revalidate_statahead(struct inode *dir, struct dentry **dentryp,
+ bool unplug)
{
struct ll_statahead_info *sai;
+ int rc = 0;
sai = ll_sai_get(dir);
- if (sai != NULL) {
- int rc;
-
- rc = revalidate_statahead_dentry(dir, sai, dentryp,
- only_unplug);
- CDEBUG(D_READA, "revalidate statahead %.*s: %d.\n",
- (*dentryp)->d_name.len, (*dentryp)->d_name.name, rc);
+ if (sai) {
+ rc = revalidate_statahead_dentry(dir, sai, dentryp, unplug);
+ CDEBUG(D_READA, "revalidate statahead %pd: rc = %d.\n",
+ *dentryp, rc);
ll_sai_put(sai);
- return rc;
}
-
- return start_statahead_thread(dir, *dentryp);
+ return rc;
}