*
* 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) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2010, 2013, Intel Corporation.
+ * Copyright (c) 2010, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*/
#define DEBUG_SUBSYSTEM S_LDLM
-#ifndef __KERNEL__
-#include <signal.h>
-#include <liblustre.h>
-#endif
+#include <lustre_errno.h>
#include <lustre_dlm.h>
#include <obd_class.h>
#include <obd.h>
#include "ldlm_internal.h"
-int ldlm_enqueue_min = OBD_TIMEOUT_DEFAULT;
-CFS_MODULE_PARM(ldlm_enqueue_min, "i", int, 0644,
- "lock enqueue timeout minimum");
+unsigned int ldlm_enqueue_min = OBD_TIMEOUT_DEFAULT;
+module_param(ldlm_enqueue_min, uint, 0644);
+MODULE_PARM_DESC(ldlm_enqueue_min, "lock enqueue timeout minimum");
/* in client side, whether the cached locks will be canceled before replay */
unsigned int ldlm_cancel_unused_locks_before_replay = 1;
}
struct lock_wait_data {
- struct ldlm_lock *lwd_lock;
- __u32 lwd_conn_cnt;
+ struct ldlm_lock *lwd_lock;
+ __u32 lwd_conn_cnt;
};
struct ldlm_async_args {
- struct lustre_handle lock_handle;
+ struct lustre_handle lock_handle;
};
+/**
+ * ldlm_request_bufsize
+ *
+ * If opcode=LDLM_ENQUEUE, 1 slot is already occupied,
+ * LDLM_LOCKREQ_HANDLE -1 slots are available.
+ * Otherwise, LDLM_LOCKREQ_HANDLE slots are available.
+ *
+ * \param[in] count
+ * \param[in] type
+ *
+ * \retval size of the request buffer
+ */
+int ldlm_request_bufsize(int count, int type)
+{
+ int avail = LDLM_LOCKREQ_HANDLES;
+
+ if (type == LDLM_ENQUEUE)
+ avail -= LDLM_ENQUEUE_CANCEL_OFF;
+
+ if (count > avail)
+ avail = (count - avail) * sizeof(struct lustre_handle);
+ else
+ avail = 0;
+
+ return sizeof(struct ldlm_request) + avail;
+}
+
int ldlm_expired_completion_wait(void *data)
{
- struct lock_wait_data *lwd = data;
- struct ldlm_lock *lock = lwd->lwd_lock;
- struct obd_import *imp;
- struct obd_device *obd;
-
- ENTRY;
- if (lock->l_conn_export == NULL) {
- static cfs_time_t next_dump = 0, last_dump = 0;
-
- LCONSOLE_WARN("lock timed out (enqueued at "CFS_TIME_T", "
- CFS_DURATION_T"s ago)\n",
- lock->l_last_activity,
- cfs_time_sub(cfs_time_current_sec(),
- lock->l_last_activity));
- LDLM_DEBUG(lock, "lock timed out (enqueued at "CFS_TIME_T", "
- CFS_DURATION_T"s ago); not entering recovery in "
- "server code, just going back to sleep",
- lock->l_last_activity,
- cfs_time_sub(cfs_time_current_sec(),
- lock->l_last_activity));
- if (cfs_time_after(cfs_time_current(), next_dump)) {
- last_dump = next_dump;
- next_dump = cfs_time_shift(300);
- ldlm_namespace_dump(D_DLMTRACE,
- ldlm_lock_to_ns(lock));
- if (last_dump == 0)
- libcfs_debug_dumplog();
- }
- RETURN(0);
- }
-
- obd = lock->l_conn_export->exp_obd;
- imp = obd->u.cli.cl_import;
- ptlrpc_fail_import(imp, lwd->lwd_conn_cnt);
- LDLM_ERROR(lock, "lock timed out (enqueued at "CFS_TIME_T", "
- CFS_DURATION_T"s ago), entering recovery for %s@%s",
- lock->l_last_activity,
- cfs_time_sub(cfs_time_current_sec(), lock->l_last_activity),
- obd2cli_tgt(obd), imp->imp_connection->c_remote_uuid.uuid);
-
- RETURN(0);
+ struct lock_wait_data *lwd = data;
+ struct ldlm_lock *lock = lwd->lwd_lock;
+ struct obd_import *imp;
+ struct obd_device *obd;
+
+ ENTRY;
+ if (lock->l_conn_export == NULL) {
+ static time64_t next_dump, last_dump;
+
+ LDLM_ERROR(lock,
+ "lock timed out (enqueued at %lld, %llds ago); not entering recovery in server code, just going back to sleep",
+ (s64)lock->l_activity,
+ (s64)(ktime_get_real_seconds() -
+ lock->l_activity));
+ if (ktime_get_seconds() > next_dump) {
+ last_dump = next_dump;
+ next_dump = ktime_get_seconds() + 300;
+ ldlm_namespace_dump(D_DLMTRACE,
+ ldlm_lock_to_ns(lock));
+ if (last_dump == 0)
+ libcfs_debug_dumplog();
+ }
+ RETURN(0);
+ }
+
+ obd = lock->l_conn_export->exp_obd;
+ imp = obd->u.cli.cl_import;
+ ptlrpc_fail_import(imp, lwd->lwd_conn_cnt);
+ LDLM_ERROR(lock,
+ "lock timed out (enqueued at %lld, %llds ago), entering recovery for %s@%s",
+ (s64)lock->l_activity,
+ (s64)(ktime_get_real_seconds() - lock->l_activity),
+ obd2cli_tgt(obd), imp->imp_connection->c_remote_uuid.uuid);
+
+ RETURN(0);
}
-EXPORT_SYMBOL(ldlm_expired_completion_wait);
-/* We use the same basis for both server side and client side functions
- from a single node. */
-int ldlm_get_enq_timeout(struct ldlm_lock *lock)
+int is_granted_or_cancelled_nolock(struct ldlm_lock *lock)
{
- int timeout = at_get(ldlm_lock_to_ns_at(lock));
- if (AT_OFF)
- return obd_timeout / 2;
- /* Since these are non-updating timeouts, we should be conservative.
- It would be nice to have some kind of "early reply" mechanism for
- lock callbacks too... */
- timeout = min_t(int, at_max, timeout + (timeout >> 1)); /* 150% */
- return max(timeout, ldlm_enqueue_min);
+ int ret = 0;
+
+ check_res_locked(lock->l_resource);
+ if (ldlm_is_granted(lock) && !ldlm_is_cp_reqd(lock))
+ ret = 1;
+ else if (ldlm_is_failed(lock) || ldlm_is_cancel(lock))
+ ret = 1;
+ return ret;
+}
+EXPORT_SYMBOL(is_granted_or_cancelled_nolock);
+
+/**
+ * Calculate the Completion timeout (covering enqueue, BL AST, data flush,
+ * lock cancel, and their replies). Used for lock completion timeout on the
+ * client side.
+ *
+ * \param[in] lock lock which is waiting the completion callback
+ *
+ * \retval timeout in seconds to wait for the server reply
+ */
+/*
+ * We use the same basis for both server side and client side functions
+ * from a single node.
+ */
+static time64_t ldlm_cp_timeout(struct ldlm_lock *lock)
+{
+ time64_t timeout;
+
+ if (AT_OFF)
+ return obd_timeout;
+
+ /*
+ * Wait a long time for enqueue - server may have to callback a
+ * lock from another client. Server will evict the other client if it
+ * doesn't respond reasonably, and then give us the lock.
+ */
+ timeout = at_get(ldlm_lock_to_ns_at(lock));
+ return max(3 * timeout, (time64_t) ldlm_enqueue_min);
}
-EXPORT_SYMBOL(ldlm_get_enq_timeout);
/**
* Helper function for ldlm_completion_ast(), updating timings when lock is
* actually granted.
*/
-static int ldlm_completion_tail(struct ldlm_lock *lock)
+static int ldlm_completion_tail(struct ldlm_lock *lock, void *data)
{
- long delay;
- int result;
+ time64_t delay;
+ int result = 0;
if (ldlm_is_destroyed(lock) || ldlm_is_failed(lock)) {
LDLM_DEBUG(lock, "client-side enqueue: destroyed");
result = -EIO;
+ } else if (data == NULL) {
+ LDLM_DEBUG(lock, "client-side enqueue: granted");
} else {
- delay = cfs_time_sub(cfs_time_current_sec(),
- lock->l_last_activity);
- LDLM_DEBUG(lock, "client-side enqueue: granted after "
- CFS_DURATION_T"s", delay);
+ /* Take into AT only CP RPC, not immediately granted locks */
+ delay = ktime_get_real_seconds() - lock->l_activity;
+ LDLM_DEBUG(lock, "client-side enqueue: granted after %llds",
+ (s64)delay);
/* Update our time estimate */
- at_measured(ldlm_lock_to_ns_at(lock),
- delay);
- result = 0;
+ at_measured(ldlm_lock_to_ns_at(lock), delay);
}
return result;
}
RETURN(0);
}
- if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
- LDLM_FL_BLOCK_CONV))) {
+ if (!(flags & LDLM_FL_BLOCKED_MASK)) {
wake_up(&lock->l_waitq);
- RETURN(ldlm_completion_tail(lock));
+ RETURN(ldlm_completion_tail(lock, data));
}
- LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
- "going forward");
- ldlm_reprocess_all(lock->l_resource);
+ LDLM_DEBUG(lock,
+ "client-side enqueue returned a blocked lock, going forward");
+ ldlm_reprocess_all(lock->l_resource, NULL);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_completion_ast_async);
*
* - to force all locks when resource is destroyed (cleanup_resource());
*
- * - during lock conversion (not used currently).
- *
* If lock is not granted in the first case, this function waits until second
* or penultimate cases happen in some other thread.
*
*/
int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
{
- /* XXX ALLOCATE - 160 bytes */
- struct lock_wait_data lwd;
- struct obd_device *obd;
- struct obd_import *imp = NULL;
- struct l_wait_info lwi;
- __u32 timeout;
- int rc = 0;
- ENTRY;
-
- if (flags == LDLM_FL_WAIT_NOREPROC) {
- LDLM_DEBUG(lock, "client-side enqueue waiting on pending lock");
- goto noreproc;
- }
-
- if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
- LDLM_FL_BLOCK_CONV))) {
+ /* XXX ALLOCATE - 160 bytes */
+ struct lock_wait_data lwd;
+ struct obd_device *obd;
+ struct obd_import *imp = NULL;
+ struct l_wait_info lwi;
+ time64_t timeout;
+ int rc = 0;
+
+ ENTRY;
+
+ if (flags == LDLM_FL_WAIT_NOREPROC) {
+ LDLM_DEBUG(lock, "client-side enqueue waiting on pending lock");
+ goto noreproc;
+ }
+
+ if (!(flags & LDLM_FL_BLOCKED_MASK)) {
wake_up(&lock->l_waitq);
RETURN(0);
}
- LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
- "sleeping");
+ LDLM_DEBUG(lock, "client-side enqueue returned a blocked locksleeping");
noreproc:
- obd = class_exp2obd(lock->l_conn_export);
+ obd = class_exp2obd(lock->l_conn_export);
- /* if this is a local lock, then there is no import */
- if (obd != NULL) {
- imp = obd->u.cli.cl_import;
- }
+ /* if this is a local lock, then there is no import */
+ if (obd != NULL)
+ imp = obd->u.cli.cl_import;
- /* Wait a long time for enqueue - server may have to callback a
- lock from another client. Server will evict the other client if it
- doesn't respond reasonably, and then give us the lock. */
- timeout = ldlm_get_enq_timeout(lock) * 2;
+ timeout = ldlm_cp_timeout(lock);
- lwd.lwd_lock = lock;
+ lwd.lwd_lock = lock;
+ lock->l_activity = ktime_get_real_seconds();
if (ldlm_is_no_timeout(lock)) {
- LDLM_DEBUG(lock, "waiting indefinitely because of NO_TIMEOUT");
- lwi = LWI_INTR(interrupted_completion_wait, &lwd);
- } else {
- lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(timeout),
- ldlm_expired_completion_wait,
- interrupted_completion_wait, &lwd);
- }
+ LDLM_DEBUG(lock, "waiting indefinitely because of NO_TIMEOUT");
+ lwi = LWI_INTR(interrupted_completion_wait, &lwd);
+ } else {
+ lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(timeout),
+ ldlm_expired_completion_wait,
+ interrupted_completion_wait, &lwd);
+ }
if (imp != NULL) {
spin_lock(&imp->imp_lock);
spin_unlock(&imp->imp_lock);
}
- if (ns_is_client(ldlm_lock_to_ns(lock)) &&
- OBD_FAIL_CHECK_RESET(OBD_FAIL_LDLM_INTR_CP_AST,
- OBD_FAIL_LDLM_CP_BL_RACE | OBD_FAIL_ONCE)) {
+ if (ns_is_client(ldlm_lock_to_ns(lock)) &&
+ OBD_FAIL_CHECK_RESET(OBD_FAIL_LDLM_INTR_CP_AST,
+ OBD_FAIL_LDLM_CP_BL_RACE | OBD_FAIL_ONCE)) {
ldlm_set_fail_loc(lock);
- rc = -EINTR;
- } else {
- /* Go to sleep until the lock is granted or cancelled. */
- rc = l_wait_event(lock->l_waitq,
- is_granted_or_cancelled(lock), &lwi);
- }
-
- if (rc) {
- LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
- rc);
- RETURN(rc);
- }
-
- RETURN(ldlm_completion_tail(lock));
+ rc = -EINTR;
+ } else {
+ /* Go to sleep until the lock is granted or cancelled. */
+ rc = l_wait_event(lock->l_waitq,
+ is_granted_or_cancelled(lock), &lwi);
+ }
+
+ if (rc) {
+ LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
+ rc);
+ RETURN(rc);
+ }
+
+ RETURN(ldlm_completion_tail(lock, data));
}
EXPORT_SYMBOL(ldlm_completion_ast);
*/
int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock)
{
- int do_ast;
- ENTRY;
+ int do_ast;
+
+ ENTRY;
ldlm_set_cbpending(lock);
- do_ast = (!lock->l_readers && !lock->l_writers);
- unlock_res_and_lock(lock);
+ do_ast = (!lock->l_readers && !lock->l_writers);
+ unlock_res_and_lock(lock);
- if (do_ast) {
- struct lustre_handle lockh;
- int rc;
+ if (do_ast) {
+ struct lustre_handle lockh;
+ int rc;
- LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
- ldlm_lock2handle(lock, &lockh);
+ LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
+ ldlm_lock2handle(lock, &lockh);
rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
- if (rc < 0)
- CERROR("ldlm_cli_cancel: %d\n", rc);
- } else {
- LDLM_DEBUG(lock, "Lock still has references, will be "
- "cancelled later");
- }
- RETURN(0);
+ if (rc < 0)
+ CERROR("ldlm_cli_cancel: %d\n", rc);
+ } else {
+ LDLM_DEBUG(lock,
+ "Lock still has references, will be cancelled later");
+ }
+ RETURN(0);
}
EXPORT_SYMBOL(ldlm_blocking_ast_nocheck);
* \see ldlm_blocking_ast_nocheck
*/
int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
- void *data, int flag)
+ void *data, int flag)
{
- ENTRY;
-
- if (flag == LDLM_CB_CANCELING) {
- /* Don't need to do anything here. */
- RETURN(0);
- }
-
- lock_res_and_lock(lock);
- /* Get this: if ldlm_blocking_ast is racing with intent_policy, such
- * that ldlm_blocking_ast is called just before intent_policy method
- * takes the lr_lock, then by the time we get the lock, we might not
- * be the correct blocking function anymore. So check, and return
- * early, if so. */
- if (lock->l_blocking_ast != ldlm_blocking_ast) {
- unlock_res_and_lock(lock);
- RETURN(0);
- }
- RETURN(ldlm_blocking_ast_nocheck(lock));
+ ENTRY;
+
+ if (flag == LDLM_CB_CANCELING) {
+ /* Don't need to do anything here. */
+ RETURN(0);
+ }
+
+ lock_res_and_lock(lock);
+ /*
+ * Get this: if ldlm_blocking_ast is racing with intent_policy, such
+ * that ldlm_blocking_ast is called just before intent_policy method
+ * takes the lr_lock, then by the time we get the lock, we might not
+ * be the correct blocking function anymore. So check, and return
+ * early, if so.
+ */
+ if (lock->l_blocking_ast != ldlm_blocking_ast) {
+ unlock_res_and_lock(lock);
+ RETURN(0);
+ }
+ RETURN(ldlm_blocking_ast_nocheck(lock));
}
EXPORT_SYMBOL(ldlm_blocking_ast);
*/
int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp)
{
- return -ELDLM_NO_LOCK_DATA;
+ return -ELDLM_NO_LOCK_DATA;
}
-EXPORT_SYMBOL(ldlm_glimpse_ast);
/**
* Enqueue a local lock (typically on a server).
*/
-int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
- const struct ldlm_res_id *res_id,
- ldlm_type_t type, ldlm_policy_data_t *policy,
- ldlm_mode_t mode, __u64 *flags,
- ldlm_blocking_callback blocking,
- ldlm_completion_callback completion,
- ldlm_glimpse_callback glimpse,
+int ldlm_cli_enqueue_local(const struct lu_env *env,
+ struct ldlm_namespace *ns,
+ const struct ldlm_res_id *res_id,
+ enum ldlm_type type, union ldlm_policy_data *policy,
+ enum ldlm_mode mode, __u64 *flags,
+ ldlm_blocking_callback blocking,
+ ldlm_completion_callback completion,
+ ldlm_glimpse_callback glimpse,
void *data, __u32 lvb_len, enum lvb_type lvb_type,
- const __u64 *client_cookie,
- struct lustre_handle *lockh)
+ const __u64 *client_cookie,
+ struct lustre_handle *lockh)
{
- struct ldlm_lock *lock;
- int err;
- const struct ldlm_callback_suite cbs = { .lcs_completion = completion,
- .lcs_blocking = blocking,
- .lcs_glimpse = glimpse,
- };
- ENTRY;
-
- LASSERT(!(*flags & LDLM_FL_REPLAY));
- if (unlikely(ns_is_client(ns))) {
- CERROR("Trying to enqueue local lock in a shadow namespace\n");
- LBUG();
- }
+ struct ldlm_lock *lock;
+ int err;
+ const struct ldlm_callback_suite cbs = { .lcs_completion = completion,
+ .lcs_blocking = blocking,
+ .lcs_glimpse = glimpse,
+ };
+
+ ENTRY;
+
+ LASSERT(!(*flags & LDLM_FL_REPLAY));
+ if (unlikely(ns_is_client(ns))) {
+ CERROR("Trying to enqueue local lock in a shadow namespace\n");
+ LBUG();
+ }
lock = ldlm_lock_create(ns, res_id, type, mode, &cbs, data, lvb_len,
lvb_type);
err = ldlm_lvbo_init(lock->l_resource);
if (err < 0) {
LDLM_ERROR(lock, "delayed lvb init failed (rc %d)", err);
+ ldlm_lock_destroy_nolock(lock);
GOTO(out, err);
}
- ldlm_lock2handle(lock, lockh);
+ ldlm_lock2handle(lock, lockh);
- /* NB: we don't have any lock now (lock_res_and_lock)
- * because it's a new lock */
- ldlm_lock_addref_internal_nolock(lock, mode);
+ /*
+ * NB: we don't have any lock now (lock_res_and_lock)
+ * because it's a new lock
+ */
+ ldlm_lock_addref_internal_nolock(lock, mode);
ldlm_set_local(lock);
- if (*flags & LDLM_FL_ATOMIC_CB)
+ if (*flags & LDLM_FL_ATOMIC_CB)
ldlm_set_atomic_cb(lock);
- if (policy != NULL)
- lock->l_policy_data = *policy;
- if (client_cookie != NULL)
- lock->l_client_cookie = *client_cookie;
+ if (*flags & LDLM_FL_CANCEL_ON_BLOCK)
+ ldlm_set_cancel_on_block(lock);
+
+ if (policy != NULL)
+ lock->l_policy_data = *policy;
+ if (client_cookie != NULL)
+ lock->l_client_cookie = *client_cookie;
if (type == LDLM_EXTENT) {
/* extent lock without policy is a bug */
if (policy == NULL)
lock->l_req_extent = policy->l_extent;
}
- err = ldlm_lock_enqueue(ns, &lock, policy, flags);
- if (unlikely(err != ELDLM_OK))
- GOTO(out, err);
+ err = ldlm_lock_enqueue(env, ns, &lock, policy, flags);
+ if (unlikely(err != ELDLM_OK))
+ GOTO(out, err);
- if (policy != NULL)
- *policy = lock->l_policy_data;
+ if (policy != NULL)
+ *policy = lock->l_policy_data;
- if (lock->l_completion_ast)
- lock->l_completion_ast(lock, *flags, NULL);
+ if (lock->l_completion_ast)
+ lock->l_completion_ast(lock, *flags, NULL);
- LDLM_DEBUG(lock, "client-side local enqueue handler, new lock created");
- EXIT;
+ LDLM_DEBUG(lock, "client-side local enqueue handler, new lock created");
+ EXIT;
out:
- LDLM_LOCK_RELEASE(lock);
+ LDLM_LOCK_RELEASE(lock);
out_nolock:
- return err;
+ return err;
}
EXPORT_SYMBOL(ldlm_cli_enqueue_local);
static void failed_lock_cleanup(struct ldlm_namespace *ns,
- struct ldlm_lock *lock, int mode)
+ struct ldlm_lock *lock, int mode)
{
- int need_cancel = 0;
-
- /* Set a flag to prevent us from sending a CANCEL (bug 407) */
- lock_res_and_lock(lock);
- /* Check that lock is not granted or failed, we might race. */
- if ((lock->l_req_mode != lock->l_granted_mode) &&
- !ldlm_is_failed(lock)) {
- /* Make sure that this lock will not be found by raced
+ int need_cancel = 0;
+
+ /* Set a flag to prevent us from sending a CANCEL (b=407) */
+ lock_res_and_lock(lock);
+ /* Check that lock is not granted or failed, we might race. */
+ if (!ldlm_is_granted(lock) && !ldlm_is_failed(lock)) {
+ /*
+ * Make sure that this lock will not be found by raced
* bl_ast and -EINVAL reply is sent to server anyways.
- * b=17645*/
+ * b=17645
+ */
lock->l_flags |= LDLM_FL_LOCAL_ONLY | LDLM_FL_FAILED |
LDLM_FL_ATOMIC_CB | LDLM_FL_CBPENDING;
need_cancel = 1;
- }
- unlock_res_and_lock(lock);
-
- if (need_cancel)
- LDLM_DEBUG(lock,
- "setting FL_LOCAL_ONLY | LDLM_FL_FAILED | "
- "LDLM_FL_ATOMIC_CB | LDLM_FL_CBPENDING");
- else
- LDLM_DEBUG(lock, "lock was granted or failed in race");
-
- /* XXX - HACK because we shouldn't call ldlm_lock_destroy()
- * from llite/file.c/ll_file_flock(). */
- /* This code makes for the fact that we do not have blocking handler on
+ }
+ unlock_res_and_lock(lock);
+
+ if (need_cancel)
+ LDLM_DEBUG(lock,
+ "setting FL_LOCAL_ONLY | LDLM_FL_FAILED | LDLM_FL_ATOMIC_CB | LDLM_FL_CBPENDING");
+ else
+ LDLM_DEBUG(lock, "lock was granted or failed in race");
+
+ /*
+ * XXX - HACK because we shouldn't call ldlm_lock_destroy()
+ * from llite/file.c/ll_file_flock().
+ */
+ /*
+ * This code makes for the fact that we do not have blocking handler on
* a client for flock locks. As such this is the place where we must
* completely kill failed locks. (interrupted and those that
- * were waiting to be granted when server evicted us. */
+ * were waiting to be granted when server evicted us.
+ */
if (lock->l_resource->lr_type == LDLM_FLOCK) {
lock_res_and_lock(lock);
if (!ldlm_is_destroyed(lock)) {
* Called after receiving reply from server.
*/
int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
- ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
- __u64 *flags, void *lvb, __u32 lvb_len,
- struct lustre_handle *lockh,int rc)
+ enum ldlm_type type, __u8 with_policy,
+ enum ldlm_mode mode, __u64 *flags, void *lvb,
+ __u32 lvb_len, const struct lustre_handle *lockh,
+ int rc)
{
- struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
- int is_replay = *flags & LDLM_FL_REPLAY;
- struct ldlm_lock *lock;
- struct ldlm_reply *reply;
- int cleanup_phase = 1;
- int size = 0;
- ENTRY;
-
- lock = ldlm_handle2lock(lockh);
- /* ldlm_cli_enqueue is holding a reference on this lock. */
- if (!lock) {
- LASSERT(type == LDLM_FLOCK);
- RETURN(-ENOLCK);
- }
+ struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
+ const struct lu_env *env = NULL;
+ int is_replay = *flags & LDLM_FL_REPLAY;
+ struct ldlm_lock *lock;
+ struct ldlm_reply *reply;
+ int cleanup_phase = 1;
+
+ ENTRY;
+
+ if (req && req->rq_svc_thread)
+ env = req->rq_svc_thread->t_env;
+
+ lock = ldlm_handle2lock(lockh);
+ /* ldlm_cli_enqueue is holding a reference on this lock. */
+ if (!lock) {
+ LASSERT(type == LDLM_FLOCK);
+ RETURN(-ENOLCK);
+ }
LASSERTF(ergo(lvb_len != 0, lvb_len == lock->l_lvb_len),
"lvb_len = %d, l_lvb_len = %d\n", lvb_len, lock->l_lvb_len);
- if (rc != ELDLM_OK) {
- LASSERT(!is_replay);
- LDLM_DEBUG(lock, "client-side enqueue END (%s)",
- rc == ELDLM_LOCK_ABORTED ? "ABORTED" : "FAILED");
+ if (rc != ELDLM_OK) {
+ LASSERT(!is_replay);
+ LDLM_DEBUG(lock, "client-side enqueue END (%s)",
+ rc == ELDLM_LOCK_ABORTED ? "ABORTED" : "FAILED");
if (rc != ELDLM_LOCK_ABORTED)
GOTO(cleanup, rc);
if (reply == NULL)
GOTO(cleanup, rc = -EPROTO);
- if (lvb_len != 0) {
- LASSERT(lvb != NULL);
+ if (lvb_len > 0) {
+ int size = 0;
size = req_capsule_get_size(&req->rq_pill, &RMF_DLM_LVB,
RCL_SERVER);
LDLM_ERROR(lock, "Fail to get lvb_len, rc = %d", size);
GOTO(cleanup, rc = size);
} else if (unlikely(size > lvb_len)) {
- LDLM_ERROR(lock, "Replied LVB is larger than "
- "expectation, expected = %d, replied = %d",
+ LDLM_ERROR(lock,
+ "Replied LVB is larger than expectation, expected = %d, replied = %d",
lvb_len, size);
GOTO(cleanup, rc = -EINVAL);
}
+ lvb_len = size;
}
if (rc == ELDLM_LOCK_ABORTED) {
- if (lvb_len != 0)
+ if (lvb_len > 0 && lvb != NULL)
rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_SERVER,
- lvb, size);
- GOTO(cleanup, rc = (rc != 0 ? rc : ELDLM_LOCK_ABORTED));
+ lvb, lvb_len);
+ GOTO(cleanup, rc = rc ? : ELDLM_LOCK_ABORTED);
}
- /* lock enqueued on the server */
- cleanup_phase = 0;
+ /* lock enqueued on the server */
+ cleanup_phase = 0;
- lock_res_and_lock(lock);
- /* Key change rehash lock in per-export hash with new key */
- if (exp->exp_lock_hash) {
- /* In the function below, .hs_keycmp resolves to
- * ldlm_export_lock_keycmp() */
+ lock_res_and_lock(lock);
+ /* Key change rehash lock in per-export hash with new key */
+ if (exp->exp_lock_hash) {
+ /*
+ * In the function below, .hs_keycmp resolves to
+ * ldlm_export_lock_keycmp()
+ */
/* coverity[overrun-buffer-val] */
- cfs_hash_rehash_key(exp->exp_lock_hash,
- &lock->l_remote_handle,
- &reply->lock_handle,
- &lock->l_exp_hash);
- } else {
- lock->l_remote_handle = reply->lock_handle;
- }
+ cfs_hash_rehash_key(exp->exp_lock_hash,
+ &lock->l_remote_handle,
+ &reply->lock_handle,
+ &lock->l_exp_hash);
+ } else {
+ lock->l_remote_handle = reply->lock_handle;
+ }
*flags = ldlm_flags_from_wire(reply->lock_flags);
lock->l_flags |= ldlm_flags_from_wire(reply->lock_flags &
LDLM_FL_INHERIT_MASK);
- unlock_res_and_lock(lock);
+ unlock_res_and_lock(lock);
- CDEBUG(D_INFO, "local: %p, remote cookie: "LPX64", flags: "LPX64"\n",
+ CDEBUG(D_INFO, "local: %p, remote cookie: %#llx, flags: %#llx\n",
lock, reply->lock_handle.cookie, *flags);
- /* If enqueue returned a blocked lock but the completion handler has
+ /*
+ * If enqueue returned a blocked lock but the completion handler has
* already run, then it fixed up the resource and we don't need to do it
- * again. */
+ * again.
+ */
if ((*flags) & LDLM_FL_LOCK_CHANGED) {
int newmode = reply->lock_desc.l_req_mode;
+
LASSERT(!is_replay);
if (newmode && newmode != lock->l_req_mode) {
LDLM_DEBUG(lock, "server returned different mode %s",
if (!ldlm_res_eq(&reply->lock_desc.l_resource.lr_name,
&lock->l_resource->lr_name)) {
- CDEBUG(D_INFO, "remote intent success, locking "DLDLMRES
- " instead of "DLDLMRES"\n",
+ CDEBUG(D_INFO,
+ "remote intent success, locking "DLDLMRES", instead of "DLDLMRES"\n",
PLDLMRES(&reply->lock_desc.l_resource),
PLDLMRES(lock->l_resource));
GOTO(cleanup, rc = -ENOMEM);
LDLM_DEBUG(lock, "client-side enqueue, new resource");
}
- if (with_policy)
- if (!(type == LDLM_IBITS &&
- !(exp_connect_flags(exp) & OBD_CONNECT_IBITS)))
- /* We assume lock type cannot change on server*/
- ldlm_convert_policy_to_local(exp,
+
+ if (with_policy) {
+ /* We assume lock type cannot change on server*/
+ ldlm_convert_policy_to_local(exp,
lock->l_resource->lr_type,
&reply->lock_desc.l_policy_data,
&lock->l_policy_data);
- if (type != LDLM_PLAIN)
- LDLM_DEBUG(lock,"client-side enqueue, new policy data");
- }
-
- if ((*flags) & LDLM_FL_AST_SENT ||
- /* Cancel extent locks as soon as possible on a liblustre client,
- * because it cannot handle asynchronous ASTs robustly (see
- * bug 7311). */
- (LIBLUSTRE_CLIENT && type == LDLM_EXTENT)) {
- lock_res_and_lock(lock);
+ }
+
+ if (type != LDLM_PLAIN)
+ LDLM_DEBUG(lock,
+ "client-side enqueue, new policy data");
+ }
+
+ if ((*flags) & LDLM_FL_AST_SENT) {
+ lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_BL_AST;
- unlock_res_and_lock(lock);
- LDLM_DEBUG(lock, "enqueue reply includes blocking AST");
- }
-
- /* If the lock has already been granted by a completion AST, don't
- * clobber the LVB with an older one. */
- if (lvb_len != 0) {
- /* We must lock or a racing completion might update lvb without
+ unlock_res_and_lock(lock);
+ LDLM_DEBUG(lock, "enqueue reply includes blocking AST");
+ }
+
+ /*
+ * If the lock has already been granted by a completion AST, don't
+ * clobber the LVB with an older one.
+ */
+ if (lvb_len > 0) {
+ /*
+ * We must lock or a racing completion might update lvb without
* letting us know and we'll clobber the correct value.
* Cannot unlock after the check either, a that still leaves
- * a tiny window for completion to get in */
+ * a tiny window for completion to get in
+ */
lock_res_and_lock(lock);
- if (lock->l_req_mode != lock->l_granted_mode)
+ if (!ldlm_is_granted(lock))
rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_SERVER,
- lock->l_lvb_data, size);
+ lock->l_lvb_data, lvb_len);
unlock_res_and_lock(lock);
if (rc < 0) {
cleanup_phase = 1;
GOTO(cleanup, rc);
}
- }
-
- if (!is_replay) {
- rc = ldlm_lock_enqueue(ns, &lock, NULL, flags);
- if (lock->l_completion_ast != NULL) {
- int err = lock->l_completion_ast(lock, *flags, NULL);
- if (!rc)
- rc = err;
- if (rc)
- cleanup_phase = 1;
- }
- }
-
- if (lvb_len && lvb != NULL) {
- /* Copy the LVB here, and not earlier, because the completion
- * AST (if any) can override what we got in the reply */
- memcpy(lvb, lock->l_lvb_data, lvb_len);
- }
-
- LDLM_DEBUG(lock, "client-side enqueue END");
- EXIT;
+ }
+
+ if (!is_replay) {
+ rc = ldlm_lock_enqueue(env, ns, &lock, NULL, flags);
+ if (lock->l_completion_ast != NULL) {
+ int err = lock->l_completion_ast(lock, *flags, NULL);
+
+ if (!rc)
+ rc = err;
+ if (rc)
+ cleanup_phase = 1;
+ }
+ }
+
+ if (lvb_len > 0 && lvb != NULL) {
+ /*
+ * Copy the LVB here, and not earlier, because the completion
+ * AST (if any) can override what we got in the reply
+ */
+ memcpy(lvb, lock->l_lvb_data, lvb_len);
+ }
+
+ LDLM_DEBUG(lock, "client-side enqueue END");
+ EXIT;
cleanup:
- if (cleanup_phase == 1 && rc)
- failed_lock_cleanup(ns, lock, mode);
- /* Put lock 2 times, the second reference is held by ldlm_cli_enqueue */
- LDLM_LOCK_PUT(lock);
- LDLM_LOCK_RELEASE(lock);
- return rc;
+ if (cleanup_phase == 1 && rc)
+ failed_lock_cleanup(ns, lock, mode);
+ /* Put lock 2 times, the second reference is held by ldlm_cli_enqueue */
+ LDLM_LOCK_PUT(lock);
+ LDLM_LOCK_RELEASE(lock);
+ return rc;
}
EXPORT_SYMBOL(ldlm_cli_enqueue_fini);
*/
static inline int ldlm_req_handles_avail(int req_size, int off)
{
- int avail;
+ int avail;
- avail = min_t(int, LDLM_MAXREQSIZE, PAGE_CACHE_SIZE - 512) - req_size;
- if (likely(avail >= 0))
- avail /= (int)sizeof(struct lustre_handle);
- else
- avail = 0;
- avail += LDLM_LOCKREQ_HANDLES - off;
+ avail = min_t(int, LDLM_MAXREQSIZE, PAGE_SIZE - 512) - req_size;
+ if (likely(avail >= 0))
+ avail /= (int)sizeof(struct lustre_handle);
+ else
+ avail = 0;
+ avail += LDLM_LOCKREQ_HANDLES - off;
- return avail;
+ return avail;
}
static inline int ldlm_capsule_handles_avail(struct req_capsule *pill,
- enum req_location loc,
- int off)
+ enum req_location loc,
+ int off)
{
- int size = req_capsule_msg_size(pill, loc);
- return ldlm_req_handles_avail(size, off);
+ __u32 size = req_capsule_msg_size(pill, loc);
+
+ return ldlm_req_handles_avail(size, off);
}
static inline int ldlm_format_handles_avail(struct obd_import *imp,
- const struct req_format *fmt,
- enum req_location loc, int off)
+ const struct req_format *fmt,
+ enum req_location loc, int off)
{
- int size = req_capsule_fmt_size(imp->imp_msg_magic, fmt, loc);
- return ldlm_req_handles_avail(size, off);
+ __u32 size = req_capsule_fmt_size(imp->imp_msg_magic, fmt, loc);
+
+ return ldlm_req_handles_avail(size, off);
}
/**
int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
int version, int opc, int canceloff,
struct list_head *cancels, int count)
- {
+{
struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
struct req_capsule *pill = &req->rq_pill;
struct ldlm_request *dlm = NULL;
struct list_head head = LIST_HEAD_INIT(head);
- int flags, avail, to_free, pack = 0;
+ enum ldlm_lru_flags lru_flags;
+ int avail, to_free, pack = 0;
int rc;
+
ENTRY;
if (cancels == NULL)
cancels = &head;
if (ns_connect_cancelset(ns)) {
- /* Estimate the amount of available space in the request. */
- req_capsule_filled_sizes(pill, RCL_CLIENT);
- avail = ldlm_capsule_handles_avail(pill, RCL_CLIENT, canceloff);
+ /* Estimate the amount of available space in the request. */
+ req_capsule_filled_sizes(pill, RCL_CLIENT);
+ avail = ldlm_capsule_handles_avail(pill, RCL_CLIENT, canceloff);
- flags = ns_connect_lru_resize(ns) ?
- LDLM_CANCEL_LRUR : LDLM_CANCEL_AGED;
- to_free = !ns_connect_lru_resize(ns) &&
- opc == LDLM_ENQUEUE ? 1 : 0;
+ lru_flags = LDLM_LRU_FLAG_NO_WAIT | (ns_connect_lru_resize(ns) ?
+ LDLM_LRU_FLAG_LRUR : LDLM_LRU_FLAG_AGED);
+ to_free = !ns_connect_lru_resize(ns) &&
+ opc == LDLM_ENQUEUE ? 1 : 0;
- /* Cancel LRU locks here _only_ if the server supports
+ /*
+ * Cancel LRU locks here _only_ if the server supports
* EARLY_CANCEL. Otherwise we have to send extra CANCEL
- * RPC, which will make us slower. */
- if (avail > count)
- count += ldlm_cancel_lru_local(ns, cancels, to_free,
- avail - count, 0, flags);
- if (avail > count)
- pack = count;
- else
- pack = avail;
- req_capsule_set_size(pill, &RMF_DLM_REQ, RCL_CLIENT,
- ldlm_request_bufsize(pack, opc));
- }
-
- rc = ptlrpc_request_pack(req, version, opc);
- if (rc) {
- ldlm_lock_list_put(cancels, l_bl_ast, count);
- RETURN(rc);
- }
+ * RPC, which will make us slower.
+ */
+ if (avail > count)
+ count += ldlm_cancel_lru_local(ns, cancels, to_free,
+ avail - count, 0,
+ lru_flags);
+ if (avail > count)
+ pack = count;
+ else
+ pack = avail;
+ req_capsule_set_size(pill, &RMF_DLM_REQ, RCL_CLIENT,
+ ldlm_request_bufsize(pack, opc));
+ }
+
+ rc = ptlrpc_request_pack(req, version, opc);
+ if (rc) {
+ ldlm_lock_list_put(cancels, l_bl_ast, count);
+ RETURN(rc);
+ }
if (ns_connect_cancelset(ns)) {
- if (canceloff) {
- dlm = req_capsule_client_get(pill, &RMF_DLM_REQ);
- LASSERT(dlm);
- /* Skip first lock handler in ldlm_request_pack(),
- * this method will incrment @lock_count according
- * to the lock handle amount actually written to
- * the buffer. */
- dlm->lock_count = canceloff;
- }
- /* Pack into the request @pack lock handles. */
- ldlm_cli_cancel_list(cancels, pack, req, 0);
+ if (canceloff) {
+ dlm = req_capsule_client_get(pill, &RMF_DLM_REQ);
+ LASSERT(dlm);
+ /*
+ * Skip first lock handler in ldlm_request_pack(),
+ * this method will increment @lock_count according
+ * to the lock handle amount actually written to
+ * the buffer.
+ */
+ dlm->lock_count = canceloff;
+ }
+ /* Pack into the request @pack lock handles. */
+ ldlm_cli_cancel_list(cancels, pack, req, 0);
/* Prepare and send separate cancel RPC for others. */
- ldlm_cli_cancel_list(cancels, count - pack, NULL, 0);
- } else {
- ldlm_lock_list_put(cancels, l_bl_ast, count);
- }
- RETURN(0);
+ ldlm_cli_cancel_list(cancels, count - pack, NULL, 0);
+ } else {
+ ldlm_lock_list_put(cancels, l_bl_ast, count);
+ }
+ RETURN(0);
}
EXPORT_SYMBOL(ldlm_prep_elc_req);
int ldlm_prep_enqueue_req(struct obd_export *exp, struct ptlrpc_request *req,
- cfs_list_t *cancels, int count)
+ struct list_head *cancels, int count)
{
- return ldlm_prep_elc_req(exp, req, LUSTRE_DLM_VERSION, LDLM_ENQUEUE,
- LDLM_ENQUEUE_CANCEL_OFF, cancels, count);
+ return ldlm_prep_elc_req(exp, req, LUSTRE_DLM_VERSION, LDLM_ENQUEUE,
+ LDLM_ENQUEUE_CANCEL_OFF, cancels, count);
}
EXPORT_SYMBOL(ldlm_prep_enqueue_req);
{
struct ptlrpc_request *req;
int rc;
+
ENTRY;
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
* pass it to the caller in \a reqp.
*/
int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
- struct ldlm_enqueue_info *einfo,
- const struct ldlm_res_id *res_id,
- ldlm_policy_data_t const *policy, __u64 *flags,
+ struct ldlm_enqueue_info *einfo,
+ const struct ldlm_res_id *res_id,
+ union ldlm_policy_data const *policy, __u64 *flags,
void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
struct lustre_handle *lockh, int async)
{
struct ldlm_namespace *ns;
- struct ldlm_lock *lock;
- struct ldlm_request *body;
- int is_replay = *flags & LDLM_FL_REPLAY;
- int req_passed_in = 1;
- int rc, err;
- struct ptlrpc_request *req;
- ENTRY;
+ struct ldlm_lock *lock;
+ struct ldlm_request *body;
+ int is_replay = *flags & LDLM_FL_REPLAY;
+ int req_passed_in = 1;
+ int rc, err;
+ struct ptlrpc_request *req;
- LASSERT(exp != NULL);
+ ENTRY;
+
+ LASSERT(exp != NULL);
ns = exp->exp_obd->obd_namespace;
- /* If we're replaying this lock, just check some invariants.
- * If we're creating a new lock, get everything all setup nice. */
- if (is_replay) {
- lock = ldlm_handle2lock_long(lockh, 0);
- LASSERT(lock != NULL);
- LDLM_DEBUG(lock, "client-side enqueue START");
- LASSERT(exp == lock->l_conn_export);
- } else {
+ /*
+ * If we're replaying this lock, just check some invariants.
+ * If we're creating a new lock, get everything all setup nice.
+ */
+ if (is_replay) {
+ lock = ldlm_handle2lock_long(lockh, 0);
+ LASSERT(lock != NULL);
+ LDLM_DEBUG(lock, "client-side enqueue START");
+ LASSERT(exp == lock->l_conn_export);
+ } else {
const struct ldlm_callback_suite cbs = {
.lcs_completion = einfo->ei_cb_cp,
.lcs_blocking = einfo->ei_cb_bl,
lvb_len, lvb_type);
if (IS_ERR(lock))
RETURN(PTR_ERR(lock));
- /* for the local lock, add the reference */
- ldlm_lock_addref_internal(lock, einfo->ei_mode);
- ldlm_lock2handle(lock, lockh);
+
+ if (einfo->ei_cb_created)
+ einfo->ei_cb_created(lock);
+
+ /* for the local lock, add the reference */
+ ldlm_lock_addref_internal(lock, einfo->ei_mode);
+ ldlm_lock2handle(lock, lockh);
if (policy != NULL)
lock->l_policy_data = *policy;
lock->l_req_extent = policy->l_extent;
}
- LDLM_DEBUG(lock, "client-side enqueue START, flags "LPX64"\n",
+ LDLM_DEBUG(lock, "client-side enqueue START, flags %#llx",
*flags);
}
lock->l_export = NULL;
lock->l_blocking_ast = einfo->ei_cb_bl;
lock->l_flags |= (*flags & (LDLM_FL_NO_LRU | LDLM_FL_EXCL));
+ lock->l_activity = ktime_get_real_seconds();
- /* lock not sent to server yet */
-
- if (reqp == NULL || *reqp == NULL) {
- req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
- &RQF_LDLM_ENQUEUE,
- LUSTRE_DLM_VERSION,
- LDLM_ENQUEUE);
- if (req == NULL) {
- failed_lock_cleanup(ns, lock, einfo->ei_mode);
- LDLM_LOCK_RELEASE(lock);
- RETURN(-ENOMEM);
- }
- req_passed_in = 0;
- if (reqp)
- *reqp = req;
- } else {
- int len;
-
- req = *reqp;
- len = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ,
- RCL_CLIENT);
- LASSERTF(len >= sizeof(*body), "buflen[%d] = %d, not %d\n",
- DLM_LOCKREQ_OFF, len, (int)sizeof(*body));
- }
-
- /* Dump lock data into the request buffer */
- body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
- ldlm_lock2desc(lock, &body->lock_desc);
+ /* lock not sent to server yet */
+ if (reqp == NULL || *reqp == NULL) {
+ req = ldlm_enqueue_pack(exp, lvb_len);
+ if (IS_ERR(req)) {
+ failed_lock_cleanup(ns, lock, einfo->ei_mode);
+ LDLM_LOCK_RELEASE(lock);
+ RETURN(PTR_ERR(req));
+ }
+
+ req_passed_in = 0;
+ if (reqp)
+ *reqp = req;
+ } else {
+ int len;
+
+ req = *reqp;
+ len = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ,
+ RCL_CLIENT);
+ LASSERTF(len >= sizeof(*body), "buflen[%d] = %d, not %d\n",
+ DLM_LOCKREQ_OFF, len, (int)sizeof(*body));
+ }
+
+ if (*flags & LDLM_FL_NDELAY) {
+ DEBUG_REQ(D_DLMTRACE, req, "enque lock with no delay\n");
+ req->rq_no_resend = req->rq_no_delay = 1;
+ /*
+ * probably set a shorter timeout value and handle ETIMEDOUT
+ * in osc_lock_upcall() correctly
+ */
+ /* lustre_msg_set_timeout(req, req->rq_timeout / 2); */
+ }
+
+ /* Dump lock data into the request buffer */
+ body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
+ ldlm_lock2desc(lock, &body->lock_desc);
body->lock_flags = ldlm_flags_to_wire(*flags);
- body->lock_handle[0] = *lockh;
-
- /* Continue as normal. */
- if (!req_passed_in) {
- if (lvb_len > 0)
- req_capsule_extend(&req->rq_pill,
- &RQF_LDLM_ENQUEUE_LVB);
- req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
- lvb_len);
- ptlrpc_request_set_replen(req);
- }
-
- /*
- * Liblustre client doesn't get extent locks, except for O_APPEND case
- * where [0, OBD_OBJECT_EOF] lock is taken, or truncate, where
- * [i_size, OBD_OBJECT_EOF] lock is taken.
- */
- LASSERT(ergo(LIBLUSTRE_CLIENT, einfo->ei_type != LDLM_EXTENT ||
- policy->l_extent.end == OBD_OBJECT_EOF));
-
- if (async) {
- LASSERT(reqp != NULL);
- RETURN(0);
- }
-
- LDLM_DEBUG(lock, "sending request");
-
- rc = ptlrpc_queue_wait(req);
-
- err = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, policy ? 1 : 0,
- einfo->ei_mode, flags, lvb, lvb_len,
- lockh, rc);
-
- /* If ldlm_cli_enqueue_fini did not find the lock, we need to free
- * one reference that we took */
- if (err == -ENOLCK)
- LDLM_LOCK_RELEASE(lock);
- else
- rc = err;
-
- if (!req_passed_in && req != NULL) {
- ptlrpc_req_finished(req);
- if (reqp)
- *reqp = NULL;
- }
-
- RETURN(rc);
+ body->lock_handle[0] = *lockh;
+
+ /* extended LDLM opcodes in client stats */
+ if (exp->exp_obd->obd_svc_stats != NULL) {
+ bool glimpse = *flags & LDLM_FL_HAS_INTENT;
+
+ /* OST glimpse has no intent buffer */
+ if (req_capsule_has_field(&req->rq_pill, &RMF_LDLM_INTENT,
+ RCL_CLIENT)) {
+ struct ldlm_intent *it;
+
+ it = req_capsule_client_get(&req->rq_pill,
+ &RMF_LDLM_INTENT);
+ glimpse = (it && (it->opc == IT_GLIMPSE));
+ }
+
+ if (!glimpse)
+ ldlm_svc_get_eopc(body, exp->exp_obd->obd_svc_stats);
+ else
+ lprocfs_counter_incr(exp->exp_obd->obd_svc_stats,
+ PTLRPC_LAST_CNTR +
+ LDLM_GLIMPSE_ENQUEUE);
+ }
+
+ if (async) {
+ LASSERT(reqp != NULL);
+ RETURN(0);
+ }
+
+ LDLM_DEBUG(lock, "sending request");
+
+ rc = ptlrpc_queue_wait(req);
+
+ err = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, policy ? 1 : 0,
+ einfo->ei_mode, flags, lvb, lvb_len,
+ lockh, rc);
+
+ /*
+ * If ldlm_cli_enqueue_fini did not find the lock, we need to free
+ * one reference that we took
+ */
+ if (err == -ENOLCK)
+ LDLM_LOCK_RELEASE(lock);
+ else
+ rc = err;
+
+ if (!req_passed_in && req != NULL) {
+ ptlrpc_req_finished(req);
+ if (reqp)
+ *reqp = NULL;
+ }
+
+ RETURN(rc);
}
EXPORT_SYMBOL(ldlm_cli_enqueue);
-static int ldlm_cli_convert_local(struct ldlm_lock *lock, int new_mode,
- __u32 *flags)
+/**
+ * Client-side lock convert reply handling.
+ *
+ * Finish client lock converting, checks for concurrent converts
+ * and clear 'converting' flag so lock can be placed back into LRU.
+ */
+static int lock_convert_interpret(const struct lu_env *env,
+ struct ptlrpc_request *req,
+ void *args, int rc)
{
- struct ldlm_resource *res;
- int rc;
- ENTRY;
- if (ns_is_client(ldlm_lock_to_ns(lock))) {
- CERROR("Trying to cancel local lock\n");
- LBUG();
- }
- LDLM_DEBUG(lock, "client-side local convert");
-
- res = ldlm_lock_convert(lock, new_mode, flags);
- if (res) {
- ldlm_reprocess_all(res);
- rc = 0;
- } else {
- rc = LUSTRE_EDEADLK;
- }
- LDLM_DEBUG(lock, "client-side local convert handler END");
- LDLM_LOCK_PUT(lock);
- RETURN(rc);
+ struct ldlm_async_args *aa = args;
+ struct ldlm_lock *lock;
+ struct ldlm_reply *reply;
+
+ ENTRY;
+
+ lock = ldlm_handle2lock(&aa->lock_handle);
+ if (!lock) {
+ LDLM_DEBUG_NOLOCK("convert ACK for unknown local cookie %#llx",
+ aa->lock_handle.cookie);
+ RETURN(-ESTALE);
+ }
+
+ LDLM_DEBUG(lock, "CONVERTED lock:");
+
+ if (rc != ELDLM_OK)
+ GOTO(out, rc);
+
+ reply = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
+ if (reply == NULL)
+ GOTO(out, rc = -EPROTO);
+
+ if (reply->lock_handle.cookie != aa->lock_handle.cookie) {
+ LDLM_ERROR(lock,
+ "convert ACK with wrong lock cookie %#llx but cookie %#llx from server %s id %s\n",
+ aa->lock_handle.cookie, reply->lock_handle.cookie,
+ req->rq_export->exp_client_uuid.uuid,
+ libcfs_id2str(req->rq_peer));
+ GOTO(out, rc = ELDLM_NO_LOCK_DATA);
+ }
+
+ lock_res_and_lock(lock);
+ /*
+ * Lock convert is sent for any new bits to drop, the converting flag
+ * is dropped when ibits on server are the same as on client. Meanwhile
+ * that can be so that more later convert will be replied first with
+ * and clear converting flag, so in case of such race just exit here.
+ * if lock has no converting bits then
+ */
+ if (!ldlm_is_converting(lock)) {
+ LDLM_DEBUG(lock,
+ "convert ACK for lock without converting flag, reply ibits %#llx",
+ reply->lock_desc.l_policy_data.l_inodebits.bits);
+ } else if (reply->lock_desc.l_policy_data.l_inodebits.bits !=
+ lock->l_policy_data.l_inodebits.bits) {
+ /*
+ * Compare server returned lock ibits and local lock ibits
+ * if they are the same we consider convertion is done,
+ * otherwise we have more converts inflight and keep
+ * converting flag.
+ */
+ LDLM_DEBUG(lock, "convert ACK with ibits %#llx\n",
+ reply->lock_desc.l_policy_data.l_inodebits.bits);
+ } else {
+ ldlm_clear_converting(lock);
+
+ /*
+ * Concurrent BL AST may arrive and cause another convert
+ * or cancel so just do nothing here if bl_ast is set,
+ * finish with convert otherwise.
+ */
+ if (!ldlm_is_bl_ast(lock)) {
+ struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
+
+ /*
+ * Drop cancel_bits since there are no more converts
+ * and put lock into LRU if it is still not used and
+ * is not there yet.
+ */
+ lock->l_policy_data.l_inodebits.cancel_bits = 0;
+ if (!lock->l_readers && !lock->l_writers &&
+ !ldlm_is_canceling(lock)) {
+ spin_lock(&ns->ns_lock);
+ /* there is check for list_empty() inside */
+ ldlm_lock_remove_from_lru_nolock(lock);
+ ldlm_lock_add_to_lru_nolock(lock);
+ spin_unlock(&ns->ns_lock);
+ }
+ }
+ }
+ unlock_res_and_lock(lock);
+out:
+ if (rc) {
+ int flag;
+
+ lock_res_and_lock(lock);
+ if (ldlm_is_converting(lock)) {
+ ldlm_clear_converting(lock);
+ ldlm_set_cbpending(lock);
+ ldlm_set_bl_ast(lock);
+ lock->l_policy_data.l_inodebits.cancel_bits = 0;
+ }
+ unlock_res_and_lock(lock);
+
+ /*
+ * fallback to normal lock cancel. If rc means there is no
+ * valid lock on server, do only local cancel
+ */
+ if (rc == ELDLM_NO_LOCK_DATA)
+ flag = LCF_LOCAL;
+ else
+ flag = LCF_ASYNC;
+
+ rc = ldlm_cli_cancel(&aa->lock_handle, flag);
+ if (rc < 0)
+ LDLM_DEBUG(lock, "failed to cancel lock: rc = %d\n",
+ rc);
+ }
+ LDLM_LOCK_PUT(lock);
+ RETURN(rc);
}
-/* FIXME: one of ldlm_cli_convert or the server side should reject attempted
- * conversion of locks which are on the waiting or converting queue */
-/* Caller of this code is supposed to take care of lock readers/writers
- accounting */
-int ldlm_cli_convert(struct lustre_handle *lockh, int new_mode, __u32 *flags)
+/**
+ * Client-side IBITS lock convert.
+ *
+ * Inform server that lock has been converted instead of canceling.
+ * Server finishes convert on own side and does reprocess to grant
+ * all related waiting locks.
+ *
+ * Since convert means only ibits downgrading, client doesn't need to
+ * wait for server reply to finish local converting process so this request
+ * is made asynchronous.
+ *
+ */
+int ldlm_cli_convert(struct ldlm_lock *lock, __u32 *flags)
{
- struct ldlm_request *body;
- struct ldlm_reply *reply;
- struct ldlm_lock *lock;
- struct ldlm_resource *res;
- struct ptlrpc_request *req;
- int rc;
- ENTRY;
-
- lock = ldlm_handle2lock(lockh);
- if (!lock) {
- LBUG();
- RETURN(-EINVAL);
- }
- *flags = 0;
-
- if (lock->l_conn_export == NULL)
- RETURN(ldlm_cli_convert_local(lock, new_mode, flags));
-
- LDLM_DEBUG(lock, "client-side convert");
-
- req = ptlrpc_request_alloc_pack(class_exp2cliimp(lock->l_conn_export),
- &RQF_LDLM_CONVERT, LUSTRE_DLM_VERSION,
- LDLM_CONVERT);
- if (req == NULL) {
- LDLM_LOCK_PUT(lock);
- RETURN(-ENOMEM);
- }
-
- body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
- body->lock_handle[0] = lock->l_remote_handle;
-
- body->lock_desc.l_req_mode = new_mode;
+ struct ldlm_request *body;
+ struct ptlrpc_request *req;
+ struct ldlm_async_args *aa;
+ struct obd_export *exp = lock->l_conn_export;
+
+ ENTRY;
+
+ if (exp == NULL) {
+ LDLM_ERROR(lock, "convert must not be called on local locks.");
+ RETURN(-EINVAL);
+ }
+
+ /*
+ * this is better to check earlier and it is done so already,
+ * but this check is kept too as final one to issue an error
+ * if any new code will miss such check.
+ */
+ if (!exp_connect_lock_convert(exp)) {
+ LDLM_ERROR(lock, "server doesn't support lock convert\n");
+ RETURN(-EPROTO);
+ }
+
+ if (lock->l_resource->lr_type != LDLM_IBITS) {
+ LDLM_ERROR(lock, "convert works with IBITS locks only.");
+ RETURN(-EINVAL);
+ }
+
+ LDLM_DEBUG(lock, "client-side convert");
+
+ req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+ &RQF_LDLM_CONVERT, LUSTRE_DLM_VERSION,
+ LDLM_CONVERT);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
+ body->lock_handle[0] = lock->l_remote_handle;
+
+ body->lock_desc.l_req_mode = lock->l_req_mode;
+ body->lock_desc.l_granted_mode = lock->l_granted_mode;
+
+ body->lock_desc.l_policy_data.l_inodebits.bits =
+ lock->l_policy_data.l_inodebits.bits;
+ body->lock_desc.l_policy_data.l_inodebits.cancel_bits = 0;
+
body->lock_flags = ldlm_flags_to_wire(*flags);
+ body->lock_count = 1;
+
+ ptlrpc_request_set_replen(req);
+ /*
+ * Use cancel portals for convert as well as high-priority handling.
+ */
+ req->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
+ req->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
- ptlrpc_request_set_replen(req);
- rc = ptlrpc_queue_wait(req);
- if (rc != ELDLM_OK)
- GOTO(out, rc);
-
- reply = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
- if (reply == NULL)
- GOTO(out, rc = -EPROTO);
-
- if (req->rq_status)
- GOTO(out, rc = req->rq_status);
-
- res = ldlm_lock_convert(lock, new_mode, &reply->lock_flags);
- if (res != NULL) {
- ldlm_reprocess_all(res);
- /* Go to sleep until the lock is granted. */
- /* FIXME: or cancelled. */
- if (lock->l_completion_ast) {
- rc = lock->l_completion_ast(lock, LDLM_FL_WAIT_NOREPROC,
- NULL);
- if (rc)
- GOTO(out, rc);
- }
- } else {
- rc = LUSTRE_EDEADLK;
- }
- EXIT;
- out:
- LDLM_LOCK_PUT(lock);
- ptlrpc_req_finished(req);
- return rc;
+ ptlrpc_at_set_req_timeout(req);
+
+ if (exp->exp_obd->obd_svc_stats != NULL)
+ lprocfs_counter_incr(exp->exp_obd->obd_svc_stats,
+ LDLM_CONVERT - LDLM_FIRST_OPC);
+
+ aa = ptlrpc_req_async_args(req);
+ ldlm_lock2handle(lock, &aa->lock_handle);
+ req->rq_interpret_reply = lock_convert_interpret;
+
+ ptlrpcd_add_req(req);
+ RETURN(0);
}
-EXPORT_SYMBOL(ldlm_cli_convert);
/**
* Cancel locks locally.
static __u64 ldlm_cli_cancel_local(struct ldlm_lock *lock)
{
__u64 rc = LDLM_FL_LOCAL_ONLY;
- ENTRY;
- if (lock->l_conn_export) {
- bool local_only;
+ ENTRY;
+
+ if (lock->l_conn_export) {
+ bool local_only;
- LDLM_DEBUG(lock, "client-side cancel");
- /* Set this flag to prevent others from getting new references*/
- lock_res_and_lock(lock);
+ LDLM_DEBUG(lock, "client-side cancel");
+ /* Set this flag to prevent others from getting new references*/
+ lock_res_and_lock(lock);
ldlm_set_cbpending(lock);
local_only = !!(lock->l_flags &
(LDLM_FL_LOCAL_ONLY|LDLM_FL_CANCEL_ON_BLOCK));
LDLM_FL_BL_AST : LDLM_FL_CANCELING;
unlock_res_and_lock(lock);
- if (local_only) {
- CDEBUG(D_DLMTRACE, "not sending request (at caller's "
- "instruction)\n");
- rc = LDLM_FL_LOCAL_ONLY;
- }
- ldlm_lock_cancel(lock);
- } else {
- if (ns_is_client(ldlm_lock_to_ns(lock))) {
- LDLM_ERROR(lock, "Trying to cancel local lock");
- LBUG();
- }
- LDLM_DEBUG(lock, "server-side local cancel");
- ldlm_lock_cancel(lock);
- ldlm_reprocess_all(lock->l_resource);
- }
-
- RETURN(rc);
+ if (local_only) {
+ CDEBUG(D_DLMTRACE,
+ "not sending request (at caller's instruction)\n");
+ rc = LDLM_FL_LOCAL_ONLY;
+ }
+ ldlm_lock_cancel(lock);
+ } else {
+ if (ns_is_client(ldlm_lock_to_ns(lock))) {
+ LDLM_ERROR(lock, "Trying to cancel local lock");
+ LBUG();
+ }
+ LDLM_DEBUG(lock, "server-side local cancel");
+ ldlm_lock_cancel(lock);
+ ldlm_reprocess_all(lock->l_resource, lock);
+ }
+
+ RETURN(rc);
}
/**
* Pack \a count locks in \a head into ldlm_request buffer of request \a req.
*/
static void ldlm_cancel_pack(struct ptlrpc_request *req,
- cfs_list_t *head, int count)
+ struct list_head *head, int count)
{
- struct ldlm_request *dlm;
- struct ldlm_lock *lock;
- int max, packed = 0;
- ENTRY;
-
- dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
- LASSERT(dlm != NULL);
-
- /* Check the room in the request buffer. */
- max = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT) -
- sizeof(struct ldlm_request);
- max /= sizeof(struct lustre_handle);
- max += LDLM_LOCKREQ_HANDLES;
- LASSERT(max >= dlm->lock_count + count);
-
- /* XXX: it would be better to pack lock handles grouped by resource.
- * so that the server cancel would call filter_lvbo_update() less
- * frequently. */
- cfs_list_for_each_entry(lock, head, l_bl_ast) {
- if (!count--)
- break;
- LASSERT(lock->l_conn_export);
- /* Pack the lock handle to the given request buffer. */
- LDLM_DEBUG(lock, "packing");
- dlm->lock_handle[dlm->lock_count++] = lock->l_remote_handle;
- packed++;
- }
- CDEBUG(D_DLMTRACE, "%d locks packed\n", packed);
- EXIT;
+ struct ldlm_request *dlm;
+ struct ldlm_lock *lock;
+ int max, packed = 0;
+
+ ENTRY;
+
+ dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
+ LASSERT(dlm != NULL);
+
+ /* Check the room in the request buffer. */
+ max = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT) -
+ sizeof(struct ldlm_request);
+ max /= sizeof(struct lustre_handle);
+ max += LDLM_LOCKREQ_HANDLES;
+ LASSERT(max >= dlm->lock_count + count);
+
+ /*
+ * XXX: it would be better to pack lock handles grouped by resource.
+ * so that the server cancel would call filter_lvbo_update() less
+ * frequently.
+ */
+ list_for_each_entry(lock, head, l_bl_ast) {
+ if (!count--)
+ break;
+ LASSERT(lock->l_conn_export);
+ /* Pack the lock handle to the given request buffer. */
+ LDLM_DEBUG(lock, "packing");
+ dlm->lock_handle[dlm->lock_count++] = lock->l_remote_handle;
+ packed++;
+ }
+ CDEBUG(D_DLMTRACE, "%d locks packed\n", packed);
+ EXIT;
}
/**
* Prepare and send a batched cancel RPC. It will include \a count lock
- * handles of locks given in \a cancels list. */
-int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *cancels,
- int count, ldlm_cancel_flags_t flags)
+ * handles of locks given in \a cancels list.
+ */
+int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *cancels,
+ int count, enum ldlm_cancel_flags flags)
{
- struct ptlrpc_request *req = NULL;
- struct obd_import *imp;
- int free, sent = 0;
- int rc = 0;
- ENTRY;
-
- LASSERT(exp != NULL);
- LASSERT(count > 0);
-
- CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_PAUSE_CANCEL, cfs_fail_val);
-
- if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_RACE))
- RETURN(count);
-
- free = ldlm_format_handles_avail(class_exp2cliimp(exp),
- &RQF_LDLM_CANCEL, RCL_CLIENT, 0);
- if (count > free)
- count = free;
-
- while (1) {
- imp = class_exp2cliimp(exp);
- if (imp == NULL || imp->imp_invalid) {
- CDEBUG(D_DLMTRACE,
- "skipping cancel on invalid import %p\n", imp);
- RETURN(count);
- }
-
- req = ptlrpc_request_alloc(imp, &RQF_LDLM_CANCEL);
- if (req == NULL)
- GOTO(out, rc = -ENOMEM);
-
- req_capsule_filled_sizes(&req->rq_pill, RCL_CLIENT);
- req_capsule_set_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT,
- ldlm_request_bufsize(count, LDLM_CANCEL));
-
- rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_CANCEL);
- if (rc) {
- ptlrpc_request_free(req);
- GOTO(out, rc);
- }
-
- req->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
- req->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
- ptlrpc_at_set_req_timeout(req);
-
- ldlm_cancel_pack(req, cancels, count);
-
- ptlrpc_request_set_replen(req);
- if (flags & LCF_ASYNC) {
- ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
- sent = count;
- GOTO(out, 0);
- } else {
- rc = ptlrpc_queue_wait(req);
- }
+ struct ptlrpc_request *req = NULL;
+ struct obd_import *imp;
+ int free, sent = 0;
+ int rc = 0;
+
+ ENTRY;
+
+ LASSERT(exp != NULL);
+ LASSERT(count > 0);
+
+ CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_PAUSE_CANCEL, cfs_fail_val);
+
+ if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_RACE))
+ RETURN(count);
+
+ free = ldlm_format_handles_avail(class_exp2cliimp(exp),
+ &RQF_LDLM_CANCEL, RCL_CLIENT, 0);
+ if (count > free)
+ count = free;
+
+ while (1) {
+ imp = class_exp2cliimp(exp);
+ if (imp == NULL || imp->imp_invalid) {
+ CDEBUG(D_DLMTRACE,
+ "skipping cancel on invalid import %p\n", imp);
+ RETURN(count);
+ }
+
+ req = ptlrpc_request_alloc(imp, &RQF_LDLM_CANCEL);
+ if (req == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ req_capsule_filled_sizes(&req->rq_pill, RCL_CLIENT);
+ req_capsule_set_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT,
+ ldlm_request_bufsize(count, LDLM_CANCEL));
+
+ rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_CANCEL);
+ if (rc) {
+ ptlrpc_request_free(req);
+ GOTO(out, rc);
+ }
+
+ /*
+ * If OSP want cancel cross-MDT lock, let's not block it in
+ * in recovery, otherwise the lock will not released, if
+ * the remote target is also in recovery, and it also need
+ * this lock, it might cause deadlock.
+ */
+ if (exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS &&
+ exp->exp_obd->obd_lu_dev != NULL &&
+ exp->exp_obd->obd_lu_dev->ld_site != NULL) {
+ struct lu_device *top_dev;
+
+ top_dev = exp->exp_obd->obd_lu_dev->ld_site->ls_top_dev;
+ if (top_dev != NULL &&
+ top_dev->ld_obd->obd_recovering)
+ req->rq_allow_replay = 1;
+ }
+
+ req->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
+ req->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
+ ptlrpc_at_set_req_timeout(req);
+
+ ldlm_cancel_pack(req, cancels, count);
+
+ ptlrpc_request_set_replen(req);
+ if (flags & LCF_ASYNC) {
+ ptlrpcd_add_req(req);
+ sent = count;
+ GOTO(out, 0);
+ }
+
+ rc = ptlrpc_queue_wait(req);
if (rc == LUSTRE_ESTALE) {
- CDEBUG(D_DLMTRACE, "client/server (nid %s) "
- "out of sync -- not fatal\n",
- libcfs_nid2str(req->rq_import->
- imp_connection->c_peer.nid));
- rc = 0;
- } else if (rc == -ETIMEDOUT && /* check there was no reconnect*/
- req->rq_import_generation == imp->imp_generation) {
- ptlrpc_req_finished(req);
- continue;
- } else if (rc != ELDLM_OK) {
+ CDEBUG(D_DLMTRACE,
+ "client/server (nid %s) out of sync -- not fatal\n",
+ libcfs_nid2str(req->rq_import->imp_connection->c_peer.nid));
+ rc = 0;
+ } else if (rc == -ETIMEDOUT && /* check there was no reconnect*/
+ req->rq_import_generation == imp->imp_generation) {
+ ptlrpc_req_finished(req);
+ continue;
+ } else if (rc != ELDLM_OK) {
/* -ESHUTDOWN is common on umount */
CDEBUG_LIMIT(rc == -ESHUTDOWN ? D_DLMTRACE : D_ERROR,
- "Got rc %d from cancel RPC: "
- "canceling anyway\n", rc);
+ "Got rc %d from cancel RPC: canceling anyway\n",
+ rc);
break;
- }
- sent = count;
- break;
- }
+ }
+ sent = count;
+ break;
+ }
- ptlrpc_req_finished(req);
- EXIT;
+ ptlrpc_req_finished(req);
+ EXIT;
out:
- return sent ? sent : rc;
+ return sent ? sent : rc;
}
-EXPORT_SYMBOL(ldlm_cli_cancel_req);
static inline struct ldlm_pool *ldlm_imp2pl(struct obd_import *imp)
{
- LASSERT(imp != NULL);
- return &imp->imp_obd->obd_namespace->ns_pool;
+ LASSERT(imp != NULL);
+ return &imp->imp_obd->obd_namespace->ns_pool;
}
/**
*/
int ldlm_cli_update_pool(struct ptlrpc_request *req)
{
- struct obd_device *obd;
- __u64 new_slv;
- __u32 new_limit;
- ENTRY;
- if (unlikely(!req->rq_import || !req->rq_import->imp_obd ||
- !imp_connect_lru_resize(req->rq_import)))
- {
- /*
- * Do nothing for corner cases.
- */
- RETURN(0);
- }
-
- /* In some cases RPC may contain SLV and limit zeroed out. This
+ struct obd_device *obd;
+ __u64 new_slv;
+ __u32 new_limit;
+
+ ENTRY;
+ if (unlikely(!req->rq_import || !req->rq_import->imp_obd ||
+ !imp_connect_lru_resize(req->rq_import)))
+ /* Do nothing for corner cases. */
+ RETURN(0);
+
+ /*
+ * In some cases RPC may contain SLV and limit zeroed out. This
* is the case when server does not support LRU resize feature.
* This is also possible in some recovery cases when server-side
* reqs have no reference to the OBD export and thus access to
- * server-side namespace is not possible. */
- if (lustre_msg_get_slv(req->rq_repmsg) == 0 ||
- lustre_msg_get_limit(req->rq_repmsg) == 0) {
- DEBUG_REQ(D_HA, req, "Zero SLV or Limit found "
- "(SLV: "LPU64", Limit: %u)",
- lustre_msg_get_slv(req->rq_repmsg),
- lustre_msg_get_limit(req->rq_repmsg));
- RETURN(0);
- }
-
- new_limit = lustre_msg_get_limit(req->rq_repmsg);
- new_slv = lustre_msg_get_slv(req->rq_repmsg);
- obd = req->rq_import->imp_obd;
-
- /* Set new SLV and limit in OBD fields to make them accessible
+ * server-side namespace is not possible.
+ */
+ if (lustre_msg_get_slv(req->rq_repmsg) == 0 ||
+ lustre_msg_get_limit(req->rq_repmsg) == 0) {
+ DEBUG_REQ(D_HA, req,
+ "Zero SLV or Limit found (SLV: %llu, Limit: %u)",
+ lustre_msg_get_slv(req->rq_repmsg),
+ lustre_msg_get_limit(req->rq_repmsg));
+ RETURN(0);
+ }
+
+ new_limit = lustre_msg_get_limit(req->rq_repmsg);
+ new_slv = lustre_msg_get_slv(req->rq_repmsg);
+ obd = req->rq_import->imp_obd;
+
+ /*
+ * Set new SLV and limit in OBD fields to make them accessible
* to the pool thread. We do not access obd_namespace and pool
* directly here as there is no reliable way to make sure that
* they are still alive at cleanup time. Evil races are possible
- * which may cause Oops at that time. */
+ * which may cause Oops at that time.
+ */
write_lock(&obd->obd_pool_lock);
- obd->obd_pool_slv = new_slv;
- obd->obd_pool_limit = new_limit;
+ obd->obd_pool_slv = new_slv;
+ obd->obd_pool_limit = new_limit;
write_unlock(&obd->obd_pool_lock);
- RETURN(0);
+ RETURN(0);
}
-EXPORT_SYMBOL(ldlm_cli_update_pool);
/**
* Client side lock cancel.
*
* Lock must not have any readers or writers by this time.
*/
-int ldlm_cli_cancel(struct lustre_handle *lockh,
- ldlm_cancel_flags_t cancel_flags)
+int ldlm_cli_cancel(const struct lustre_handle *lockh,
+ enum ldlm_cancel_flags cancel_flags)
{
struct obd_export *exp;
- int avail, flags, count = 1;
+ enum ldlm_lru_flags lru_flags;
+ int avail, count = 1;
__u64 rc = 0;
struct ldlm_namespace *ns;
struct ldlm_lock *lock;
struct list_head cancels = LIST_HEAD_INIT(cancels);
+
ENTRY;
- /* concurrent cancels on the same handle can happen */
- lock = ldlm_handle2lock_long(lockh, LDLM_FL_CANCELING);
+ lock = ldlm_handle2lock_long(lockh, 0);
if (lock == NULL) {
LDLM_DEBUG_NOLOCK("lock is already being destroyed");
RETURN(0);
}
+ /* Convert lock bits instead of cancel for IBITS locks */
+ if (cancel_flags & LCF_CONVERT) {
+ LASSERT(lock->l_resource->lr_type == LDLM_IBITS);
+ LASSERT(lock->l_policy_data.l_inodebits.cancel_bits != 0);
+
+ rc = ldlm_cli_dropbits(lock,
+ lock->l_policy_data.l_inodebits.cancel_bits);
+ if (rc == 0) {
+ LDLM_LOCK_RELEASE(lock);
+ RETURN(0);
+ }
+ }
+
+ lock_res_and_lock(lock);
+ /* Lock is being canceled and the caller doesn't want to wait */
+ if (ldlm_is_canceling(lock)) {
+ if (cancel_flags & LCF_ASYNC) {
+ unlock_res_and_lock(lock);
+ } else {
+ struct l_wait_info lwi = { 0 };
+
+ unlock_res_and_lock(lock);
+ l_wait_event(lock->l_waitq, is_bl_done(lock), &lwi);
+ }
+ LDLM_LOCK_RELEASE(lock);
+ RETURN(0);
+ }
+
+ /*
+ * Lock is being converted, cancel it immediately.
+ * When convert will end, it releases lock and it will be gone.
+ */
+ if (ldlm_is_converting(lock)) {
+ /* set back flags removed by convert */
+ ldlm_set_cbpending(lock);
+ ldlm_set_bl_ast(lock);
+ }
+
+ ldlm_set_canceling(lock);
+ unlock_res_and_lock(lock);
+
+ if (cancel_flags & LCF_LOCAL)
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_LOCAL_CANCEL_PAUSE,
+ cfs_fail_val);
+
rc = ldlm_cli_cancel_local(lock);
if (rc == LDLM_FL_LOCAL_ONLY || cancel_flags & LCF_LOCAL) {
LDLM_LOCK_RELEASE(lock);
RETURN(0);
}
- /* Even if the lock is marked as LDLM_FL_BL_AST, this is a LDLM_CANCEL
+ /*
+ * Even if the lock is marked as LDLM_FL_BL_AST, this is a LDLM_CANCEL
* RPC which goes to canceld portal, so we can cancel other LRU locks
- * here and send them all as one LDLM_CANCEL RPC. */
- LASSERT(cfs_list_empty(&lock->l_bl_ast));
- cfs_list_add(&lock->l_bl_ast, &cancels);
-
- exp = lock->l_conn_export;
- if (exp_connect_cancelset(exp)) {
- avail = ldlm_format_handles_avail(class_exp2cliimp(exp),
- &RQF_LDLM_CANCEL,
- RCL_CLIENT, 0);
- LASSERT(avail > 0);
-
- ns = ldlm_lock_to_ns(lock);
- flags = ns_connect_lru_resize(ns) ?
- LDLM_CANCEL_LRUR : LDLM_CANCEL_AGED;
- count += ldlm_cancel_lru_local(ns, &cancels, 0, avail - 1,
- LCF_BL_AST, flags);
- }
+ * here and send them all as one LDLM_CANCEL RPC.
+ */
+ LASSERT(list_empty(&lock->l_bl_ast));
+ list_add(&lock->l_bl_ast, &cancels);
+
+ exp = lock->l_conn_export;
+ if (exp_connect_cancelset(exp)) {
+ avail = ldlm_format_handles_avail(class_exp2cliimp(exp),
+ &RQF_LDLM_CANCEL,
+ RCL_CLIENT, 0);
+ LASSERT(avail > 0);
+
+ ns = ldlm_lock_to_ns(lock);
+ lru_flags = ns_connect_lru_resize(ns) ?
+ LDLM_LRU_FLAG_LRUR : LDLM_LRU_FLAG_AGED;
+ count += ldlm_cancel_lru_local(ns, &cancels, 0, avail - 1,
+ LCF_BL_AST, lru_flags);
+ }
ldlm_cli_cancel_list(&cancels, count, NULL, cancel_flags);
- RETURN(0);
+ RETURN(0);
}
EXPORT_SYMBOL(ldlm_cli_cancel);
* Locally cancel up to \a count locks in list \a cancels.
* Return the number of cancelled locks.
*/
-int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
- ldlm_cancel_flags_t flags)
+int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
+ enum ldlm_cancel_flags cancel_flags)
{
struct list_head head = LIST_HEAD_INIT(head);
struct ldlm_lock *lock, *next;
int left = 0, bl_ast = 0;
__u64 rc;
- left = count;
- cfs_list_for_each_entry_safe(lock, next, cancels, l_bl_ast) {
- if (left-- == 0)
- break;
-
- if (flags & LCF_LOCAL) {
- rc = LDLM_FL_LOCAL_ONLY;
- ldlm_lock_cancel(lock);
- } else {
- rc = ldlm_cli_cancel_local(lock);
- }
- /* Until we have compound requests and can send LDLM_CANCEL
+ left = count;
+ list_for_each_entry_safe(lock, next, cancels, l_bl_ast) {
+ if (left-- == 0)
+ break;
+
+ if (cancel_flags & LCF_LOCAL) {
+ rc = LDLM_FL_LOCAL_ONLY;
+ ldlm_lock_cancel(lock);
+ } else {
+ rc = ldlm_cli_cancel_local(lock);
+ }
+ /*
+ * Until we have compound requests and can send LDLM_CANCEL
* requests batched with generic RPCs, we need to send cancels
* with the LDLM_FL_BL_AST flag in a separate RPC from
- * the one being generated now. */
- if (!(flags & LCF_BL_AST) && (rc == LDLM_FL_BL_AST)) {
+ * the one being generated now.
+ */
+ if (!(cancel_flags & LCF_BL_AST) && (rc == LDLM_FL_BL_AST)) {
LDLM_DEBUG(lock, "Cancel lock separately");
- cfs_list_del_init(&lock->l_bl_ast);
- cfs_list_add(&lock->l_bl_ast, &head);
+ list_move(&lock->l_bl_ast, &head);
bl_ast++;
- continue;
- }
- if (rc == LDLM_FL_LOCAL_ONLY) {
- /* CANCEL RPC should not be sent to server. */
- cfs_list_del_init(&lock->l_bl_ast);
- LDLM_LOCK_RELEASE(lock);
- count--;
- }
- }
- if (bl_ast > 0) {
- count -= bl_ast;
- ldlm_cli_cancel_list(&head, bl_ast, NULL, 0);
- }
-
- RETURN(count);
+ continue;
+ }
+ if (rc == LDLM_FL_LOCAL_ONLY) {
+ /* CANCEL RPC should not be sent to server. */
+ list_del_init(&lock->l_bl_ast);
+ LDLM_LOCK_RELEASE(lock);
+ count--;
+ }
+ }
+ if (bl_ast > 0) {
+ count -= bl_ast;
+ ldlm_cli_cancel_list(&head, bl_ast, NULL, 0);
+ }
+
+ RETURN(count);
}
-EXPORT_SYMBOL(ldlm_cli_cancel_list_local);
/**
* Cancel as many locks as possible w/o sending any RPCs (e.g. to write back
* dirty data, to close a file, ...) or waiting for any RPCs in-flight (e.g.
* readahead requests, ...)
*/
-static ldlm_policy_res_t ldlm_cancel_no_wait_policy(struct ldlm_namespace *ns,
- struct ldlm_lock *lock,
- int unused, int added,
- int count)
+static enum ldlm_policy_res
+ldlm_cancel_no_wait_policy(struct ldlm_namespace *ns, struct ldlm_lock *lock,
+ int unused, int added, int count)
{
- ldlm_policy_res_t result = LDLM_POLICY_CANCEL_LOCK;
+ enum ldlm_policy_res result = LDLM_POLICY_CANCEL_LOCK;
- /* don't check added & count since we want to process all locks
+ /*
+ * don't check added & count since we want to process all locks
* from unused list.
* It's fine to not take lock to access lock->l_resource since
- * the lock has already been granted so it won't change. */
+ * the lock has already been granted so it won't change.
+ */
switch (lock->l_resource->lr_type) {
case LDLM_EXTENT:
case LDLM_IBITS:
break;
default:
result = LDLM_POLICY_SKIP_LOCK;
- lock_res_and_lock(lock);
- ldlm_set_skipped(lock);
- unlock_res_and_lock(lock);
break;
}
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
-static ldlm_policy_res_t ldlm_cancel_lrur_policy(struct ldlm_namespace *ns,
- struct ldlm_lock *lock,
- int unused, int added,
- int count)
+static enum ldlm_policy_res ldlm_cancel_lrur_policy(struct ldlm_namespace *ns,
+ struct ldlm_lock *lock,
+ int unused, int added,
+ int count)
{
- cfs_time_t cur = cfs_time_current();
+ ktime_t cur = ktime_get();
struct ldlm_pool *pl = &ns->ns_pool;
- __u64 slv, lvf, lv;
- cfs_time_t la;
+ u64 slv, lvf, lv;
+ s64 la;
- /* Stop LRU processing when we reach past @count or have checked all
- * locks in LRU. */
+ /*
+ * Stop LRU processing when we reach past @count or have checked all
+ * locks in LRU.
+ */
if (count && added >= count)
return LDLM_POLICY_KEEP_LOCK;
+ /*
+ * Despite of the LV, It doesn't make sense to keep the lock which
+ * is unused for ns_max_age time.
+ */
+ if (ktime_after(ktime_get(),
+ ktime_add(lock->l_last_used, ns->ns_max_age)))
+ return LDLM_POLICY_CANCEL_LOCK;
+
slv = ldlm_pool_get_slv(pl);
lvf = ldlm_pool_get_lvf(pl);
- la = cfs_duration_sec(cfs_time_sub(cur,
- lock->l_last_used));
+ la = div_u64(ktime_to_ns(ktime_sub(cur, lock->l_last_used)),
+ NSEC_PER_SEC);
lv = lvf * la * unused;
- /* Inform pool about current CLV to see it via proc. */
+ /* Inform pool about current CLV to see it via debugfs. */
ldlm_pool_set_clv(pl, lv);
- /* Stop when SLV is not yet come from server or lv is smaller than
- * it is. */
+ /*
+ * Stop when SLV is not yet come from server or lv is smaller than
+ * it is.
+ */
if (slv == 0 || lv < slv)
return LDLM_POLICY_KEEP_LOCK;
- if (ns->ns_cancel != NULL && ns->ns_cancel(lock) == 0)
- return LDLM_POLICY_KEEP_LOCK;
-
return LDLM_POLICY_CANCEL_LOCK;
}
+static enum ldlm_policy_res
+ldlm_cancel_lrur_no_wait_policy(struct ldlm_namespace *ns,
+ struct ldlm_lock *lock,
+ int unused, int added,
+ int count)
+{
+ enum ldlm_policy_res result;
+
+ result = ldlm_cancel_lrur_policy(ns, lock, unused, added, count);
+ if (result == LDLM_POLICY_KEEP_LOCK)
+ return result;
+
+ return ldlm_cancel_no_wait_policy(ns, lock, unused, added, count);
+}
+
/**
- * Callback function for proc used policy. Makes decision whether to keep
+ * Callback function for debugfs used policy. Makes decision whether to keep
* \a lock in LRU for current \a LRU size \a unused, added in current scan \a
* added and number of locks to be preferably canceled \a count.
*
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
-static ldlm_policy_res_t ldlm_cancel_passed_policy(struct ldlm_namespace *ns,
- struct ldlm_lock *lock,
- int unused, int added,
- int count)
+static enum ldlm_policy_res ldlm_cancel_passed_policy(struct ldlm_namespace *ns,
+ struct ldlm_lock *lock,
+ int unused, int added,
+ int count)
{
- /* Stop LRU processing when we reach past @count or have checked all
- * locks in LRU. */
+ /*
+ * Stop LRU processing when we reach past @count or have checked all
+ * locks in LRU.
+ */
return (added >= count) ?
LDLM_POLICY_KEEP_LOCK : LDLM_POLICY_CANCEL_LOCK;
}
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
-static ldlm_policy_res_t ldlm_cancel_aged_policy(struct ldlm_namespace *ns,
- struct ldlm_lock *lock,
- int unused, int added,
- int count)
+static enum ldlm_policy_res ldlm_cancel_aged_policy(struct ldlm_namespace *ns,
+ struct ldlm_lock *lock,
+ int unused, int added,
+ int count)
{
- if (added >= count)
+ if ((added >= count) &&
+ ktime_before(ktime_get(),
+ ktime_add(lock->l_last_used, ns->ns_max_age)))
return LDLM_POLICY_KEEP_LOCK;
- if (cfs_time_before(cfs_time_current(),
- cfs_time_add(lock->l_last_used, ns->ns_max_age)))
- return LDLM_POLICY_KEEP_LOCK;
+ return LDLM_POLICY_CANCEL_LOCK;
+}
+
+static enum ldlm_policy_res
+ldlm_cancel_aged_no_wait_policy(struct ldlm_namespace *ns,
+ struct ldlm_lock *lock,
+ int unused, int added, int count)
+{
+ enum ldlm_policy_res result;
- if (ns->ns_cancel != NULL && ns->ns_cancel(lock) == 0)
- return LDLM_POLICY_KEEP_LOCK;
+ result = ldlm_cancel_aged_policy(ns, lock, unused, added, count);
+ if (result == LDLM_POLICY_KEEP_LOCK)
+ return result;
- return LDLM_POLICY_CANCEL_LOCK;
+ return ldlm_cancel_no_wait_policy(ns, lock, unused, added, count);
}
/**
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
-static ldlm_policy_res_t ldlm_cancel_default_policy(struct ldlm_namespace *ns,
- struct ldlm_lock *lock,
- int unused, int added,
- int count)
+static
+enum ldlm_policy_res ldlm_cancel_default_policy(struct ldlm_namespace *ns,
+ struct ldlm_lock *lock,
+ int unused, int added,
+ int count)
{
- /* Stop LRU processing when we reach past count or have checked all
- * locks in LRU. */
- return (added >= count) ?
- LDLM_POLICY_KEEP_LOCK : LDLM_POLICY_CANCEL_LOCK;
+ /*
+ * Stop LRU processing when we reach past count or have checked all
+ * locks in LRU.
+ */
+ return (added >= count) ?
+ LDLM_POLICY_KEEP_LOCK : LDLM_POLICY_CANCEL_LOCK;
}
-typedef ldlm_policy_res_t (*ldlm_cancel_lru_policy_t)(struct ldlm_namespace *,
- struct ldlm_lock *, int,
- int, int);
+typedef enum ldlm_policy_res
+(*ldlm_cancel_lru_policy_t)(struct ldlm_namespace *ns, struct ldlm_lock *lock,
+ int unused, int added, int count);
static ldlm_cancel_lru_policy_t
-ldlm_cancel_lru_policy(struct ldlm_namespace *ns, int flags)
+ldlm_cancel_lru_policy(struct ldlm_namespace *ns, enum ldlm_lru_flags lru_flags)
{
- if (flags & LDLM_CANCEL_NO_WAIT)
- return ldlm_cancel_no_wait_policy;
-
- if (ns_connect_lru_resize(ns)) {
- if (flags & LDLM_CANCEL_SHRINK)
- /* We kill passed number of old locks. */
- return ldlm_cancel_passed_policy;
- else if (flags & LDLM_CANCEL_LRUR)
- return ldlm_cancel_lrur_policy;
- else if (flags & LDLM_CANCEL_PASSED)
- return ldlm_cancel_passed_policy;
- } else {
- if (flags & LDLM_CANCEL_AGED)
- return ldlm_cancel_aged_policy;
- }
-
- return ldlm_cancel_default_policy;
+ if (ns_connect_lru_resize(ns)) {
+ if (lru_flags & LDLM_LRU_FLAG_SHRINK)
+ /* We kill passed number of old locks. */
+ return ldlm_cancel_passed_policy;
+ if (lru_flags & LDLM_LRU_FLAG_LRUR) {
+ if (lru_flags & LDLM_LRU_FLAG_NO_WAIT)
+ return ldlm_cancel_lrur_no_wait_policy;
+ else
+ return ldlm_cancel_lrur_policy;
+ }
+ if (lru_flags & LDLM_LRU_FLAG_PASSED)
+ return ldlm_cancel_passed_policy;
+ } else {
+ if (lru_flags & LDLM_LRU_FLAG_AGED) {
+ if (lru_flags & LDLM_LRU_FLAG_NO_WAIT)
+ return ldlm_cancel_aged_no_wait_policy;
+ else
+ return ldlm_cancel_aged_policy;
+ }
+ }
+ if (lru_flags & LDLM_LRU_FLAG_NO_WAIT)
+ return ldlm_cancel_no_wait_policy;
+
+ return ldlm_cancel_default_policy;
}
/**
*
* Calling policies for enabled LRU resize:
* ----------------------------------------
- * flags & LDLM_CANCEL_LRUR - use LRU resize policy (SLV from server) to
- * cancel not more than \a count locks;
+ * flags & LDLM_LRU_FLAG_LRUR - use LRU resize policy (SLV from server) to
+ * cancel not more than \a count locks;
*
- * flags & LDLM_CANCEL_PASSED - cancel \a count number of old locks (located at
- * the beginning of LRU list);
+ * flags & LDLM_LRU_FLAG_PASSED - cancel \a count number of old locks (located
+ * at the beginning of LRU list);
*
- * flags & LDLM_CANCEL_SHRINK - cancel not more than \a count locks according to
- * memory pressre policy function;
+ * flags & LDLM_LRU_FLAG_SHRINK - cancel not more than \a count locks according
+ * to memory pressre policy function;
*
- * flags & LDLM_CANCEL_AGED - cancel \a count locks according to "aged policy".
+ * flags & LDLM_LRU_FLAG_AGED - cancel \a count locks according to "aged policy"
*
- * flags & LDLM_CANCEL_NO_WAIT - cancel as many unused locks as possible
- * (typically before replaying locks) w/o
- * sending any RPCs or waiting for any
- * outstanding RPC to complete.
+ * flags & LDLM_LRU_FLAG_NO_WAIT - cancel as many unused locks as possible
+ * (typically before replaying locks) w/o
+ * sending any RPCs or waiting for any
+ * outstanding RPC to complete.
+ *
+ * flags & LDLM_CANCEL_CLEANUP - when cancelling read locks, do not check for
+ * other read locks covering the same pages, just
+ * discard those pages.
*/
-static int ldlm_prepare_lru_list(struct ldlm_namespace *ns, cfs_list_t *cancels,
- int count, int max, int flags)
+static int ldlm_prepare_lru_list(struct ldlm_namespace *ns,
+ struct list_head *cancels, int count, int max,
+ enum ldlm_lru_flags lru_flags)
{
ldlm_cancel_lru_policy_t pf;
- struct ldlm_lock *lock, *next;
- int added = 0, unused, remained;
- ENTRY;
+ int added = 0;
+ int no_wait = lru_flags & LDLM_LRU_FLAG_NO_WAIT;
- spin_lock(&ns->ns_lock);
- unused = ns->ns_nr_unused;
- remained = unused;
-
- if (!ns_connect_lru_resize(ns))
- count += unused - ns->ns_max_unused;
+ ENTRY;
- pf = ldlm_cancel_lru_policy(ns, flags);
- LASSERT(pf != NULL);
+ if (!ns_connect_lru_resize(ns))
+ count += ns->ns_nr_unused - ns->ns_max_unused;
- while (!cfs_list_empty(&ns->ns_unused_list)) {
- ldlm_policy_res_t result;
+ pf = ldlm_cancel_lru_policy(ns, lru_flags);
+ LASSERT(pf != NULL);
- /* all unused locks */
- if (remained-- <= 0)
- break;
+ /* For any flags, stop scanning if @max is reached. */
+ while (!list_empty(&ns->ns_unused_list) && (max == 0 || added < max)) {
+ struct ldlm_lock *lock;
+ struct list_head *item, *next;
+ enum ldlm_policy_res result;
+ ktime_t last_use = ktime_set(0, 0);
- /* For any flags, stop scanning if @max is reached. */
- if (max && added >= max)
- break;
+ spin_lock(&ns->ns_lock);
+ item = no_wait ? ns->ns_last_pos : &ns->ns_unused_list;
+ for (item = item->next, next = item->next;
+ item != &ns->ns_unused_list;
+ item = next, next = item->next) {
+ lock = list_entry(item, struct ldlm_lock, l_lru);
- cfs_list_for_each_entry_safe(lock, next, &ns->ns_unused_list,
- l_lru) {
- /* No locks which got blocking requests. */
+ /* No locks which got blocking requests. */
LASSERT(!ldlm_is_bl_ast(lock));
- if (flags & LDLM_CANCEL_NO_WAIT &&
- ldlm_is_skipped(lock))
- /* already processed */
- continue;
-
- /* Somebody is already doing CANCEL. No need for this
- * lock in LRU, do not traverse it again. */
- if (!ldlm_is_canceling(lock))
- break;
+ if (!ldlm_is_canceling(lock) &&
+ !ldlm_is_converting(lock))
+ break;
+ /*
+ * Somebody is already doing CANCEL. No need for this
+ * lock in LRU, do not traverse it again.
+ */
ldlm_lock_remove_from_lru_nolock(lock);
}
- if (&lock->l_lru == &ns->ns_unused_list)
+ if (item == &ns->ns_unused_list) {
+ spin_unlock(&ns->ns_lock);
break;
+ }
+
+ last_use = lock->l_last_used;
LDLM_LOCK_GET(lock);
spin_unlock(&ns->ns_lock);
lu_ref_add(&lock->l_reference, __FUNCTION__, current);
- /* Pass the lock through the policy filter and see if it
+ /*
+ * Pass the lock through the policy filter and see if it
* should stay in LRU.
*
* Even for shrinker policy we stop scanning if
* That is, for shrinker policy we drop only
* old locks, but additionally choose them by
* their weight. Big extent locks will stay in
- * the cache. */
- result = pf(ns, lock, unused, added, count);
+ * the cache.
+ */
+ result = pf(ns, lock, ns->ns_nr_unused, added, count);
if (result == LDLM_POLICY_KEEP_LOCK) {
- lu_ref_del(&lock->l_reference,
- __FUNCTION__, current);
+ lu_ref_del(&lock->l_reference, __func__, current);
LDLM_LOCK_RELEASE(lock);
- spin_lock(&ns->ns_lock);
break;
}
+
if (result == LDLM_POLICY_SKIP_LOCK) {
- lu_ref_del(&lock->l_reference,
- __func__, current);
+ lu_ref_del(&lock->l_reference, __func__, current);
+ if (no_wait) {
+ spin_lock(&ns->ns_lock);
+ if (!list_empty(&lock->l_lru) &&
+ lock->l_lru.prev == ns->ns_last_pos)
+ ns->ns_last_pos = &lock->l_lru;
+ spin_unlock(&ns->ns_lock);
+ }
+
LDLM_LOCK_RELEASE(lock);
- spin_lock(&ns->ns_lock);
continue;
}
lock_res_and_lock(lock);
/* Check flags again under the lock. */
- if (ldlm_is_canceling(lock) ||
- (ldlm_lock_remove_from_lru(lock) == 0)) {
- /* Another thread is removing lock from LRU, or
+ if (ldlm_is_canceling(lock) || ldlm_is_converting(lock) ||
+ ldlm_lock_remove_from_lru_check(lock, last_use) == 0) {
+ /*
+ * Another thread is removing lock from LRU, or
* somebody is already doing CANCEL, or there
* is a blocking request which will send cancel
- * by itself, or the lock is no longer unused. */
+ * by itself, or the lock is no longer unused or
+ * the lock has been used since the pf() call and
+ * pages could be put under it.
+ */
unlock_res_and_lock(lock);
lu_ref_del(&lock->l_reference, __FUNCTION__, current);
LDLM_LOCK_RELEASE(lock);
- spin_lock(&ns->ns_lock);
continue;
}
LASSERT(!lock->l_readers && !lock->l_writers);
- /* If we have chosen to cancel this lock voluntarily, we
+ /*
+ * If we have chosen to cancel this lock voluntarily, we
* better send cancel notification to server, so that it
* frees appropriate state. This might lead to a race
* where while we are doing cancel here, server is also
- * silently cancelling this lock. */
+ * silently cancelling this lock.
+ */
ldlm_clear_cancel_on_block(lock);
- /* Setting the CBPENDING flag is a little misleading,
+ /*
+ * Setting the CBPENDING flag is a little misleading,
* but prevents an important race; namely, once
* CBPENDING is set, the lock can accumulate no more
* readers/writers. Since readers and writers are
* already zero here, ldlm_lock_decref() won't see
- * this flag and call l_blocking_ast */
+ * this flag and call l_blocking_ast
+ */
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_CANCELING;
- /* We can't re-add to l_lru as it confuses the
+ if ((lru_flags & LDLM_LRU_FLAG_CLEANUP) &&
+ (lock->l_resource->lr_type == LDLM_EXTENT ||
+ ldlm_has_dom(lock)) && lock->l_granted_mode == LCK_PR)
+ ldlm_set_discard_data(lock);
+
+ /*
+ * We can't re-add to l_lru as it confuses the
* refcounting in ldlm_lock_remove_from_lru() if an AST
* arrives after we drop lr_lock below. We use l_bl_ast
* and can't use l_pending_chain as it is used both on
- * server and client nevertheless bug 5666 says it is
- * used only on server */
- LASSERT(cfs_list_empty(&lock->l_bl_ast));
- cfs_list_add(&lock->l_bl_ast, cancels);
+ * server and client nevertheless b=5666 says it is
+ * used only on server
+ */
+ LASSERT(list_empty(&lock->l_bl_ast));
+ list_add(&lock->l_bl_ast, cancels);
unlock_res_and_lock(lock);
lu_ref_del(&lock->l_reference, __FUNCTION__, current);
- spin_lock(&ns->ns_lock);
added++;
- unused--;
}
- spin_unlock(&ns->ns_lock);
RETURN(added);
}
-int ldlm_cancel_lru_local(struct ldlm_namespace *ns, cfs_list_t *cancels,
- int count, int max, ldlm_cancel_flags_t cancel_flags,
- int flags)
+int ldlm_cancel_lru_local(struct ldlm_namespace *ns, struct list_head *cancels,
+ int count, int max,
+ enum ldlm_cancel_flags cancel_flags,
+ enum ldlm_lru_flags lru_flags)
{
- int added;
- added = ldlm_prepare_lru_list(ns, cancels, count, max, flags);
- if (added <= 0)
- return added;
- return ldlm_cli_cancel_list_local(cancels, added, cancel_flags);
+ int added;
+
+ added = ldlm_prepare_lru_list(ns, cancels, count, max, lru_flags);
+ if (added <= 0)
+ return added;
+
+ return ldlm_cli_cancel_list_local(cancels, added, cancel_flags);
}
/**
* callback will be performed in this function.
*/
int ldlm_cancel_lru(struct ldlm_namespace *ns, int nr,
- ldlm_cancel_flags_t cancel_flags,
- int flags)
+ enum ldlm_cancel_flags cancel_flags,
+ enum ldlm_lru_flags lru_flags)
{
struct list_head cancels = LIST_HEAD_INIT(cancels);
int count, rc;
+
ENTRY;
-#ifndef __KERNEL__
- cancel_flags &= ~LCF_ASYNC; /* force to be sync in user space */
-#endif
- /* Just prepare the list of locks, do not actually cancel them yet.
- * Locks are cancelled later in a separate thread. */
- count = ldlm_prepare_lru_list(ns, &cancels, nr, 0, flags);
+ /*
+ * Just prepare the list of locks, do not actually cancel them yet.
+ * Locks are cancelled later in a separate thread.
+ */
+ count = ldlm_prepare_lru_list(ns, &cancels, nr, 0, lru_flags);
rc = ldlm_bl_to_thread_list(ns, NULL, &cancels, count, cancel_flags);
if (rc == 0)
RETURN(count);
* list.
*/
int ldlm_cancel_resource_local(struct ldlm_resource *res,
- cfs_list_t *cancels,
- ldlm_policy_data_t *policy,
- ldlm_mode_t mode, __u64 lock_flags,
- ldlm_cancel_flags_t cancel_flags, void *opaque)
+ struct list_head *cancels,
+ union ldlm_policy_data *policy,
+ enum ldlm_mode mode, __u64 lock_flags,
+ enum ldlm_cancel_flags cancel_flags,
+ void *opaque)
{
- struct ldlm_lock *lock;
- int count = 0;
- ENTRY;
-
- lock_res(res);
- cfs_list_for_each_entry(lock, &res->lr_granted, l_res_link) {
- if (opaque != NULL && lock->l_ast_data != opaque) {
- LDLM_ERROR(lock, "data %p doesn't match opaque %p",
- lock->l_ast_data, opaque);
- //LBUG();
- continue;
- }
-
- if (lock->l_readers || lock->l_writers)
- continue;
-
- /* If somebody is already doing CANCEL, or blocking AST came,
- * skip this lock. */
- if (ldlm_is_bl_ast(lock) || ldlm_is_canceling(lock))
+ struct ldlm_lock *lock;
+ int count = 0;
+
+ ENTRY;
+
+ lock_res(res);
+ list_for_each_entry(lock, &res->lr_granted, l_res_link) {
+ if (opaque != NULL && lock->l_ast_data != opaque) {
+ LDLM_ERROR(lock, "data %p doesn't match opaque %p",
+ lock->l_ast_data, opaque);
+ continue;
+ }
+
+ if (lock->l_readers || lock->l_writers)
continue;
- if (lockmode_compat(lock->l_granted_mode, mode))
- continue;
+ /*
+ * If somebody is already doing CANCEL, or blocking AST came,
+ * or lock is being converted then skip this lock.
+ */
+ if (ldlm_is_bl_ast(lock) || ldlm_is_canceling(lock) ||
+ ldlm_is_converting(lock))
+ continue;
- /* If policy is given and this is IBITS lock, add to list only
- * those locks that match by policy. */
- if (policy && (lock->l_resource->lr_type == LDLM_IBITS) &&
- !(lock->l_policy_data.l_inodebits.bits &
- policy->l_inodebits.bits))
- continue;
+ if (lockmode_compat(lock->l_granted_mode, mode))
+ continue;
+
+ /*
+ * If policy is given and this is IBITS lock, add to list only
+ * those locks that match by policy.
+ */
+ if (policy && (lock->l_resource->lr_type == LDLM_IBITS)) {
+ if (!(lock->l_policy_data.l_inodebits.bits &
+ policy->l_inodebits.bits))
+ continue;
+ /* Skip locks with DoM bit if it is not set in policy
+ * to don't flush data by side-bits. Lock convert will
+ * drop those bits separately.
+ */
+ if (ldlm_has_dom(lock) &&
+ !(policy->l_inodebits.bits & MDS_INODELOCK_DOM))
+ continue;
+ }
/* See CBPENDING comment in ldlm_cancel_lru */
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_CANCELING |
lock_flags;
- LASSERT(cfs_list_empty(&lock->l_bl_ast));
- cfs_list_add(&lock->l_bl_ast, cancels);
- LDLM_LOCK_GET(lock);
- count++;
- }
- unlock_res(res);
+ LASSERT(list_empty(&lock->l_bl_ast));
+ list_add(&lock->l_bl_ast, cancels);
+ LDLM_LOCK_GET(lock);
+ count++;
+ }
+ unlock_res(res);
- RETURN(ldlm_cli_cancel_list_local(cancels, count, cancel_flags));
+ RETURN(ldlm_cli_cancel_list_local(cancels, count, cancel_flags));
}
EXPORT_SYMBOL(ldlm_cancel_resource_local);
* buffer at the offset \a off.
* Destroy \a cancels at the end.
*/
-int ldlm_cli_cancel_list(cfs_list_t *cancels, int count,
- struct ptlrpc_request *req, ldlm_cancel_flags_t flags)
+int ldlm_cli_cancel_list(struct list_head *cancels, int count,
+ struct ptlrpc_request *req,
+ enum ldlm_cancel_flags flags)
{
- struct ldlm_lock *lock;
- int res = 0;
- ENTRY;
-
- if (cfs_list_empty(cancels) || count == 0)
- RETURN(0);
-
- /* XXX: requests (both batched and not) could be sent in parallel.
- * Usually it is enough to have just 1 RPC, but it is possible that
- * there are too many locks to be cancelled in LRU or on a resource.
- * It would also speed up the case when the server does not support
- * the feature. */
- while (count > 0) {
- LASSERT(!cfs_list_empty(cancels));
- lock = cfs_list_entry(cancels->next, struct ldlm_lock,
- l_bl_ast);
- LASSERT(lock->l_conn_export);
-
- if (exp_connect_cancelset(lock->l_conn_export)) {
- res = count;
- if (req)
- ldlm_cancel_pack(req, cancels, count);
- else
- res = ldlm_cli_cancel_req(lock->l_conn_export,
- cancels, count,
- flags);
- } else {
- res = ldlm_cli_cancel_req(lock->l_conn_export,
- cancels, 1, flags);
- }
+ struct ldlm_lock *lock;
+ int res = 0;
+
+ ENTRY;
+
+ if (list_empty(cancels) || count == 0)
+ RETURN(0);
+
+ /*
+ * XXX: requests (both batched and not) could be sent in parallel.
+ * Usually it is enough to have just 1 RPC, but it is possible that
+ * there are too many locks to be cancelled in LRU or on a resource.
+ * It would also speed up the case when the server does not support
+ * the feature.
+ */
+ while (count > 0) {
+ LASSERT(!list_empty(cancels));
+ lock = list_entry(cancels->next, struct ldlm_lock,
+ l_bl_ast);
+ LASSERT(lock->l_conn_export);
+
+ if (exp_connect_cancelset(lock->l_conn_export)) {
+ res = count;
+ if (req)
+ ldlm_cancel_pack(req, cancels, count);
+ else
+ res = ldlm_cli_cancel_req(lock->l_conn_export,
+ cancels, count,
+ flags);
+ } else {
+ res = ldlm_cli_cancel_req(lock->l_conn_export,
+ cancels, 1, flags);
+ }
if (res < 0) {
CDEBUG_LIMIT(res == -ESHUTDOWN ? D_DLMTRACE : D_ERROR,
res = count;
}
- count -= res;
- ldlm_lock_list_put(cancels, l_bl_ast, res);
- }
- LASSERT(count == 0);
- RETURN(0);
+ count -= res;
+ ldlm_lock_list_put(cancels, l_bl_ast, res);
+ }
+ LASSERT(count == 0);
+ RETURN(0);
}
EXPORT_SYMBOL(ldlm_cli_cancel_list);
* Cancel all locks on a resource that have 0 readers/writers.
*
* If flags & LDLM_FL_LOCAL_ONLY, throw the locks away without trying
- * to notify the server. */
+ * to notify the server.
+ */
int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
const struct ldlm_res_id *res_id,
- ldlm_policy_data_t *policy,
- ldlm_mode_t mode,
- ldlm_cancel_flags_t flags,
- void *opaque)
+ union ldlm_policy_data *policy,
+ enum ldlm_mode mode,
+ enum ldlm_cancel_flags flags, void *opaque)
{
struct ldlm_resource *res;
struct list_head cancels = LIST_HEAD_INIT(cancels);
int count;
int rc;
+
ENTRY;
res = ldlm_resource_get(ns, NULL, res_id, 0, 0);
if (IS_ERR(res)) {
/* This is not a problem. */
- CDEBUG(D_INFO, "No resource "LPU64"\n", res_id->name[0]);
+ CDEBUG(D_INFO, "No resource %llu\n", res_id->name[0]);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_cli_cancel_unused_resource);
struct ldlm_cli_cancel_arg {
- int lc_flags;
- void *lc_opaque;
+ int lc_flags;
+ void *lc_opaque;
};
-static int ldlm_cli_hash_cancel_unused(cfs_hash_t *hs, cfs_hash_bd_t *bd,
- cfs_hlist_node_t *hnode, void *arg)
+static int
+ldlm_cli_hash_cancel_unused(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode, void *arg)
{
struct ldlm_resource *res = cfs_hash_object(hs, hnode);
struct ldlm_cli_cancel_arg *lc = arg;
* that have 0 readers/writers.
*
* If flags & LCF_LOCAL, throw the locks away without trying
- * to notify the server. */
+ * to notify the server.
+ */
int ldlm_cli_cancel_unused(struct ldlm_namespace *ns,
- const struct ldlm_res_id *res_id,
- ldlm_cancel_flags_t flags, void *opaque)
+ const struct ldlm_res_id *res_id,
+ enum ldlm_cancel_flags flags, void *opaque)
{
- struct ldlm_cli_cancel_arg arg = {
- .lc_flags = flags,
- .lc_opaque = opaque,
- };
-
- ENTRY;
-
- if (ns == NULL)
- RETURN(ELDLM_OK);
-
- if (res_id != NULL) {
- RETURN(ldlm_cli_cancel_unused_resource(ns, res_id, NULL,
- LCK_MINMODE, flags,
- opaque));
- } else {
- cfs_hash_for_each_nolock(ns->ns_rs_hash,
- ldlm_cli_hash_cancel_unused, &arg);
- RETURN(ELDLM_OK);
- }
+ struct ldlm_cli_cancel_arg arg = {
+ .lc_flags = flags,
+ .lc_opaque = opaque,
+ };
+
+ ENTRY;
+
+ if (ns == NULL)
+ RETURN(ELDLM_OK);
+
+ if (res_id != NULL) {
+ RETURN(ldlm_cli_cancel_unused_resource(ns, res_id, NULL,
+ LCK_MINMODE, flags,
+ opaque));
+ } else {
+ cfs_hash_for_each_nolock(ns->ns_rs_hash,
+ ldlm_cli_hash_cancel_unused, &arg, 0);
+ RETURN(ELDLM_OK);
+ }
}
-EXPORT_SYMBOL(ldlm_cli_cancel_unused);
/* Lock iterators. */
int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
- void *closure)
+ void *closure)
{
- cfs_list_t *tmp, *next;
- struct ldlm_lock *lock;
- int rc = LDLM_ITER_CONTINUE;
-
- ENTRY;
-
- if (!res)
- RETURN(LDLM_ITER_CONTINUE);
+ struct list_head *tmp, *next;
+ struct ldlm_lock *lock;
+ int rc = LDLM_ITER_CONTINUE;
- lock_res(res);
- cfs_list_for_each_safe(tmp, next, &res->lr_granted) {
- lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
+ ENTRY;
- if (iter(lock, closure) == LDLM_ITER_STOP)
- GOTO(out, rc = LDLM_ITER_STOP);
- }
+ if (!res)
+ RETURN(LDLM_ITER_CONTINUE);
- cfs_list_for_each_safe(tmp, next, &res->lr_converting) {
- lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
+ lock_res(res);
+ list_for_each_safe(tmp, next, &res->lr_granted) {
+ lock = list_entry(tmp, struct ldlm_lock, l_res_link);
- if (iter(lock, closure) == LDLM_ITER_STOP)
- GOTO(out, rc = LDLM_ITER_STOP);
- }
+ if (iter(lock, closure) == LDLM_ITER_STOP)
+ GOTO(out, rc = LDLM_ITER_STOP);
+ }
- cfs_list_for_each_safe(tmp, next, &res->lr_waiting) {
- lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
+ list_for_each_safe(tmp, next, &res->lr_waiting) {
+ lock = list_entry(tmp, struct ldlm_lock, l_res_link);
- if (iter(lock, closure) == LDLM_ITER_STOP)
- GOTO(out, rc = LDLM_ITER_STOP);
- }
- out:
- unlock_res(res);
- RETURN(rc);
+ if (iter(lock, closure) == LDLM_ITER_STOP)
+ GOTO(out, rc = LDLM_ITER_STOP);
+ }
+out:
+ unlock_res(res);
+ RETURN(rc);
}
-EXPORT_SYMBOL(ldlm_resource_foreach);
struct iter_helper_data {
- ldlm_iterator_t iter;
- void *closure;
+ ldlm_iterator_t iter;
+ void *closure;
};
static int ldlm_iter_helper(struct ldlm_lock *lock, void *closure)
{
- struct iter_helper_data *helper = closure;
- return helper->iter(lock, helper->closure);
+ struct iter_helper_data *helper = closure;
+
+ return helper->iter(lock, helper->closure);
}
-static int ldlm_res_iter_helper(cfs_hash_t *hs, cfs_hash_bd_t *bd,
- cfs_hlist_node_t *hnode, void *arg)
+static int ldlm_res_iter_helper(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode, void *arg)
{
- struct ldlm_resource *res = cfs_hash_object(hs, hnode);
+ struct ldlm_resource *res = cfs_hash_object(hs, hnode);
- return ldlm_resource_foreach(res, ldlm_iter_helper, arg) ==
- LDLM_ITER_STOP;
+ return ldlm_resource_foreach(res, ldlm_iter_helper, arg) ==
+ LDLM_ITER_STOP;
}
void ldlm_namespace_foreach(struct ldlm_namespace *ns,
- ldlm_iterator_t iter, void *closure)
+ ldlm_iterator_t iter, void *closure)
{
struct iter_helper_data helper = { .iter = iter, .closure = closure };
- cfs_hash_for_each_nolock(ns->ns_rs_hash,
- ldlm_res_iter_helper, &helper);
+ cfs_hash_for_each_nolock(ns->ns_rs_hash,
+ ldlm_res_iter_helper, &helper, 0);
}
-EXPORT_SYMBOL(ldlm_namespace_foreach);
-/* non-blocking function to manipulate a lock whose cb_data is being put away.
+/*
+ * non-blocking function to manipulate a lock whose cb_data is being put away.
* return 0: find no resource
* > 0: must be LDLM_ITER_STOP/LDLM_ITER_CONTINUE.
* < 0: errors
{
struct ldlm_resource *res;
int rc;
+
ENTRY;
LASSERTF(ns != NULL, "must pass in namespace\n");
EXPORT_SYMBOL(ldlm_resource_iterate);
/* Lock replay */
-
static int ldlm_chain_lock_for_replay(struct ldlm_lock *lock, void *closure)
{
- cfs_list_t *list = closure;
-
- /* we use l_pending_chain here, because it's unused on clients. */
- LASSERTF(cfs_list_empty(&lock->l_pending_chain),
- "lock %p next %p prev %p\n",
- lock, &lock->l_pending_chain.next,&lock->l_pending_chain.prev);
- /* bug 9573: don't replay locks left after eviction, or
- * bug 17614: locks being actively cancelled. Get a reference
- * on a lock so that it does not disapear under us (e.g. due to cancel)
- */
- if (!(lock->l_flags & (LDLM_FL_FAILED|LDLM_FL_CANCELING))) {
- cfs_list_add(&lock->l_pending_chain, list);
- LDLM_LOCK_GET(lock);
- }
-
- return LDLM_ITER_CONTINUE;
+ struct list_head *list = closure;
+
+ /* we use l_pending_chain here, because it's unused on clients. */
+ LASSERTF(list_empty(&lock->l_pending_chain),
+ "lock %p next %p prev %p\n",
+ lock, &lock->l_pending_chain.next,
+ &lock->l_pending_chain.prev);
+ /*
+ * b=9573: don't replay locks left after eviction, or
+ * b=17614: locks being actively cancelled. Get a reference
+ * on a lock so that it does not disapear under us (e.g. due to cancel)
+ */
+ if (!(lock->l_flags & (LDLM_FL_FAILED|LDLM_FL_BL_DONE))) {
+ list_add(&lock->l_pending_chain, list);
+ LDLM_LOCK_GET(lock);
+ }
+
+ return LDLM_ITER_CONTINUE;
}
static int replay_lock_interpret(const struct lu_env *env,
- struct ptlrpc_request *req,
- struct ldlm_async_args *aa, int rc)
+ struct ptlrpc_request *req, void *args, int rc)
{
+ struct ldlm_async_args *aa = args;
struct ldlm_lock *lock;
struct ldlm_reply *reply;
struct obd_export *exp;
if (rc != ELDLM_OK)
GOTO(out, rc);
- reply = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
- if (reply == NULL)
- GOTO(out, rc = -EPROTO);
-
- lock = ldlm_handle2lock(&aa->lock_handle);
- if (!lock) {
- CERROR("received replay ack for unknown local cookie "LPX64
- " remote cookie "LPX64 " from server %s id %s\n",
- aa->lock_handle.cookie, reply->lock_handle.cookie,
- req->rq_export->exp_client_uuid.uuid,
- libcfs_id2str(req->rq_peer));
- GOTO(out, rc = -ESTALE);
- }
-
- /* Key change rehash lock in per-export hash with new key */
- exp = req->rq_export;
- if (exp && exp->exp_lock_hash) {
- /* In the function below, .hs_keycmp resolves to
- * ldlm_export_lock_keycmp() */
+ reply = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
+ if (reply == NULL)
+ GOTO(out, rc = -EPROTO);
+
+ lock = ldlm_handle2lock(&aa->lock_handle);
+ if (!lock) {
+ CERROR("received replay ack for unknown local cookie %#llx remote cookie %#llx from server %s id %s\n",
+ aa->lock_handle.cookie, reply->lock_handle.cookie,
+ req->rq_export->exp_client_uuid.uuid,
+ libcfs_id2str(req->rq_peer));
+ GOTO(out, rc = -ESTALE);
+ }
+
+ /* Key change rehash lock in per-export hash with new key */
+ exp = req->rq_export;
+ if (exp && exp->exp_lock_hash) {
+ /*
+ * In the function below, .hs_keycmp resolves to
+ * ldlm_export_lock_keycmp()
+ */
/* coverity[overrun-buffer-val] */
- cfs_hash_rehash_key(exp->exp_lock_hash,
- &lock->l_remote_handle,
- &reply->lock_handle,
- &lock->l_exp_hash);
- } else {
- lock->l_remote_handle = reply->lock_handle;
- }
-
- LDLM_DEBUG(lock, "replayed lock:");
- ptlrpc_import_recovery_state_machine(req->rq_import);
- LDLM_LOCK_PUT(lock);
+ cfs_hash_rehash_key(exp->exp_lock_hash,
+ &lock->l_remote_handle,
+ &reply->lock_handle,
+ &lock->l_exp_hash);
+ } else {
+ lock->l_remote_handle = reply->lock_handle;
+ }
+
+ LDLM_DEBUG(lock, "replayed lock:");
+ ptlrpc_import_recovery_state_machine(req->rq_import);
+ LDLM_LOCK_PUT(lock);
out:
- if (rc != ELDLM_OK)
- ptlrpc_connect_import(req->rq_import);
+ if (rc != ELDLM_OK)
+ ptlrpc_connect_import(req->rq_import);
- RETURN(rc);
+ RETURN(rc);
}
static int replay_one_lock(struct obd_import *imp, struct ldlm_lock *lock)
{
- struct ptlrpc_request *req;
- struct ldlm_async_args *aa;
- struct ldlm_request *body;
- int flags;
- ENTRY;
+ struct ptlrpc_request *req;
+ struct ldlm_async_args *aa;
+ struct ldlm_request *body;
+ int flags;
+ ENTRY;
- /* Bug 11974: Do not replay a lock which is actively being canceled */
- if (ldlm_is_canceling(lock)) {
- LDLM_DEBUG(lock, "Not replaying canceled lock:");
- RETURN(0);
- }
- /* If this is reply-less callback lock, we cannot replay it, since
- * server might have long dropped it, but notification of that event was
- * lost by network. (and server granted conflicting lock already) */
+ /* b=11974: Do not replay a lock which is actively being canceled */
+ if (ldlm_is_bl_done(lock)) {
+ LDLM_DEBUG(lock, "Not replaying canceled lock:");
+ RETURN(0);
+ }
+
+ /*
+ * If this is reply-less callback lock, we cannot replay it, since
+ * server might have long dropped it, but notification of that event was
+ * lost by network. (and server granted conflicting lock already)
+ */
if (ldlm_is_cancel_on_block(lock)) {
- LDLM_DEBUG(lock, "Not replaying reply-less lock:");
- ldlm_lock_cancel(lock);
- RETURN(0);
- }
-
- /*
- * If granted mode matches the requested mode, this lock is granted.
- *
- * If they differ, but we have a granted mode, then we were granted
- * one mode and now want another: ergo, converting.
- *
- * If we haven't been granted anything and are on a resource list,
- * then we're blocked/waiting.
- *
- * If we haven't been granted anything and we're NOT on a resource list,
- * then we haven't got a reply yet and don't have a known disposition.
- * This happens whenever a lock enqueue is the request that triggers
- * recovery.
- */
- if (lock->l_granted_mode == lock->l_req_mode)
- flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_GRANTED;
- else if (lock->l_granted_mode)
- flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_CONV;
- else if (!cfs_list_empty(&lock->l_res_link))
- flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_WAIT;
- else
- flags = LDLM_FL_REPLAY;
-
- req = ptlrpc_request_alloc_pack(imp, &RQF_LDLM_ENQUEUE,
- LUSTRE_DLM_VERSION, LDLM_ENQUEUE);
- if (req == NULL)
- RETURN(-ENOMEM);
-
- /* We're part of recovery, so don't wait for it. */
- req->rq_send_state = LUSTRE_IMP_REPLAY_LOCKS;
-
- body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
- ldlm_lock2desc(lock, &body->lock_desc);
+ LDLM_DEBUG(lock, "Not replaying reply-less lock:");
+ ldlm_lock_cancel(lock);
+ RETURN(0);
+ }
+
+ /*
+ * If granted mode matches the requested mode, this lock is granted.
+ *
+ * If we haven't been granted anything and are on a resource list,
+ * then we're blocked/waiting.
+ *
+ * If we haven't been granted anything and we're NOT on a resource list,
+ * then we haven't got a reply yet and don't have a known disposition.
+ * This happens whenever a lock enqueue is the request that triggers
+ * recovery.
+ */
+ if (ldlm_is_granted(lock))
+ flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_GRANTED;
+ else if (!list_empty(&lock->l_res_link))
+ flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_WAIT;
+ else
+ flags = LDLM_FL_REPLAY;
+
+ req = ptlrpc_request_alloc_pack(imp, &RQF_LDLM_ENQUEUE,
+ LUSTRE_DLM_VERSION, LDLM_ENQUEUE);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ /* We're part of recovery, so don't wait for it. */
+ req->rq_send_state = LUSTRE_IMP_REPLAY_LOCKS;
+
+ body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
+ ldlm_lock2desc(lock, &body->lock_desc);
body->lock_flags = ldlm_flags_to_wire(flags);
- ldlm_lock2handle(lock, &body->lock_handle[0]);
+ ldlm_lock2handle(lock, &body->lock_handle[0]);
if (lock->l_lvb_len > 0)
req_capsule_extend(&req->rq_pill, &RQF_LDLM_ENQUEUE_LVB);
req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
lock->l_lvb_len);
- ptlrpc_request_set_replen(req);
- /* notify the server we've replayed all requests.
- * also, we mark the request to be put on a dedicated
- * queue to be processed after all request replayes.
- * bug 6063 */
+ ptlrpc_request_set_replen(req);
+ /*
+ * notify the server we've replayed all requests.
+ * also, we mark the request to be put on a dedicated
+ * queue to be processed after all request replayes.
+ * b=6063
+ */
lustre_msg_set_flags(req->rq_reqmsg, MSG_REQ_REPLAY_DONE);
LDLM_DEBUG(lock, "replaying lock:");
CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
aa = ptlrpc_req_async_args(req);
aa->lock_handle = body->lock_handle[0];
- req->rq_interpret_reply = (ptlrpc_interpterer_t)replay_lock_interpret;
- ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
+ req->rq_interpret_reply = replay_lock_interpret;
+ ptlrpcd_add_req(req);
RETURN(0);
}
int canceled;
struct list_head cancels = LIST_HEAD_INIT(cancels);
- CDEBUG(D_DLMTRACE, "Dropping as many unused locks as possible before"
- "replay for namespace %s (%d)\n",
- ldlm_ns_name(ns), ns->ns_nr_unused);
+ CDEBUG(D_DLMTRACE,
+ "Dropping as many unused locks as possible before replay for namespace %s (%d)\n",
+ ldlm_ns_name(ns), ns->ns_nr_unused);
- /* We don't need to care whether or not LRU resize is enabled
- * because the LDLM_CANCEL_NO_WAIT policy doesn't use the
- * count parameter */
+ /*
+ * We don't need to care whether or not LRU resize is enabled
+ * because the LDLM_LRU_FLAG_NO_WAIT policy doesn't use the
+ * count parameter
+ */
canceled = ldlm_cancel_lru_local(ns, &cancels, ns->ns_nr_unused, 0,
- LCF_LOCAL, LDLM_CANCEL_NO_WAIT);
+ LCF_LOCAL, LDLM_LRU_FLAG_NO_WAIT);
CDEBUG(D_DLMTRACE, "Canceled %d unused locks from namespace %s\n",
canceled, ldlm_ns_name(ns));
ldlm_namespace_foreach(ns, ldlm_chain_lock_for_replay, &list);
- cfs_list_for_each_entry_safe(lock, next, &list, l_pending_chain) {
- cfs_list_del_init(&lock->l_pending_chain);
+ list_for_each_entry_safe(lock, next, &list, l_pending_chain) {
+ list_del_init(&lock->l_pending_chain);
if (rc) {
LDLM_LOCK_RELEASE(lock);
continue; /* or try to do the rest? */
RETURN(rc);
}
-EXPORT_SYMBOL(ldlm_replay_locks);