/**
* Callback handler for receiving incoming glimpse ASTs.
*
* This only can happen on client side. After handling the glimpse AST
* we also consider dropping the lock here if it is unused locally for a
* long time.
*/
static void ldlm_handle_gl_callback(struct ptlrpc_request *req,
struct ldlm_namespace *ns,
struct ldlm_request *dlm_req,
struct ldlm_lock *lock)
{
int rc = -ENOSYS;
LDLM_DEBUG(lock, "client glimpse AST callback handler");
if (lock->l_glimpse_ast != NULL)
rc = lock->l_glimpse_ast(lock, req);
if (req->rq_repmsg != NULL) {
ptlrpc_reply(req);
} else {
req->rq_status = rc;
ptlrpc_error(req);
}
lock_res_and_lock(lock);
if (lock->l_granted_mode == LCK_PW &&
!lock->l_readers && !lock->l_writers &&
cfs_time_after(cfs_time_current(),
cfs_time_add(lock->l_last_used,
cfs_time_seconds(10)))) {
unlock_res_and_lock(lock);
if (ldlm_bl_to_thread_lock(ns, NULL, lock))
ldlm_handle_bl_callback(ns, NULL, lock);
return;
}
unlock_res_and_lock(lock);
LDLM_LOCK_RELEASE(lock);
}
/**
* Callback handler for receiving incoming blocking ASTs.
*
* This can only happen on client side.
*/
void ldlm_handle_bl_callback(struct ldlm_namespace *ns,
struct ldlm_lock_desc *ld, struct ldlm_lock *lock)
{
int do_ast;
LDLM_DEBUG(lock, "client blocking AST callback handler");
lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_CBPENDING;
if (lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)
lock->l_flags |= LDLM_FL_CANCEL;
do_ast = (!lock->l_readers && !lock->l_writers);
unlock_res_and_lock(lock);
if (do_ast) {
CDEBUG(D_DLMTRACE, "Lock %p already unused, calling callback (%p)\n",
lock, lock->l_blocking_ast);
if (lock->l_blocking_ast != NULL)
lock->l_blocking_ast(lock, ld, lock->l_ast_data,
LDLM_CB_BLOCKING);
} else {
CDEBUG(D_DLMTRACE, "Lock %p is referenced, will be cancelled later\n",
lock);
}
LDLM_DEBUG(lock, "client blocking callback handler END");
LDLM_LOCK_RELEASE(lock);
}
/**
* Breaks a link between osc_lock and dlm_lock.
*/
static void osc_lock_detach(const struct lu_env *env, struct osc_lock *olck)
{
struct ldlm_lock *dlmlock;
spin_lock(&osc_ast_guard);
dlmlock = olck->ols_lock;
if (dlmlock == NULL) {
spin_unlock(&osc_ast_guard);
return;
}
olck->ols_lock = NULL;
/* wb(); --- for all who checks (ols->ols_lock != NULL) before
* call to osc_lock_detach() */
dlmlock->l_ast_data = NULL;
olck->ols_handle.cookie = 0ULL;
spin_unlock(&osc_ast_guard);
lock_res_and_lock(dlmlock);
if (dlmlock->l_granted_mode == dlmlock->l_req_mode) {
struct cl_object *obj = olck->ols_cl.cls_obj;
struct cl_attr *attr = &osc_env_info(env)->oti_attr;
__u64 old_kms;
cl_object_attr_lock(obj);
/* Must get the value under the lock to avoid possible races. */
old_kms = cl2osc(obj)->oo_oinfo->loi_kms;
/* Update the kms. Need to loop all granted locks.
* Not a problem for the client */
attr->cat_kms = ldlm_extent_shift_kms(dlmlock, old_kms);
cl_object_attr_set(env, obj, attr, CAT_KMS);
cl_object_attr_unlock(obj);
}
unlock_res_and_lock(dlmlock);
/* release a reference taken in osc_lock_upcall0(). */
LASSERT(olck->ols_has_ref);
lu_ref_del(&dlmlock->l_reference, "osc_lock", olck);
LDLM_LOCK_RELEASE(dlmlock);
olck->ols_has_ref = 0;
}
/* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
static int ldlm_callback_handler(struct ptlrpc_request *req)
{
struct ldlm_namespace *ns;
struct ldlm_request *dlm_req;
struct ldlm_lock *lock;
int rc;
/* Requests arrive in sender's byte order. The ptlrpc service
* handler has already checked and, if necessary, byte-swapped the
* incoming request message body, but I am responsible for the
* message buffers. */
/* do nothing for sec context finalize */
if (lustre_msg_get_opc(req->rq_reqmsg) == SEC_CTX_FINI)
return 0;
req_capsule_init(&req->rq_pill, req, RCL_SERVER);
if (req->rq_export == NULL) {
rc = ldlm_callback_reply(req, -ENOTCONN);
ldlm_callback_errmsg(req, "Operate on unconnected server",
rc, NULL);
return 0;
}
LASSERT(req->rq_export != NULL);
LASSERT(req->rq_export->exp_obd != NULL);
switch (lustre_msg_get_opc(req->rq_reqmsg)) {
case LDLM_BL_CALLBACK:
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
return 0;
break;
case LDLM_CP_CALLBACK:
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CP_CALLBACK_NET))
return 0;
break;
case LDLM_GL_CALLBACK:
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_GL_CALLBACK_NET))
return 0;
break;
case LDLM_SET_INFO:
rc = ldlm_handle_setinfo(req);
ldlm_callback_reply(req, rc);
return 0;
case OBD_LOG_CANCEL: /* remove this eventually - for 1.4.0 compat */
CERROR("shouldn't be handling OBD_LOG_CANCEL on DLM thread\n");
req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
return 0;
rc = llog_origin_handle_cancel(req);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
return 0;
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_CREATE:
req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_open(req);
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
req_capsule_set(&req->rq_pill,
&RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_next_block(req);
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_READ_HEADER:
req_capsule_set(&req->rq_pill,
&RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_read_header(req);
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_CLOSE:
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_close(req);
ldlm_callback_reply(req, rc);
return 0;
case OBD_QC_CALLBACK:
req_capsule_set(&req->rq_pill, &RQF_QC_CALLBACK);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_QC_CALLBACK_NET))
return 0;
rc = ldlm_handle_qc_callback(req);
ldlm_callback_reply(req, rc);
return 0;
default:
CERROR("unknown opcode %u\n",
lustre_msg_get_opc(req->rq_reqmsg));
ldlm_callback_reply(req, -EPROTO);
return 0;
}
ns = req->rq_export->exp_obd->obd_namespace;
LASSERT(ns != NULL);
req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK);
dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
if (dlm_req == NULL) {
rc = ldlm_callback_reply(req, -EPROTO);
ldlm_callback_errmsg(req, "Operate without parameter", rc,
NULL);
return 0;
}
/* Force a known safe race, send a cancel to the server for a lock
* which the server has already started a blocking callback on. */
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE) &&
lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) {
rc = ldlm_cli_cancel(&dlm_req->lock_handle[0], 0);
if (rc < 0)
CERROR("ldlm_cli_cancel: %d\n", rc);
}
lock = ldlm_handle2lock_long(&dlm_req->lock_handle[0], 0);
if (!lock) {
CDEBUG(D_DLMTRACE, "callback on lock "LPX64" - lock "
"disappeared\n", dlm_req->lock_handle[0].cookie);
rc = ldlm_callback_reply(req, -EINVAL);
ldlm_callback_errmsg(req, "Operate with invalid parameter", rc,
&dlm_req->lock_handle[0]);
return 0;
}
if ((lock->l_flags & LDLM_FL_FAIL_LOC) &&
lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK)
OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
/* Copy hints/flags (e.g. LDLM_FL_DISCARD_DATA) from AST. */
lock_res_and_lock(lock);
lock->l_flags |= ldlm_flags_from_wire(dlm_req->lock_flags &
LDLM_AST_FLAGS);
if (lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) {
/* If somebody cancels lock and cache is already dropped,
* or lock is failed before cp_ast received on client,
* we can tell the server we have no lock. Otherwise, we
* should send cancel after dropping the cache. */
if (((lock->l_flags & LDLM_FL_CANCELING) &&
(lock->l_flags & LDLM_FL_BL_DONE)) ||
(lock->l_flags & LDLM_FL_FAILED)) {
LDLM_DEBUG(lock, "callback on lock "
LPX64" - lock disappeared\n",
dlm_req->lock_handle[0].cookie);
unlock_res_and_lock(lock);
LDLM_LOCK_RELEASE(lock);
rc = ldlm_callback_reply(req, -EINVAL);
ldlm_callback_errmsg(req, "Operate on stale lock", rc,
&dlm_req->lock_handle[0]);
return 0;
}
/* BL_AST locks are not needed in LRU.
* Let ldlm_cancel_lru() be fast. */
ldlm_lock_remove_from_lru(lock);
lock->l_flags |= LDLM_FL_BL_AST;
}
unlock_res_and_lock(lock);
/* We want the ost thread to get this reply so that it can respond
* to ost requests (write cache writeback) that might be triggered
* in the callback.
*
* But we'd also like to be able to indicate in the reply that we're
* cancelling right now, because it's unused, or have an intent result
* in the reply, so we might have to push the responsibility for sending
* the reply down into the AST handlers, alas. */
switch (lustre_msg_get_opc(req->rq_reqmsg)) {
case LDLM_BL_CALLBACK:
CDEBUG(D_INODE, "blocking ast\n");
req_capsule_extend(&req->rq_pill, &RQF_LDLM_BL_CALLBACK);
if (!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)) {
rc = ldlm_callback_reply(req, 0);
if (req->rq_no_reply || rc)
ldlm_callback_errmsg(req, "Normal process", rc,
&dlm_req->lock_handle[0]);
}
if (ldlm_bl_to_thread_lock(ns, &dlm_req->lock_desc, lock))
ldlm_handle_bl_callback(ns, &dlm_req->lock_desc, lock);
break;
case LDLM_CP_CALLBACK:
CDEBUG(D_INODE, "completion ast\n");
req_capsule_extend(&req->rq_pill, &RQF_LDLM_CP_CALLBACK);
ldlm_callback_reply(req, 0);
ldlm_handle_cp_callback(req, ns, dlm_req, lock);
break;
case LDLM_GL_CALLBACK:
CDEBUG(D_INODE, "glimpse ast\n");
req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
ldlm_handle_gl_callback(req, ns, dlm_req, lock);
break;
default:
LBUG(); /* checked above */
}
return 0;
}
/**
* Callback handler for receiving incoming completion ASTs.
*
* This only can happen on client side.
*/
static void ldlm_handle_cp_callback(struct ptlrpc_request *req,
struct ldlm_namespace *ns,
struct ldlm_request *dlm_req,
struct ldlm_lock *lock)
{
int lvb_len;
LIST_HEAD(ast_list);
int rc = 0;
LDLM_DEBUG(lock, "client completion callback handler START");
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) {
int to = cfs_time_seconds(1);
while (to > 0) {
schedule_timeout_and_set_state(
TASK_INTERRUPTIBLE, to);
if (lock->l_granted_mode == lock->l_req_mode ||
lock->l_flags & LDLM_FL_DESTROYED)
break;
}
}
lvb_len = req_capsule_get_size(&req->rq_pill, &RMF_DLM_LVB, RCL_CLIENT);
if (lvb_len < 0) {
LDLM_ERROR(lock, "Fail to get lvb_len, rc = %d", lvb_len);
GOTO(out, rc = lvb_len);
} else if (lvb_len > 0) {
if (lock->l_lvb_len > 0) {
/* for extent lock, lvb contains ost_lvb{}. */
LASSERT(lock->l_lvb_data != NULL);
if (unlikely(lock->l_lvb_len < lvb_len)) {
LDLM_ERROR(lock, "Replied LVB is larger than "
"expectation, expected = %d, "
"replied = %d",
lock->l_lvb_len, lvb_len);
GOTO(out, rc = -EINVAL);
}
} else if (ldlm_has_layout(lock)) { /* for layout lock, lvb has
* variable length */
void *lvb_data;
OBD_ALLOC(lvb_data, lvb_len);
if (lvb_data == NULL) {
LDLM_ERROR(lock, "No memory: %d.\n", lvb_len);
GOTO(out, rc = -ENOMEM);
}
lock_res_and_lock(lock);
LASSERT(lock->l_lvb_data == NULL);
lock->l_lvb_data = lvb_data;
lock->l_lvb_len = lvb_len;
unlock_res_and_lock(lock);
}
}
lock_res_and_lock(lock);
if ((lock->l_flags & LDLM_FL_DESTROYED) ||
lock->l_granted_mode == lock->l_req_mode) {
/* bug 11300: the lock has already been granted */
unlock_res_and_lock(lock);
LDLM_DEBUG(lock, "Double grant race happened");
GOTO(out, rc = 0);
}
/* If we receive the completion AST before the actual enqueue returned,
* then we might need to switch lock modes, resources, or extents. */
if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
LDLM_DEBUG(lock, "completion AST, new lock mode");
}
if (lock->l_resource->lr_type != LDLM_PLAIN) {
ldlm_convert_policy_to_local(req->rq_export,
dlm_req->lock_desc.l_resource.lr_type,
&dlm_req->lock_desc.l_policy_data,
&lock->l_policy_data);
LDLM_DEBUG(lock, "completion AST, new policy data");
}
ldlm_resource_unlink_lock(lock);
if (memcmp(&dlm_req->lock_desc.l_resource.lr_name,
&lock->l_resource->lr_name,
sizeof(lock->l_resource->lr_name)) != 0) {
unlock_res_and_lock(lock);
rc = ldlm_lock_change_resource(ns, lock,
&dlm_req->lock_desc.l_resource.lr_name);
if (rc < 0) {
LDLM_ERROR(lock, "Failed to allocate resource");
GOTO(out, rc);
}
LDLM_DEBUG(lock, "completion AST, new resource");
CERROR("change resource!\n");
lock_res_and_lock(lock);
}
if (dlm_req->lock_flags & LDLM_FL_AST_SENT) {
/* BL_AST locks are not needed in LRU.
* Let ldlm_cancel_lru() be fast. */
ldlm_lock_remove_from_lru(lock);
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_BL_AST;
LDLM_DEBUG(lock, "completion AST includes blocking AST");
}
if (lock->l_lvb_len > 0) {
rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_CLIENT,
lock->l_lvb_data, lvb_len);
if (rc < 0) {
unlock_res_and_lock(lock);
GOTO(out, rc);
}
}
ldlm_grant_lock(lock, &ast_list);
unlock_res_and_lock(lock);
LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
/* Let Enqueue to call osc_lock_upcall() and initialize
* l_ast_data */
OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 2);
ldlm_run_ast_work(ns, &ast_list, LDLM_WORK_CP_AST);
LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
lock);
GOTO(out, rc);
out:
if (rc < 0) {
lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_FAILED;
unlock_res_and_lock(lock);
wake_up(&lock->l_waitq);
}
LDLM_LOCK_RELEASE(lock);
}
