/*
* Blocking callback handler for global index lock
*
* \param lock - is the lock for which ast occurred.
* \param desc - is the description of a conflicting lock in case of blocking
* ast.
* \param data - is the value of lock->l_ast_data
* \param flag - LDLM_CB_BLOCKING or LDLM_CB_CANCELING. Used to distinguish
* cancellation and blocking ast's.
*/
static int qsd_glb_blocking_ast(struct ldlm_lock *lock,
struct ldlm_lock_desc *desc, void *data,
int flag)
{
int rc = 0;
ENTRY;
switch(flag) {
case LDLM_CB_BLOCKING: {
struct lustre_handle lockh;
LDLM_DEBUG(lock, "blocking AST on global quota lock");
ldlm_lock2handle(lock, &lockh);
rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
break;
}
case LDLM_CB_CANCELING: {
struct qsd_qtype_info *qqi;
LDLM_DEBUG(lock, "canceling global quota lock");
qqi = qsd_glb_ast_data_get(lock, true);
if (qqi == NULL)
break;
/* we are losing the global index lock, so let's mark the
* global & slave indexes as not up-to-date any more */
write_lock(&qqi->qqi_qsd->qsd_lock);
qqi->qqi_glb_uptodate = false;
qqi->qqi_slv_uptodate = false;
if (lock->l_handle.h_cookie == qqi->qqi_lockh.cookie)
memset(&qqi->qqi_lockh, 0, sizeof(qqi->qqi_lockh));
write_unlock(&qqi->qqi_qsd->qsd_lock);
CDEBUG(D_QUOTA, "%s: losing global index lock for %s type\n",
qqi->qqi_qsd->qsd_svname, QTYPE_NAME((qqi->qqi_qtype)));
/* kick off reintegration thread if not running already, if
* it's just local cancel (for stack clean up or eviction),
* don't re-trigger the reintegration. */
if (!ldlm_is_local_only(lock))
qsd_start_reint_thread(qqi);
lu_ref_del(&qqi->qqi_reference, "ast_data_get", lock);
qqi_putref(qqi);
break;
}
default:
LASSERTF(0, "invalid flags for blocking ast %d", flag);
}
RETURN(rc);
}
static void mdc_close_intent_pack(struct ptlrpc_request *req,
struct md_op_data *op_data)
{
struct close_data *data;
struct ldlm_lock *lock;
enum mds_op_bias bias = op_data->op_bias;
if (!(bias & (MDS_CLOSE_INTENT | MDS_CLOSE_MIGRATE)))
return;
data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
LASSERT(data != NULL);
lock = ldlm_handle2lock(&op_data->op_lease_handle);
if (lock != NULL) {
data->cd_handle = lock->l_remote_handle;
LDLM_LOCK_PUT(lock);
}
ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
data->cd_data_version = op_data->op_data_version;
data->cd_fid = op_data->op_fid2;
if (bias & MDS_CLOSE_LAYOUT_SPLIT) {
data->cd_mirror_id = op_data->op_mirror_id;
} else if (bias & MDS_CLOSE_RESYNC_DONE) {
struct close_data_resync_done *sync = &data->cd_resync;
CLASSERT(sizeof(data->cd_resync) <= sizeof(data->cd_reserved));
sync->resync_count = op_data->op_data_size / sizeof(__u32);
if (sync->resync_count <= INLINE_RESYNC_ARRAY_SIZE) {
memcpy(sync->resync_ids_inline, op_data->op_data,
op_data->op_data_size);
} else {
size_t count = sync->resync_count;
memcpy(req_capsule_client_get(&req->rq_pill, &RMF_U32),
op_data->op_data, count * sizeof(__u32));
}
}
}
/* 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;
}
int llu_md_blocking_ast(struct ldlm_lock *lock,
struct ldlm_lock_desc *desc,
void *data, int flag)
{
struct lustre_handle lockh;
int rc;
ENTRY;
switch (flag) {
case LDLM_CB_BLOCKING:
ldlm_lock2handle(lock, &lockh);
rc = ldlm_cli_cancel(&lockh, 0);
if (rc < 0) {
CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
RETURN(rc);
}
break;
case LDLM_CB_CANCELING: {
struct inode *inode = llu_inode_from_resource_lock(lock);
struct llu_inode_info *lli;
struct intnl_stat *st;
__u64 bits = lock->l_policy_data.l_inodebits.bits;
struct lu_fid *fid;
/* Inode is set to lock->l_resource->lr_lvb_inode
* for mdc - bug 24555 */
LASSERT(lock->l_ast_data == NULL);
/* Invalidate all dentries associated with this inode */
if (inode == NULL)
break;
lli = llu_i2info(inode);
st = llu_i2stat(inode);
if (bits & MDS_INODELOCK_UPDATE)
lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
fid = &lli->lli_fid;
if (!fid_res_name_eq(fid, &lock->l_resource->lr_name))
LDLM_ERROR(lock, "data mismatch with object "
DFID" (%p)", PFID(fid), inode);
if (S_ISDIR(st->st_mode) &&
(bits & MDS_INODELOCK_UPDATE)) {
CDEBUG(D_INODE, "invalidating inode %llu\n",
(long long)st->st_ino);
llu_invalidate_inode_pages(inode);
}
/*
if (inode->i_sb->s_root &&
inode != inode->i_sb->s_root->d_inode)
ll_unhash_aliases(inode);
*/
I_RELE(inode);
break;
}
default:
LBUG();
}
RETURN(0);
}
/**
* Blocking callback handler for per-ID lock
*
* \param lock - is the lock for which ast occurred.
* \param desc - is the description of a conflicting lock in case of blocking
* ast.
* \param data - is the value of lock->l_ast_data
* \param flag - LDLM_CB_BLOCKING or LDLM_CB_CANCELING. Used to distinguish
* cancellation and blocking ast's.
*/
static int qsd_id_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
void *data, int flag)
{
struct lustre_handle lockh;
int rc = 0;
ENTRY;
switch(flag) {
case LDLM_CB_BLOCKING: {
LDLM_DEBUG(lock, "blocking AST on ID quota lock");
ldlm_lock2handle(lock, &lockh);
rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
break;
}
case LDLM_CB_CANCELING: {
struct lu_env *env;
struct lquota_entry *lqe;
bool rel = false;
LDLM_DEBUG(lock, "canceling global quota lock");
lqe = qsd_id_ast_data_get(lock, true);
if (lqe == NULL)
break;
LQUOTA_DEBUG(lqe, "losing ID lock");
/* just local cancel (for stack clean up or eviction), don't
* release quota space in this case */
if (ldlm_is_local_only(lock)) {
lqe_putref(lqe);
break;
}
/* allocate environment */
OBD_ALLOC_PTR(env);
if (env == NULL) {
lqe_putref(lqe);
rc = -ENOMEM;
break;
}
/* initialize environment */
rc = lu_env_init(env, LCT_DT_THREAD);
if (rc) {
OBD_FREE_PTR(env);
lqe_putref(lqe);
break;
}
ldlm_lock2handle(lock, &lockh);
lqe_write_lock(lqe);
if (lustre_handle_equal(&lockh, &lqe->lqe_lockh)) {
/* Clear lqe_lockh & reset qunit to 0 */
qsd_set_qunit(lqe, 0);
memset(&lqe->lqe_lockh, 0, sizeof(lqe->lqe_lockh));
lqe->lqe_edquot = false;
rel = true;
}
lqe_write_unlock(lqe);
/* If there is qqacq inflight, the release will be skipped
* at this time, and triggered on dqacq completion later,
* which means there could be a short window that slave is
* holding spare grant wihtout per-ID lock. */
if (rel)
rc = qsd_adjust(env, lqe);
/* release lqe reference grabbed by qsd_id_ast_data_get() */
lqe_putref(lqe);
lu_env_fini(env);
OBD_FREE_PTR(env);
break;
}
default:
LASSERTF(0, "invalid flags for blocking ast %d", flag);
}
RETURN(rc);
}
/**
* Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
* and ldlm_lock caches.
*/
static int osc_dlm_blocking_ast0(const struct lu_env *env,
struct ldlm_lock *dlmlock,
void *data, int flag)
{
struct osc_lock *olck;
struct cl_lock *lock;
int result;
int cancel;
LASSERT(flag == LDLM_CB_BLOCKING || flag == LDLM_CB_CANCELING);
cancel = 0;
olck = osc_ast_data_get(dlmlock);
if (olck != NULL) {
lock = olck->ols_cl.cls_lock;
cl_lock_mutex_get(env, lock);
LINVRNT(osc_lock_invariant(olck));
if (olck->ols_ast_wait) {
/* wake up osc_lock_use() */
cl_lock_signal(env, lock);
olck->ols_ast_wait = 0;
}
/*
* Lock might have been canceled while this thread was
* sleeping for lock mutex, but olck is pinned in memory.
*/
if (olck == dlmlock->l_ast_data) {
/*
* NOTE: DLM sends blocking AST's for failed locks
* (that are still in pre-OLS_GRANTED state)
* too, and they have to be canceled otherwise
* DLM lock is never destroyed and stuck in
* the memory.
*
* Alternatively, ldlm_cli_cancel() can be
* called here directly for osc_locks with
* ols_state < OLS_GRANTED to maintain an
* invariant that ->clo_cancel() is only called
* for locks that were granted.
*/
LASSERT(data == olck);
osc_lock_blocking(env, dlmlock,
olck, flag == LDLM_CB_BLOCKING);
} else
cancel = 1;
cl_lock_mutex_put(env, lock);
osc_ast_data_put(env, olck);
} else
/*
* DLM lock exists, but there is no cl_lock attached to it.
* This is a `normal' race. cl_object and its cl_lock's can be
* removed by memory pressure, together with all pages.
*/
cancel = (flag == LDLM_CB_BLOCKING);
if (cancel) {
struct lustre_handle *lockh;
lockh = &osc_env_info(env)->oti_handle;
ldlm_lock2handle(dlmlock, lockh);
result = ldlm_cli_cancel(lockh, LCF_ASYNC);
} else
result = 0;
return result;
}