* GPL HEADER END
*/
/*
- * Copyright (c) 2011, 2012, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* lustre/target/tgt_handler.c
#define DEBUG_SUBSYSTEM S_CLASS
+#include <linux/user_namespace.h>
+#ifdef HAVE_UIDGID_HEADER
+# include <linux/uidgid.h>
+#endif
+
#include <obd.h>
#include <obd_class.h>
#include <obd_cksum.h>
#include <md_object.h>
#include <lustre_lfsck.h>
+#include <lustre_nodemap.h>
#include "tgt_internal.h"
tsi->tsi_mdt_body = body;
- if (!(body->valid & OBD_MD_FLID))
+ if (!(body->mbo_valid & OBD_MD_FLID))
RETURN(0);
/* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
* in any case in pre-2.5 clients. Fix that here if needed */
- if (unlikely(fid_is_zero(&body->fid1)))
+ if (unlikely(fid_is_zero(&body->mbo_fid1)))
RETURN(0);
- if (!fid_is_sane(&body->fid1)) {
+ if (!fid_is_sane(&body->mbo_fid1)) {
CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
- PFID(&body->fid1));
+ PFID(&body->mbo_fid1));
RETURN(-EINVAL);
}
obj = lu_object_find(tsi->tsi_env,
&tsi->tsi_tgt->lut_bottom->dd_lu_dev,
- &body->fid1, NULL);
+ &body->mbo_fid1, NULL);
if (!IS_ERR(obj)) {
if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
lu_object_put(tsi->tsi_env, obj);
- /* for capability renew ENOENT will be handled in
- * mdt_renew_capa */
- if (body->valid & OBD_MD_FLOSSCAPA)
- rc = 0;
- else
- rc = -ENOENT;
+ rc = -ENOENT;
} else {
tsi->tsi_corpus = obj;
rc = 0;
rc = PTR_ERR(obj);
}
- tsi->tsi_fid = body->fid1;
+ tsi->tsi_fid = body->mbo_fid1;
RETURN(rc);
}
int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
{
struct ost_id *oi = &oa->o_oi;
- obd_seq seq = ostid_seq(oi);
- obd_id id = ostid_id(oi);
+ u64 seq = ostid_seq(oi);
+ u64 id = ostid_id(oi);
int rc;
ENTRY;
{
struct ost_body *body;
struct req_capsule *pill = tsi->tsi_pill;
- struct lustre_capa *capa;
+ struct lu_nodemap *nodemap;
int rc;
ENTRY;
if (rc)
RETURN(rc);
- if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
- capa = req_capsule_client_get(pill, &RMF_CAPA1);
- if (capa == NULL) {
- CERROR("%s: OSSCAPA flag is set without capability\n",
- tgt_name(tsi->tsi_tgt));
- RETURN(-EFAULT);
- }
- }
+ nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
+
+ body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
+ NODEMAP_CLIENT_TO_FS,
+ body->oa.o_uid);
+ body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
+ NODEMAP_CLIENT_TO_FS,
+ body->oa.o_gid);
tsi->tsi_ost_body = body;
tsi->tsi_fid = body->oa.o_oi.oi_fid;
{
int serious = 0;
int rc;
+ __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
ENTRY;
+
+ /* When dealing with sec context requests, no export is associated yet,
+ * because these requests are sent before *_CONNECT requests.
+ * A NULL req->rq_export means the normal *_common_slice handlers will
+ * not be called, because there is no reference to the target.
+ * So deal with them by hand and jump directly to target_send_reply().
+ */
+ switch (opc) {
+ case SEC_CTX_INIT:
+ case SEC_CTX_INIT_CONT:
+ case SEC_CTX_FINI:
+ CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
+ GOTO(out, rc = 0);
+ }
+
/*
* Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
* to put same checks into handlers like mdt_close(), mdt_reint(),
*/
if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
RETURN(0);
+ if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
+ OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
+ RETURN(0);
rc = tgt_request_preprocess(tsi, h, req);
/* pack reply if reply format is fixed */
RCL_SERVER))
req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
RCL_SERVER,
- tsi->tsi_mdt_body->eadatasize);
+ tsi->tsi_mdt_body->mbo_eadatasize);
if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
RCL_SERVER))
req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
if (likely(rc == 0 && req->rq_export))
target_committed_to_req(req);
+out:
target_send_reply(req, rc, tsi->tsi_reply_fail_id);
RETURN(0);
}
*process = 1;
RETURN(0);
case MDS_CLOSE:
- case MDS_DONE_WRITING:
case MDS_SYNC: /* used in unmounting */
case OBD_PING:
case MDS_REINT:
case OUT_UPDATE:
case SEQ_QUERY:
case FLD_QUERY:
+ case FLD_READ:
case LDLM_ENQUEUE:
case OST_CREATE:
case OST_DESTROY:
case OST_SETATTR:
case OST_SYNC:
case OST_WRITE:
+ case MDS_HSM_PROGRESS:
+ case MDS_HSM_STATE_SET:
+ case MDS_HSM_REQUEST:
*process = target_queue_recovery_request(req, obd);
RETURN(0);
* -ve: abort immediately with the given error code;
* 0: send reply with error code in req->rq_status;
*/
-int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
+static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
{
ENTRY;
/* sanity check: if the xid matches, the request must be marked as a
* resent or replayed */
- if (req_xid_is_last(req)) {
+ if (req_can_reconstruct(req, NULL)) {
if (!(lustre_msg_get_flags(req->rq_reqmsg) &
(MSG_RESENT | MSG_REPLAY))) {
DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
- "last_xid, expected REPLAY or RESENT flag "
+ "saved xid, expected REPLAY or RESENT flag "
"(%x)", req->rq_xid,
lustre_msg_get_flags(req->rq_reqmsg));
req->rq_status = -ENOTCONN;
ENTRY;
tgt = class_exp2tgt(req->rq_export);
+ if (unlikely(tgt == NULL)) {
+ DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
+ class_exp2obd(req->rq_export)->obd_name);
+ RETURN(ERR_PTR(-EINVAL));
+ }
for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
RETURN(h);
}
+static int process_req_last_xid(struct ptlrpc_request *req)
+{
+ __u64 last_xid;
+ ENTRY;
+
+ /* check request's xid is consistent with export's last_xid */
+ last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
+ if (last_xid > req->rq_export->exp_last_xid)
+ req->rq_export->exp_last_xid = last_xid;
+
+ if (req->rq_xid == 0 ||
+ (req->rq_xid <= req->rq_export->exp_last_xid)) {
+ DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. "
+ "last_xid %llx\n", req->rq_xid,
+ req->rq_export->exp_last_xid);
+ /* Some request is allowed to be sent during replay,
+ * such as OUT update requests, FLD requests, so it
+ * is possible that replay requests has smaller XID
+ * than the exp_last_xid.
+ *
+ * Some non-replay requests may have smaller XID as
+ * well:
+ *
+ * - Client send a no_resend RPC, like statfs;
+ * - The RPC timedout (or some other error) on client,
+ * then it's removed from the unreplied list;
+ * - Client send some other request to bump the
+ * exp_last_xid on server;
+ * - The former RPC got chance to be processed;
+ */
+ if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
+ RETURN(-EPROTO);
+ }
+
+ /* try to release in-memory reply data */
+ if (tgt_is_multimodrpcs_client(req->rq_export)) {
+ tgt_handle_received_xid(req->rq_export,
+ lustre_msg_get_last_xid(req->rq_reqmsg));
+ if (!(lustre_msg_get_flags(req->rq_reqmsg) &
+ (MSG_RESENT | MSG_REPLAY)))
+ tgt_handle_tag(req->rq_export,
+ lustre_msg_get_tag(req->rq_reqmsg));
+ }
+ RETURN(0);
+}
+
int tgt_request_handle(struct ptlrpc_request *req)
{
struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
struct lu_target *tgt;
int request_fail_id = 0;
__u32 opc = lustre_msg_get_opc(msg);
+ struct obd_device *obd;
int rc;
-
+ bool is_connect = false;
ENTRY;
/* Refill the context, to make sure all thread keys are allocated */
* target, otherwise that should be connect operation */
if (opc == MDS_CONNECT || opc == OST_CONNECT ||
opc == MGS_CONNECT) {
+ is_connect = true;
req_capsule_set(&req->rq_pill, &RQF_CONNECT);
rc = target_handle_connect(req);
if (rc != 0) {
}
if (unlikely(!class_connected_export(req->rq_export))) {
+ if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
+ opc == SEC_CTX_FINI) {
+ /* sec context initialization has to be handled
+ * by hand in tgt_handle_request0() */
+ tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
+ h = NULL;
+ GOTO(handle_recov, rc = 0);
+ }
CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
opc, libcfs_id2str(req->rq_peer));
req->rq_status = -ENOTCONN;
else
tsi->tsi_jobid = NULL;
+ if (tgt == NULL) {
+ DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
+ class_exp2obd(req->rq_export)->obd_name);
+ req->rq_status = -EINVAL;
+ rc = ptlrpc_error(req);
+ GOTO(out, rc);
+ }
+
+ /* Skip last_xid processing for the recovery thread, otherwise, the
+ * last_xid on same request could be processed twice: first time when
+ * processing the incoming request, second time when the request is
+ * being processed by recovery thread. */
+ obd = class_exp2obd(req->rq_export);
+ if (is_connect) {
+ /* reset the exp_last_xid on each connection. */
+ req->rq_export->exp_last_xid = 0;
+ } else if (obd->obd_recovery_data.trd_processing_task !=
+ current_pid()) {
+ rc = process_req_last_xid(req);
+ if (rc) {
+ req->rq_status = rc;
+ rc = ptlrpc_error(req);
+ GOTO(out, rc);
+ }
+ }
+
request_fail_id = tgt->lut_request_fail_id;
tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
GOTO(out, rc);
}
+ LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
+ h->th_opc, opc);
+
if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
GOTO(out, rc = 0);
GOTO(out, rc);
}
+handle_recov:
rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
if (likely(rc == 1)) {
- LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
- h->th_opc, opc);
rc = tgt_handle_request0(tsi, h, req);
if (rc)
GOTO(out, rc);
static inline void tgt_init_sec_none(struct obd_connect_data *reply)
{
reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
- OBD_CONNECT_RMT_CLIENT_FORCE |
- OBD_CONNECT_MDS_CAPA |
- OBD_CONNECT_OSS_CAPA);
+ OBD_CONNECT_RMT_CLIENT_FORCE);
}
static int tgt_init_sec_level(struct ptlrpc_request *req)
{
struct lu_target *tgt = class_exp2tgt(req->rq_export);
- char *client = libcfs_nid2str(req->rq_peer.nid);
+ char *client;
struct obd_connect_data *data, *reply;
int rc = 0;
bool remote;
-
ENTRY;
data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
RETURN(0);
}
+ if (unlikely(tgt == NULL)) {
+ DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
+ class_exp2obd(req->rq_export)->obd_name);
+ RETURN(-EINVAL);
+ }
+
+ client = libcfs_nid2str(req->rq_peer.nid);
/* no GSS support case */
if (!req->rq_auth_gss) {
if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
"as remote by default.\n", client, tgt_name(tgt));
}
- if (remote) {
- if (!tgt->lut_oss_capa) {
- CDEBUG(D_SEC,
- "client %s -> target %s is set as remote,"
- " but OSS capabilities are not enabled: %d.\n",
- client, tgt_name(tgt), tgt->lut_oss_capa);
- RETURN(-EACCES);
- }
- } else {
+ if (remote == 0) {
if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
CDEBUG(D_SEC, "client %s -> target %s: user is not "
"authenticated!\n", client, tgt_name(tgt));
break;
reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
OBD_CONNECT_RMT_CLIENT_FORCE);
- if (!tgt->lut_oss_capa)
- reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
- if (!tgt->lut_mds_capa)
- reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
+ reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
+ reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
break;
default:
RETURN(-EINVAL);
spin_lock(&exp->exp_lock);
exp->exp_sp_peer = req->rq_sp_from;
exp->exp_flvr = flvr;
+
+ /* when on mgs, if no restriction is set, or if client
+ * is loopback, allow any flavor */
+ if ((strcmp(exp->exp_obd->obd_type->typ_name,
+ LUSTRE_MGS_NAME) == 0) &&
+ (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
+ LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
+ == LOLND))
+ exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
+
if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
CERROR("%s: unauthorized rpc flavor %x from %s, "
struct sptlrpc_rule_set tmp_rset;
int rc;
+ if (unlikely(tgt == NULL)) {
+ CERROR("No target passed");
+ return -EINVAL;
+ }
+
sptlrpc_rule_set_init(&tmp_rset);
rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
if (rc) {
{
return err_serious(-EOPNOTSUPP);
}
-EXPORT_SYMBOL(tgt_obd_log_cancel);
-int tgt_obd_qc_callback(struct tgt_session_info *tsi)
+int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
{
- return err_serious(-EOPNOTSUPP);
+ struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
+ struct obd_export *exp = req->rq_export;
+ struct ptlrpc_bulk_desc *desc;
+ struct l_wait_info *lwi = &tti->tti_u.update.tti_wait_info;
+ int i;
+ int rc;
+
+ ENTRY;
+
+ desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
+ PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
+ MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
+ if (desc == NULL)
+ RETURN(-ENOMEM);
+
+ for (i = 0; i < rdbuf->rb_nbufs; i++)
+ desc->bd_frag_ops->add_iov_frag(desc,
+ rdbuf->rb_bufs[i].lb_buf,
+ rdbuf->rb_bufs[i].lb_len);
+
+ rc = target_bulk_io(exp, desc, lwi);
+ ptlrpc_free_bulk(desc);
+ RETURN(rc);
}
-EXPORT_SYMBOL(tgt_obd_qc_callback);
+EXPORT_SYMBOL(tgt_send_buffer);
int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
{
ENTRY;
- desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
- MDS_BULK_PORTAL);
+ desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
+ PTLRPC_BULK_PUT_SOURCE |
+ PTLRPC_BULK_BUF_KIOV,
+ MDS_BULK_PORTAL,
+ &ptlrpc_bulk_kiov_pin_ops);
if (desc == NULL)
RETURN(-ENOMEM);
for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
i++, tmpcount -= tmpsize) {
tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
- ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
+ desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
+ tmpsize);
}
LASSERT(desc->bd_nob == nob);
rc = target_bulk_io(exp, desc, lwi);
- ptlrpc_free_bulk_pin(desc);
+ ptlrpc_free_bulk(desc);
RETURN(rc);
}
EXPORT_SYMBOL(tgt_sendpage);
/*
* OBD_IDX_READ handler
*/
-int tgt_obd_idx_read(struct tgt_session_info *tsi)
+static int tgt_obd_idx_read(struct tgt_session_info *tsi)
{
struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
}
return rc;
}
-EXPORT_SYMBOL(tgt_obd_idx_read);
struct tgt_handler tgt_obd_handlers[] = {
TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
-TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
};
EXPORT_SYMBOL(tgt_obd_handlers);
int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
- struct dt_object *obj)
+ struct dt_object *obj, __u64 start, __u64 end)
{
int rc = 0;
rc = dt_sync(env, tgt->lut_bottom);
} else if (dt_version_get(env, obj) >
tgt->lut_obd->obd_last_committed) {
- rc = dt_object_sync(env, obj);
+ rc = dt_object_sync(env, obj, start, end);
}
+ atomic_inc(&tgt->lut_sync_count);
RETURN(rc);
}
* Unified target DLM handlers.
*/
-/* Ensure that data and metadata are synced to the disk when lock is cancelled
- * (if requested) */
-int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
- void *data, int flag)
+/**
+ * Unified target BAST
+ *
+ * Ensure data and metadata are synced to disk when lock is canceled if Sync on
+ * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
+ * but if it's cross-MDT lock, because remote object version is not set, a
+ * filesystem sync is needed.
+ *
+ * \param lock server side lock
+ * \param desc lock desc
+ * \param data ldlm_cb_set_arg
+ * \param flag indicates whether this cancelling or blocking callback
+ * \retval 0 on success
+ * \retval negative number on error
+ */
+static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
+ void *data, int flag)
{
struct lu_env env;
struct lu_target *tgt;
- struct dt_object *obj;
+ struct dt_object *obj = NULL;
struct lu_fid fid;
int rc = 0;
tgt = class_exp2tgt(lock->l_export);
+ if (unlikely(tgt == NULL)) {
+ CDEBUG(D_ERROR, "%s: No target for connected export\n",
+ class_exp2obd(lock->l_export)->obd_name);
+ RETURN(-EINVAL);
+ }
+
if (flag == LDLM_CB_CANCELING &&
- (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
+ (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
(tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
(tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
- lock->l_flags & LDLM_FL_CBPENDING))) {
+ ldlm_is_cbpending(lock))) &&
+ ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
+ lock->l_resource->lr_type == LDLM_EXTENT)) {
+ __u64 start = 0;
+ __u64 end = OBD_OBJECT_EOF;
+
rc = lu_env_init(&env, LCT_DT_THREAD);
if (unlikely(rc != 0))
RETURN(rc);
ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
tgt->lut_lsd.lsd_osd_index);
- obj = dt_locate(&env, tgt->lut_bottom, &fid);
- if (IS_ERR(obj))
- GOTO(err_env, rc = PTR_ERR(obj));
- if (!dt_object_exists(obj))
- GOTO(err_put, rc = -ENOENT);
+ if (lock->l_resource->lr_type == LDLM_EXTENT) {
+ obj = dt_locate(&env, tgt->lut_bottom, &fid);
+ if (IS_ERR(obj))
+ GOTO(err_env, rc = PTR_ERR(obj));
+
+ if (!dt_object_exists(obj))
+ GOTO(err_put, rc = -ENOENT);
+
+ start = lock->l_policy_data.l_extent.start;
+ end = lock->l_policy_data.l_extent.end;
+ }
- rc = tgt_sync(&env, tgt, obj);
+ rc = tgt_sync(&env, tgt, obj, start, end);
if (rc < 0) {
CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
"cancel: rc = %d\n",
lock->l_policy_data.l_extent.end, rc);
}
err_put:
- lu_object_put(&env, &obj->do_lu);
+ if (obj != NULL)
+ lu_object_put(&env, &obj->do_lu);
err_env:
lu_env_fini(&env);
}
RETURN(rc);
}
-struct ldlm_callback_suite tgt_dlm_cbs = {
+static struct ldlm_callback_suite tgt_dlm_cbs = {
.lcs_completion = ldlm_server_completion_ast,
.lcs_blocking = tgt_blocking_ast,
.lcs_glimpse = ldlm_server_glimpse_ast
if (rc)
RETURN(err_serious(rc));
+ switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
+ case LUT_FAIL_MDT:
+ tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
+ break;
+ case LUT_FAIL_OST:
+ tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
+ break;
+ case LUT_FAIL_MGT:
+ tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
+ break;
+ default:
+ tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
+ break;
+ }
RETURN(req->rq_status);
}
EXPORT_SYMBOL(tgt_enqueue);
RETURN(req->rq_status);
}
-EXPORT_SYMBOL(tgt_convert);
int tgt_bl_callback(struct tgt_session_info *tsi)
{
return err_serious(-EOPNOTSUPP);
}
-EXPORT_SYMBOL(tgt_bl_callback);
int tgt_cp_callback(struct tgt_session_info *tsi)
{
return err_serious(-EOPNOTSUPP);
}
-EXPORT_SYMBOL(tgt_cp_callback);
/* generic LDLM target handler */
struct tgt_handler tgt_dlm_handlers[] = {
RETURN(rc);
}
-EXPORT_SYMBOL(tgt_llog_destroy);
int tgt_llog_read_header(struct tgt_session_info *tsi)
{
* sec context handlers
*/
/* XXX: Implement based on mdt_sec_ctx_handle()? */
-int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
+static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
{
return 0;
}
int (*tgt_lfsck_in_notify)(const struct lu_env *env,
struct dt_device *key,
- struct lfsck_request *lr) = NULL;
+ struct lfsck_request *lr,
+ struct thandle *th) = NULL;
void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
struct dt_device *,
- struct lfsck_request *))
+ struct lfsck_request *,
+ struct thandle *))
{
tgt_lfsck_in_notify = notify;
}
static int (*tgt_lfsck_query)(const struct lu_env *env,
struct dt_device *key,
- struct lfsck_request *lr) = NULL;
+ struct lfsck_request *req,
+ struct lfsck_reply *rep,
+ struct lfsck_query *que) = NULL;
void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
struct dt_device *,
- struct lfsck_request *))
+ struct lfsck_request *,
+ struct lfsck_reply *,
+ struct lfsck_query *))
{
tgt_lfsck_query = query;
}
if (lr == NULL)
RETURN(-EPROTO);
- rc = tgt_lfsck_in_notify(env, key, lr);
+ rc = tgt_lfsck_in_notify(env, key, lr, NULL);
RETURN(rc);
}
{
struct lfsck_request *request;
struct lfsck_reply *reply;
- int rc = 0;
+ int rc;
ENTRY;
request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
if (reply == NULL)
RETURN(-ENOMEM);
- reply->lr_status = tgt_lfsck_query(tsi->tsi_env,
- tsi->tsi_tgt->lut_bottom, request);
- if (reply->lr_status < 0)
- rc = reply->lr_status;
+ rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
+ request, reply, NULL);
- RETURN(rc);
+ RETURN(rc < 0 ? rc : 0);
}
struct tgt_handler tgt_lfsck_handlers[] = {
__u64 start, __u64 end, struct lustre_handle *lh,
int mode, __u64 *flags)
{
- ldlm_policy_data_t policy;
- int rc;
+ union ldlm_policy_data policy;
+ int rc;
ENTRY;
LASSERT(!lustre_handle_is_used(lh));
policy.l_extent.gid = 0;
- policy.l_extent.start = start & CFS_PAGE_MASK;
+ policy.l_extent.start = start & PAGE_MASK;
/*
* If ->o_blocks is EOF it means "lock till the end of the file".
if (end == OBD_OBJECT_EOF || end < start)
policy.l_extent.end = OBD_OBJECT_EOF;
else
- policy.l_extent.end = end | ~CFS_PAGE_MASK;
+ policy.l_extent.end = end | ~PAGE_MASK;
rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
flags, ldlm_blocking_ast,
}
EXPORT_SYMBOL(tgt_extent_lock);
-void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
+void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
{
LASSERT(lustre_handle_is_used(lh));
ldlm_lock_decref(lh, mode);
int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
struct obd_ioobj *obj, struct niobuf_remote *nb,
- struct lustre_handle *lh, int mode)
+ struct lustre_handle *lh, enum ldlm_mode mode)
{
__u64 flags = 0;
int nrbufs = obj->ioo_bufcnt;
LASSERT(mode == LCK_PR || mode == LCK_PW);
LASSERT(!lustre_handle_is_used(lh));
- if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
+ if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
RETURN(0);
for (i = 1; i < nrbufs; i++)
- if (!(nb[i].flags & OBD_BRW_SRVLOCK))
+ if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
RETURN(-EFAULT);
- RETURN(tgt_extent_lock(ns, res_id, nb[0].offset,
- nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1,
+ RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
+ nb[nrbufs - 1].rnb_offset +
+ nb[nrbufs - 1].rnb_len - 1,
lh, mode, &flags));
}
-EXPORT_SYMBOL(tgt_brw_lock);
void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
- struct lustre_handle *lh, int mode)
+ struct lustre_handle *lh, enum ldlm_mode mode)
{
ENTRY;
LASSERT(mode == LCK_PR || mode == LCK_PW);
- LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
+ LASSERT((obj->ioo_bufcnt > 0 &&
+ (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
lustre_handle_is_used(lh));
+
if (lustre_handle_is_used(lh))
tgt_extent_unlock(lh, mode);
EXIT;
}
-EXPORT_SYMBOL(tgt_brw_unlock);
static __u32 tgt_checksum_bulk(struct lu_target *tgt,
struct ptlrpc_bulk_desc *desc, int opc,
unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
__u32 cksum;
+ LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
+
hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
if (IS_ERR(hdesc)) {
CERROR("%s: unable to initialize checksum hash %s\n",
* simulate a client->OST data error */
if (i == 0 && opc == OST_WRITE &&
OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
- int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
- int len = desc->bd_iov[i].kiov_len;
+ int off = BD_GET_KIOV(desc, i).kiov_offset &
+ ~PAGE_MASK;
+ int len = BD_GET_KIOV(desc, i).kiov_len;
struct page *np = tgt_page_to_corrupt;
- char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
+ char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
if (np) {
char *ptr2 = kmap(np) + off;
memcpy(ptr2, ptr, len);
memcpy(ptr2, "bad3", min(4, len));
kunmap(np);
- desc->bd_iov[i].kiov_page = np;
+ BD_GET_KIOV(desc, i).kiov_page = np;
} else {
CERROR("%s: can't alloc page for corruption\n",
tgt_name(tgt));
}
}
- cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
- desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
- desc->bd_iov[i].kiov_len);
+ cfs_crypto_hash_update_page(hdesc,
+ BD_GET_KIOV(desc, i).kiov_page,
+ BD_GET_KIOV(desc, i).kiov_offset &
+ ~PAGE_MASK,
+ BD_GET_KIOV(desc, i).kiov_len);
/* corrupt the data after we compute the checksum, to
* simulate an OST->client data error */
if (i == 0 && opc == OST_READ &&
OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
- int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
- int len = desc->bd_iov[i].kiov_len;
+ int off = BD_GET_KIOV(desc, i).kiov_offset
+ & ~PAGE_MASK;
+ int len = BD_GET_KIOV(desc, i).kiov_len;
struct page *np = tgt_page_to_corrupt;
- char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
+ char *ptr =
+ kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
if (np) {
char *ptr2 = kmap(np) + off;
memcpy(ptr2, ptr, len);
memcpy(ptr2, "bad4", min(4, len));
kunmap(np);
- desc->bd_iov[i].kiov_page = np;
+ BD_GET_KIOV(desc, i).kiov_page = np;
} else {
CERROR("%s: can't alloc page for corruption\n",
tgt_name(tgt));
}
}
- bufsize = 4;
+ bufsize = sizeof(cksum);
err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
- if (err)
- cfs_crypto_hash_final(hdesc, NULL, NULL);
return cksum;
}
struct ost_body *body, *repbody;
struct l_wait_info lwi;
struct lustre_handle lockh = { 0 };
- int niocount, npages, nob = 0, rc, i;
- int no_reply = 0;
+ int npages, nob = 0, rc, i, no_reply = 0;
struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
ENTRY;
ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
- niocount = ioo->ioo_bufcnt;
remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
npages = PTLRPC_MAX_BRW_PAGES;
rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
- ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
+ ioo, remote_nb, &npages, local_nb);
if (rc != 0)
GOTO(out_lock, rc);
desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
- BULK_PUT_SOURCE, OST_BULK_PORTAL);
+ PTLRPC_BULK_PUT_SOURCE |
+ PTLRPC_BULK_BUF_KIOV,
+ OST_BULK_PORTAL,
+ &ptlrpc_bulk_kiov_nopin_ops);
if (desc == NULL)
GOTO(out_commitrw, rc = -ENOMEM);
nob = 0;
for (i = 0; i < npages; i++) {
- int page_rc = local_nb[i].rc;
+ int page_rc = local_nb[i].lnb_rc;
if (page_rc < 0) {
rc = page_rc;
nob += page_rc;
if (page_rc != 0) { /* some data! */
- LASSERT(local_nb[i].page != NULL);
- ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
- local_nb[i].lnb_page_offset,
- page_rc);
+ LASSERT(local_nb[i].lnb_page != NULL);
+ desc->bd_frag_ops->add_kiov_frag
+ (desc, local_nb[i].lnb_page,
+ local_nb[i].lnb_page_offset,
+ page_rc);
}
- if (page_rc != local_nb[i].len) { /* short read */
+ if (page_rc != local_nb[i].lnb_len) { /* short read */
/* All subsequent pages should be 0 */
while (++i < npages)
- LASSERT(local_nb[i].rc == 0);
+ LASSERT(local_nb[i].lnb_rc == 0);
break;
}
}
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
+ nob != cfs_fail_val)
+ rc = -E2BIG;
if (body->oa.o_valid & OBD_MD_FLCKSUM) {
cksum_type_t cksum_type =
/* Check if client was evicted while we were doing i/o before touching
* network */
if (likely(rc == 0 &&
- !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
+ !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2) &&
+ !CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_BULK))) {
rc = target_bulk_io(exp, desc, &lwi);
no_reply = rc != 0;
}
out_commitrw:
/* Must commit after prep above in all cases */
- rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
- &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
- NULL, rc);
+ rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
+ remote_nb, npages, local_nb, rc);
if (rc == 0)
tgt_drop_id(exp, &repbody->oa);
out_lock:
tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
- ptlrpc_free_bulk_nopin(desc);
+ ptlrpc_free_bulk(desc);
LASSERT(rc <= 0);
if (rc == 0) {
lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
l_wait_event(waitq, 0, &lwi1);
target_bulk_io(exp, desc, &lwi);
- ptlrpc_free_bulk_nopin(desc);
+ ptlrpc_free_bulk(desc);
}
RETURN(rc);
static void tgt_warn_on_cksum(struct ptlrpc_request *req,
struct ptlrpc_bulk_desc *desc,
struct niobuf_local *local_nb, int npages,
- obd_count client_cksum, obd_count server_cksum,
+ u32 client_cksum, u32 server_cksum,
bool mmap)
{
struct obd_export *exp = req->rq_export;
struct ost_body *body;
- char *router;
- char *via;
+ char *router = "";
+ char *via = "";
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
LASSERT(body != NULL);
- if (req->rq_peer.nid == desc->bd_sender) {
- via = router = "";
- } else {
+ if (req->rq_peer.nid != desc->bd_sender) {
via = " via ";
router = libcfs_nid2str(desc->bd_sender);
}
POSTID(&body->oa.o_oi),
local_nb[0].lnb_file_offset,
local_nb[npages-1].lnb_file_offset +
- local_nb[npages-1].len - 1,
+ local_nb[npages - 1].lnb_len - 1,
client_cksum, server_cksum);
}
sizeof(*remote_nb))
RETURN(err_serious(-EPROTO));
- if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
+ if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
(exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
memory_pressure_set();
npages = PTLRPC_MAX_BRW_PAGES;
rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
- objcount, ioo, remote_nb, &npages, local_nb, NULL,
- BYPASS_CAPA);
+ objcount, ioo, remote_nb, &npages, local_nb);
if (rc < 0)
GOTO(out_lock, rc);
desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
- BULK_GET_SINK, OST_BULK_PORTAL);
+ PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
+ OST_BULK_PORTAL,
+ &ptlrpc_bulk_kiov_nopin_ops);
if (desc == NULL)
GOTO(skip_transfer, rc = -ENOMEM);
/* NB Having prepped, we must commit... */
for (i = 0; i < npages; i++)
- ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
- local_nb[i].lnb_page_offset,
- local_nb[i].len);
+ desc->bd_frag_ops->add_kiov_frag(desc,
+ local_nb[i].lnb_page,
+ local_nb[i].lnb_page_offset,
+ local_nb[i].lnb_len);
rc = sptlrpc_svc_prep_bulk(req, desc);
if (rc != 0)
/* Must commit after prep above in all cases */
rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
- objcount, ioo, remote_nb, npages, local_nb, NULL,
- rc);
+ objcount, ioo, remote_nb, npages, local_nb, rc);
if (rc == -ENOTCONN)
/* quota acquire process has been given up because
* either the client has been evicted or the client
/* set per-requested niobuf return codes */
for (i = j = 0; i < niocount; i++) {
- int len = remote_nb[i].len;
+ int len = remote_nb[i].rnb_len;
nob += len;
rcs[i] = 0;
do {
LASSERT(j < npages);
- if (local_nb[j].rc < 0)
- rcs[i] = local_nb[j].rc;
- len -= local_nb[j].len;
+ if (local_nb[j].lnb_rc < 0)
+ rcs[i] = local_nb[j].lnb_rc;
+ len -= local_nb[j].lnb_len;
j++;
} while (len > 0);
LASSERT(len == 0);
out_lock:
tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
if (desc)
- ptlrpc_free_bulk_nopin(desc);
+ ptlrpc_free_bulk(desc);
out:
if (no_reply) {
req->rq_no_reply = 1;
RETURN(rc);
}
EXPORT_SYMBOL(tgt_brw_write);
+
+/* Check if request can be reconstructed from saved reply data
+ * A copy of the reply data is returned in @trd if the pointer is not NULL
+ */
+bool req_can_reconstruct(struct ptlrpc_request *req,
+ struct tg_reply_data *trd)
+{
+ struct tg_export_data *ted = &req->rq_export->exp_target_data;
+ struct lsd_client_data *lcd = ted->ted_lcd;
+ bool found;
+
+ if (tgt_is_multimodrpcs_client(req->rq_export))
+ return tgt_lookup_reply(req, trd);
+
+ mutex_lock(&ted->ted_lcd_lock);
+ found = req->rq_xid == lcd->lcd_last_xid ||
+ req->rq_xid == lcd->lcd_last_close_xid;
+
+ if (found && trd != NULL) {
+ if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
+ trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
+ trd->trd_reply.lrd_transno =
+ lcd->lcd_last_close_transno;
+ trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
+ } else {
+ trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
+ trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
+ trd->trd_reply.lrd_result = lcd->lcd_last_result;
+ trd->trd_reply.lrd_data = lcd->lcd_last_data;
+ trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
+ trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
+ trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
+ trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
+ }
+ }
+ mutex_unlock(&ted->ted_lcd_lock);
+
+ return found;
+}
+EXPORT_SYMBOL(req_can_reconstruct);
+