static inline int out_check_resent(const struct lu_env *env,
struct dt_device *dt,
struct dt_object *obj,
struct ptlrpc_request *req,
out_reconstruct_t reconstruct,
struct object_update_reply *reply,
int index)
{
if (likely(!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)))
return 0;
if (req_xid_is_last(req)) {
struct lsd_client_data *lcd;
/* XXX this does not support mulitple transactions yet, i.e.
* only 1 update RPC each time betwee MDTs */
lcd = req->rq_export->exp_target_data.ted_lcd;
req->rq_transno = lcd->lcd_last_transno;
req->rq_status = lcd->lcd_last_result;
if (req->rq_status != 0)
req->rq_transno = 0;
lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
lustre_msg_set_status(req->rq_repmsg, req->rq_status);
DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"status %d",
req->rq_transno, req->rq_status);
reconstruct(env, dt, obj, reply, index);
return 1;
}
DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
return 0;
}
/**
* Disconnect a bulk desc from the network. Idempotent. Not
* thread-safe (i.e. only interlocks with completion callback).
* Returns 1 on success or 0 if network unregistration failed for whatever
* reason.
*/
int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async)
{
struct ptlrpc_bulk_desc *desc = req->rq_bulk;
wait_queue_head_t *wq;
int rc;
LASSERT(!in_interrupt()); /* might sleep */
/* Let's setup deadline for reply unlink. */
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
async && req->rq_bulk_deadline == 0 && cfs_fail_val == 0)
req->rq_bulk_deadline = ktime_get_real_seconds() + LONG_UNLINK;
if (ptlrpc_client_bulk_active(req) == 0) /* completed or */
return 1; /* never registered */
LASSERT(desc->bd_req == req); /* bd_req NULL until registered */
/* the unlink ensures the callback happens ASAP and is the last
* one. If it fails, it must be because completion just happened,
* but we must still wait_event() in this case to give liblustre
* a chance to run client_bulk_callback()
*/
mdunlink_iterate_helper(desc->bd_mds, desc->bd_md_max_brw);
if (ptlrpc_client_bulk_active(req) == 0) /* completed or */
return 1; /* never registered */
/* Move to "Unregistering" phase as bulk was not unlinked yet. */
ptlrpc_rqphase_move(req, RQ_PHASE_UNREG_BULK);
/* Do not wait for unlink to finish. */
if (async)
return 0;
if (req->rq_set)
wq = &req->rq_set->set_waitq;
else
wq = &req->rq_reply_waitq;
for (;;) {
/* Network access will complete in finite time but the HUGE
* timeout lets us CWARN for visibility of sluggish LNDs
*/
int cnt = 0;
while (cnt < LONG_UNLINK &&
(rc = wait_event_idle_timeout(*wq,
!ptlrpc_client_bulk_active(req),
HZ)) == 0)
cnt += 1;
if (rc > 0) {
ptlrpc_rqphase_move(req, req->rq_next_phase);
return 1;
}
DEBUG_REQ(D_WARNING, req, "Unexpectedly long timeout: desc %p",
desc);
}
return 0;
}
/*
* Client's outgoing request callback
*/
void request_out_callback(lnet_event_t *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_request *req = cbid->cbid_arg;
LASSERT(ev->type == LNET_EVENT_SEND ||
ev->type == LNET_EVENT_UNLINK);
LASSERT(ev->unlinked);
DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
sptlrpc_request_out_callback(req);
req->rq_real_sent = cfs_time_current_sec();
if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) {
/* Failed send: make it seem like the reply timed out, just
* like failing sends in client.c does currently... */
spin_lock(&req->rq_lock);
req->rq_net_err = 1;
spin_unlock(&req->rq_lock);
ptlrpc_client_wake_req(req);
}
ptlrpc_req_finished(req);
}
/**
* Disconnect a bulk desc from the network. Idempotent. Not
* thread-safe (i.e. only interlocks with completion callback).
* Returns 1 on success or 0 if network unregistration failed for whatever
* reason.
*/
int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async)
{
struct ptlrpc_bulk_desc *desc = req->rq_bulk;
struct l_wait_info lwi;
int rc;
ENTRY;
LASSERT(!in_interrupt()); /* might sleep */
/* Let's setup deadline for reply unlink. */
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
async && req->rq_bulk_deadline == 0)
req->rq_bulk_deadline = cfs_time_current_sec() + LONG_UNLINK;
if (ptlrpc_client_bulk_active(req) == 0) /* completed or */
RETURN(1); /* never registered */
LASSERT(desc->bd_req == req); /* bd_req NULL until registered */
/* the unlink ensures the callback happens ASAP and is the last
* one. If it fails, it must be because completion just happened,
* but we must still l_wait_event() in this case to give liblustre
* a chance to run client_bulk_callback() */
mdunlink_iterate_helper(desc->bd_mds, desc->bd_md_max_brw);
if (ptlrpc_client_bulk_active(req) == 0) /* completed or */
RETURN(1); /* never registered */
/* Move to "Unregistering" phase as bulk was not unlinked yet. */
ptlrpc_rqphase_move(req, RQ_PHASE_UNREGISTERING);
/* Do not wait for unlink to finish. */
if (async)
RETURN(0);
for (;;) {
#ifdef __KERNEL__
/* The wq argument is ignored by user-space wait_event macros */
wait_queue_head_t *wq = (req->rq_set != NULL) ?
&req->rq_set->set_waitq :
&req->rq_reply_waitq;
#endif
/* Network access will complete in finite time but the HUGE
* timeout lets us CWARN for visibility of sluggish NALs */
lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(LONG_UNLINK),
cfs_time_seconds(1), NULL, NULL);
rc = l_wait_event(*wq, !ptlrpc_client_bulk_active(req), &lwi);
if (rc == 0) {
ptlrpc_rqphase_move(req, req->rq_next_phase);
RETURN(1);
}
LASSERT(rc == -ETIMEDOUT);
DEBUG_REQ(D_WARNING, req, "Unexpectedly long timeout: desc %p",
desc);
}
RETURN(0);
}
/* Setinfo coming from Server (eg MDT) to Client (eg MDC)! */
static int ldlm_handle_setinfo(struct ptlrpc_request *req)
{
struct obd_device *obd = req->rq_export->exp_obd;
char *key;
void *val;
int keylen, vallen;
int rc = -ENOSYS;
DEBUG_REQ(D_HSM, req, "%s: handle setinfo\n", obd->obd_name);
req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
if (key == NULL) {
DEBUG_REQ(D_IOCTL, req, "no set_info key");
return -EFAULT;
}
keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
RCL_CLIENT);
val = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
if (val == NULL) {
DEBUG_REQ(D_IOCTL, req, "no set_info val");
return -EFAULT;
}
vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
RCL_CLIENT);
/* We are responsible for swabbing contents of val */
if (KEY_IS(KEY_HSM_COPYTOOL_SEND))
/* Pass it on to mdc (the "export" in this case) */
rc = obd_set_info_async(req->rq_svc_thread->t_env,
req->rq_export,
sizeof(KEY_HSM_COPYTOOL_SEND),
KEY_HSM_COPYTOOL_SEND,
vallen, val, NULL);
else
DEBUG_REQ(D_WARNING, req, "ignoring unknown key %s", key);
return rc;
}
static inline void ldlm_callback_errmsg(struct ptlrpc_request *req,
const char *msg, int rc,
struct lustre_handle *handle)
{
DEBUG_REQ((req->rq_no_reply || rc) ? D_WARNING : D_DLMTRACE, req,
"%s: [nid %s] [rc %d] [lock "LPX64"]",
msg, libcfs_id2str(req->rq_peer), rc,
handle ? handle->cookie : 0);
if (req->rq_no_reply)
CWARN("No reply was sent, maybe cause bug 21636.\n");
else if (rc)
CWARN("Send reply failed, maybe cause bug 21636.\n");
}
/* Partial request to create object only */
static int mdt_md_mkobj(struct mdt_thread_info *info)
{
struct mdt_device *mdt = info->mti_mdt;
struct mdt_object *o;
struct mdt_body *repbody;
struct md_attr *ma = &info->mti_attr;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, mdt_info_req(info), "Partial create "DFID"",
PFID(info->mti_rr.rr_fid2));
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
o = mdt_object_find(info->mti_env, mdt, info->mti_rr.rr_fid2,
MDT_OBJ_MAY_NOT_EXIST);
if (!IS_ERR(o)) {
struct md_object *next = mdt_object_child(o);
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
/*
* Cross-ref create can encounter already created obj in case of
* recovery, just get attr in that case.
*/
if (mdt_object_exists(o) == 1) {
rc = mo_attr_get(info->mti_env, next, ma);
} else {
/*
* Here, NO permission check for object_create,
* such check has been done on the original MDS.
*/
rc = mo_object_create(info->mti_env, next,
&info->mti_spec, ma);
}
if (rc == 0) {
/* Return fid & attr to client. */
if (ma->ma_valid & MA_INODE)
mdt_pack_attr2body(info, repbody, &ma->ma_attr,
mdt_object_fid(o));
}
mdt_object_put(info->mti_env, o);
} else
rc = PTR_ERR(o);
mdt_create_pack_capa(info, rc, o, repbody);
RETURN(rc);
}
static void ptlrpc_at_set_reply(struct ptlrpc_request *req, int flags)
{
struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
struct ptlrpc_service *svc = svcpt->scp_service;
int service_time = max_t(int, cfs_time_current_sec() -
req->rq_arrival_time.tv_sec, 1);
if (!(flags & PTLRPC_REPLY_EARLY) &&
(req->rq_type != PTL_RPC_MSG_ERR) &&
(req->rq_reqmsg != NULL) &&
!(lustre_msg_get_flags(req->rq_reqmsg) &
(MSG_RESENT | MSG_REPLAY |
MSG_REQ_REPLAY_DONE | MSG_LOCK_REPLAY_DONE))) {
/* early replies, errors and recovery requests don't count
* toward our service time estimate */
int oldse = at_measured(&svcpt->scp_at_estimate, service_time);
if (oldse != 0) {
DEBUG_REQ(D_ADAPTTO, req,
"svc %s changed estimate from %d to %d",
svc->srv_name, oldse,
at_get(&svcpt->scp_at_estimate));
}
}
/* Report actual service time for client latency calc */
lustre_msg_set_service_time(req->rq_repmsg, service_time);
/* Report service time estimate for future client reqs, but report 0
* (to be ignored by client) if it's a error reply during recovery.
* (bz15815) */
if (req->rq_type == PTL_RPC_MSG_ERR &&
(req->rq_export == NULL || req->rq_export->exp_obd->obd_recovering))
lustre_msg_set_timeout(req->rq_repmsg, 0);
else
lustre_msg_set_timeout(req->rq_repmsg,
at_get(&svcpt->scp_at_estimate));
if (req->rq_reqmsg &&
!(lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
CDEBUG(D_ADAPTTO, "No early reply support: flags=%#x "
"req_flags=%#x magic=%d:%x/%x len=%d\n",
flags, lustre_msg_get_flags(req->rq_reqmsg),
lustre_msg_is_v1(req->rq_reqmsg),
lustre_msg_get_magic(req->rq_reqmsg),
lustre_msg_get_magic(req->rq_repmsg), req->rq_replen);
}
}
static int ptlrpc_ping(struct obd_import *imp)
{
struct ptlrpc_request *req;
req = ptlrpc_prep_ping(imp);
if (!req) {
CERROR("OOM trying to ping %s->%s\n",
imp->imp_obd->obd_uuid.uuid,
obd2cli_tgt(imp->imp_obd));
return -ENOMEM;
}
DEBUG_REQ(D_INFO, req, "pinging %s->%s",
imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
ptlrpcd_add_req(req);
return 0;
}
static inline int out_check_resent(const struct lu_env *env,
struct dt_device *dt,
struct dt_object *obj,
struct ptlrpc_request *req,
out_reconstruct_t reconstruct,
struct update_reply *reply,
int index)
{
if (likely(!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)))
return 0;
if (req_xid_is_last(req)) {
reconstruct(env, dt, obj, reply, index);
return 1;
}
DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
return 0;
}
int ptlrpc_ping(struct obd_import *imp)
{
struct ptlrpc_request *req;
ENTRY;
req = ptlrpc_prep_ping(imp);
if (req == NULL) {
CERROR("OOM trying to ping %s->%s\n",
imp->imp_obd->obd_uuid.uuid,
obd2cli_tgt(imp->imp_obd));
RETURN(-ENOMEM);
}
DEBUG_REQ(D_INFO, req, "pinging %s->%s",
imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
RETURN(0);
}
/*
* Add a new non-blocking request to a non-blocking set.
*/
int
pscrpc_nbreqset_add(struct pscrpc_request_set *set,
struct pscrpc_request *rq)
{
int rc;
rq->rq_compl = &set->set_compl;
rc = pscrpc_push_req(rq);
if (rc) {
rq->rq_compl = NULL;
DEBUG_REQ(PLL_ERROR, rq, "send failure: %s",
pfl_strerror(rc));
} else {
/*
* Adding after is OK because it will sit on the list
* marked COMPLETED if it finishes before this add
* occurs.
*/
pscrpc_set_add_new_req(set, rq);
}
return (rc);
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
/*
* step 1: lock the parent. Note, this may be child in case of
* remote operation denoted by ->mti_cross_ref flag.
*/
parent_lh = &info->mti_lh[MDT_LH_PARENT];
if (info->mti_cross_ref) {
/*
* Init reg lock for cross ref case when we need to do only
* ref del locally.
*/
mdt_lock_reg_init(parent_lh, LCK_PW);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
}
mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
MDS_INODELOCK_UPDATE, MDT_OBJ_MUST_EXIST);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
/* errors are possible here in cross-ref cases, see below */
if (info->mti_cross_ref)
rc = 0;
GOTO(out, rc);
}
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(out_unlock_parent, rc);
mdt_reint_init_ma(info, ma);
if (!ma->ma_lmm || !ma->ma_cookie)
GOTO(out_unlock_parent, rc = -EINVAL);
if (info->mti_cross_ref) {
/*
* Remote partial operation. It is possible that replay may
* happen on parent MDT and this operation will be repeated.
* Therefore the object absense is allowed case and nothing
* should be done here.
*/
if (mdt_object_exists(mp) > 0) {
mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
rc = mo_ref_del(info->mti_env,
mdt_object_child(mp), ma);
if (rc == 0)
mdt_handle_last_unlink(info, mp, ma);
} else
rc = 0;
GOTO(out_unlock_parent, rc);
}
/* step 2: find & lock the child */
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
/* lookup child object along with version checking */
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0)
GOTO(out_unlock_parent, rc);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid,
MDT_OBJ_MUST_EXIST);
if (IS_ERR(mc))
GOTO(out_unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mc);
GOTO(out_unlock_parent, rc);
}
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req->rq_export, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req->rq_export, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
mdt_object_unlock_put(info, mc, child_lh, rc);
out_unlock_parent:
mdt_object_unlock_put(info, mp, parent_lh, rc);
out:
return rc;
}
/**
* Send request \a request.
* if \a noreply is set, don't expect any reply back and don't set up
* reply buffers.
* Returns 0 on success or error code.
*/
int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
{
int rc;
int rc2;
int mpflag = 0;
struct ptlrpc_connection *connection;
lnet_handle_me_t reply_me_h;
lnet_md_t reply_md;
struct obd_import *imp = request->rq_import;
struct obd_device *obd = imp->imp_obd;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_RPC))
RETURN(0);
LASSERT(request->rq_type == PTL_RPC_MSG_REQUEST);
LASSERT(request->rq_wait_ctx == 0);
/* If this is a re-transmit, we're required to have disengaged
* cleanly from the previous attempt */
LASSERT(!request->rq_receiving_reply);
LASSERT(!((lustre_msg_get_flags(request->rq_reqmsg) & MSG_REPLAY) &&
(imp->imp_state == LUSTRE_IMP_FULL)));
if (unlikely(obd != NULL && obd->obd_fail)) {
CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",
obd->obd_name);
/* this prevents us from waiting in ptlrpc_queue_wait */
spin_lock(&request->rq_lock);
request->rq_err = 1;
spin_unlock(&request->rq_lock);
request->rq_status = -ENODEV;
RETURN(-ENODEV);
}
connection = imp->imp_connection;
lustre_msg_set_handle(request->rq_reqmsg,
&imp->imp_remote_handle);
lustre_msg_set_type(request->rq_reqmsg, PTL_RPC_MSG_REQUEST);
lustre_msg_set_conn_cnt(request->rq_reqmsg,
imp->imp_conn_cnt);
lustre_msghdr_set_flags(request->rq_reqmsg,
imp->imp_msghdr_flags);
/* If it's the first time to resend the request for EINPROGRESS,
* we need to allocate a new XID (see after_reply()), it's different
* from the resend for reply timeout. */
if (request->rq_nr_resend != 0 &&
list_empty(&request->rq_unreplied_list)) {
__u64 min_xid = 0;
/* resend for EINPROGRESS, allocate new xid to avoid reply
* reconstruction */
spin_lock(&imp->imp_lock);
ptlrpc_assign_next_xid_nolock(request);
request->rq_mbits = request->rq_xid;
min_xid = ptlrpc_known_replied_xid(imp);
spin_unlock(&imp->imp_lock);
lustre_msg_set_last_xid(request->rq_reqmsg, min_xid);
DEBUG_REQ(D_RPCTRACE, request, "Allocating new xid for "
"resend on EINPROGRESS");
} else if (request->rq_bulk != NULL) {
ptlrpc_set_bulk_mbits(request);
lustre_msg_set_mbits(request->rq_reqmsg, request->rq_mbits);
}
if (list_empty(&request->rq_unreplied_list) ||
request->rq_xid <= imp->imp_known_replied_xid) {
DEBUG_REQ(D_ERROR, request, "xid: "LPU64", replied: "LPU64", "
"list_empty:%d\n", request->rq_xid,
imp->imp_known_replied_xid,
list_empty(&request->rq_unreplied_list));
LBUG();
}
/** For enabled AT all request should have AT_SUPPORT in the
* FULL import state when OBD_CONNECT_AT is set */
LASSERT(AT_OFF || imp->imp_state != LUSTRE_IMP_FULL ||
(imp->imp_msghdr_flags & MSGHDR_AT_SUPPORT) ||
!(imp->imp_connect_data.ocd_connect_flags &
OBD_CONNECT_AT));
if (request->rq_resend) {
lustre_msg_add_flags(request->rq_reqmsg, MSG_RESENT);
if (request->rq_resend_cb != NULL)
request->rq_resend_cb(request, &request->rq_async_args);
}
if (request->rq_memalloc)
mpflag = cfs_memory_pressure_get_and_set();
rc = sptlrpc_cli_wrap_request(request);
if (rc == -ENOMEM)
/* set rq_sent so that this request is treated
* as a delayed send in the upper layers */
request->rq_sent = cfs_time_current_sec();
if (rc)
GOTO(out, rc);
/* bulk register should be done after wrap_request() */
if (request->rq_bulk != NULL) {
rc = ptlrpc_register_bulk (request);
if (rc != 0)
GOTO(out, rc);
}
if (!noreply) {
LASSERT (request->rq_replen != 0);
if (request->rq_repbuf == NULL) {
LASSERT(request->rq_repdata == NULL);
LASSERT(request->rq_repmsg == NULL);
rc = sptlrpc_cli_alloc_repbuf(request,
request->rq_replen);
if (rc) {
/* this prevents us from looping in
* ptlrpc_queue_wait */
spin_lock(&request->rq_lock);
request->rq_err = 1;
spin_unlock(&request->rq_lock);
request->rq_status = rc;
GOTO(cleanup_bulk, rc);
}
} else {
request->rq_repdata = NULL;
request->rq_repmsg = NULL;
}
rc = LNetMEAttach(request->rq_reply_portal,/*XXX FIXME bug 249*/
connection->c_peer, request->rq_xid, 0,
LNET_UNLINK, LNET_INS_AFTER, &reply_me_h);
if (rc != 0) {
CERROR("LNetMEAttach failed: %d\n", rc);
LASSERT (rc == -ENOMEM);
GOTO(cleanup_bulk, rc = -ENOMEM);
}
}
spin_lock(&request->rq_lock);
/* We are responsible for unlinking the reply buffer */
request->rq_reply_unlinked = noreply;
request->rq_receiving_reply = !noreply;
/* Clear any flags that may be present from previous sends. */
request->rq_req_unlinked = 0;
request->rq_replied = 0;
request->rq_err = 0;
request->rq_timedout = 0;
request->rq_net_err = 0;
request->rq_resend = 0;
request->rq_restart = 0;
request->rq_reply_truncated = 0;
spin_unlock(&request->rq_lock);
if (!noreply) {
reply_md.start = request->rq_repbuf;
reply_md.length = request->rq_repbuf_len;
/* Allow multiple early replies */
reply_md.threshold = LNET_MD_THRESH_INF;
/* Manage remote for early replies */
reply_md.options = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT |
LNET_MD_MANAGE_REMOTE |
LNET_MD_TRUNCATE; /* allow to make EOVERFLOW error */;
reply_md.user_ptr = &request->rq_reply_cbid;
reply_md.eq_handle = ptlrpc_eq_h;
/* We must see the unlink callback to set rq_reply_unlinked,
* so we can't auto-unlink */
rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
&request->rq_reply_md_h);
if (rc != 0) {
CERROR("LNetMDAttach failed: %d\n", rc);
LASSERT (rc == -ENOMEM);
spin_lock(&request->rq_lock);
/* ...but the MD attach didn't succeed... */
request->rq_receiving_reply = 0;
spin_unlock(&request->rq_lock);
GOTO(cleanup_me, rc = -ENOMEM);
}
CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid "LPU64
", portal %u\n",
request->rq_repbuf_len, request->rq_xid,
request->rq_reply_portal);
}
/* add references on request for request_out_callback */
ptlrpc_request_addref(request);
if (obd != NULL && obd->obd_svc_stats != NULL)
lprocfs_counter_add(obd->obd_svc_stats, PTLRPC_REQACTIVE_CNTR,
atomic_read(&imp->imp_inflight));
OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_SEND, request->rq_timeout + 5);
do_gettimeofday(&request->rq_sent_tv);
request->rq_sent = cfs_time_current_sec();
/* We give the server rq_timeout secs to process the req, and
add the network latency for our local timeout. */
request->rq_deadline = request->rq_sent + request->rq_timeout +
ptlrpc_at_get_net_latency(request);
ptlrpc_pinger_sending_on_import(imp);
DEBUG_REQ(D_INFO, request, "send flg=%x",
lustre_msg_get_flags(request->rq_reqmsg));
rc = ptl_send_buf(&request->rq_req_md_h,
request->rq_reqbuf, request->rq_reqdata_len,
LNET_NOACK_REQ, &request->rq_req_cbid,
connection,
request->rq_request_portal,
request->rq_xid, 0);
if (likely(rc == 0))
GOTO(out, rc);
request->rq_req_unlinked = 1;
ptlrpc_req_finished(request);
if (noreply)
GOTO(out, rc);
cleanup_me:
/* MEUnlink is safe; the PUT didn't even get off the ground, and
* nobody apart from the PUT's target has the right nid+XID to
* access the reply buffer. */
rc2 = LNetMEUnlink(reply_me_h);
LASSERT (rc2 == 0);
/* UNLINKED callback called synchronously */
LASSERT(!request->rq_receiving_reply);
cleanup_bulk:
/* We do sync unlink here as there was no real transfer here so
* the chance to have long unlink to sluggish net is smaller here. */
ptlrpc_unregister_bulk(request, 0);
out:
if (request->rq_memalloc)
cfs_memory_pressure_restore(mpflag);
return rc;
}
/* partial operation for rename */
static int mdt_reint_rename_tgt(struct mdt_thread_info *info)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct mdt_object *mtgtdir;
struct mdt_object *mtgt = NULL;
struct mdt_lock_handle *lh_tgtdir;
struct mdt_lock_handle *lh_tgt = NULL;
struct lu_fid *tgt_fid = &info->mti_tmp_fid1;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, req, "rename_tgt: insert (%s->"DFID") in "DFID,
rr->rr_tgt, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
/* step 1: lookup & lock the tgt dir. */
lh_tgtdir = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh_tgtdir, LCK_PW, rr->rr_tgt,
rr->rr_tgtlen);
mtgtdir = mdt_object_find_lock(info, rr->rr_fid1, lh_tgtdir,
MDS_INODELOCK_UPDATE,
MDT_OBJ_MUST_EXIST);
if (IS_ERR(mtgtdir))
RETURN(PTR_ERR(mtgtdir));
/* step 2: find & lock the target object if exists. */
mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
lname, tgt_fid, &info->mti_spec);
if (rc != 0 && rc != -ENOENT) {
GOTO(out_unlock_tgtdir, rc);
} else if (rc == 0) {
/*
* In case of replay that name can be already inserted, check
* that and do nothing if so.
*/
if (lu_fid_eq(tgt_fid, rr->rr_fid2))
GOTO(out_unlock_tgtdir, rc);
lh_tgt = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(lh_tgt, LCK_EX);
mtgt = mdt_object_find_lock(info, tgt_fid, lh_tgt,
MDS_INODELOCK_LOOKUP,
MDT_OBJ_MUST_EXIST);
if (IS_ERR(mtgt))
GOTO(out_unlock_tgtdir, rc = PTR_ERR(mtgt));
mdt_reint_init_ma(info, ma);
if (!ma->ma_lmm || !ma->ma_cookie)
GOTO(out_unlock_tgt, rc = -EINVAL);
rc = mdo_rename_tgt(info->mti_env, mdt_object_child(mtgtdir),
mdt_object_child(mtgt), rr->rr_fid2,
lname, ma);
} else /* -ENOENT */ {
rc = mdo_name_insert(info->mti_env, mdt_object_child(mtgtdir),
lname, rr->rr_fid2, ma);
}
/* handle last link of tgt object */
if (rc == 0 && mtgt)
mdt_handle_last_unlink(info, mtgt, ma);
EXIT;
out_unlock_tgt:
if (mtgt)
mdt_object_unlock_put(info, mtgt, lh_tgt, rc);
out_unlock_tgtdir:
mdt_object_unlock_put(info, mtgtdir, lh_tgtdir, rc);
return rc;
}
/*
* VBR: we save three versions in reply:
* 0 - parent. Check that parent version is the same during replay.
* 1 - name. Version of 'name' if file exists with the same name or
* ENOENT_VERSION, it is needed because file may appear due to missed replays.
* 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
* check.
*/
static int mdt_md_create(struct mdt_thread_info *info)
{
struct mdt_device *mdt = info->mti_mdt;
struct mdt_object *parent;
struct mdt_object *child;
struct mdt_lock_handle *lh;
struct mdt_body *repbody;
struct md_attr *ma = &info->mti_attr;
struct mdt_reint_record *rr = &info->mti_rr;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
lh = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
MDS_INODELOCK_UPDATE);
if (IS_ERR(parent))
RETURN(PTR_ERR(parent));
if (mdt_object_obf(parent))
GOTO(out_put_parent, rc = -EPERM);
rc = mdt_version_get_check_save(info, parent, 0);
if (rc)
GOTO(out_put_parent, rc);
/*
* Check child name version during replay.
* During create replay a file may exist with same name.
*/
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
rc = mdt_lookup_version_check(info, parent, lname,
&info->mti_tmp_fid1, 1);
if (rc == 0)
GOTO(out_put_parent, rc = -EEXIST);
/* -ENOENT is expected here */
if (rc != -ENOENT)
GOTO(out_put_parent, rc);
/* save version of file name for replay, it must be ENOENT here */
mdt_enoent_version_save(info, 1);
child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
if (likely(!IS_ERR(child))) {
struct md_object *next = mdt_object_child(parent);
if (mdt_object_remote(child)) {
struct seq_server_site *ss;
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: Creating remote dir is only "
"permitted for administrator: rc = %d\n",
mdt2obd_dev(mdt)->obd_name, -EPERM);
GOTO(out_put_child, rc = -EPERM);
}
ss = mdt_seq_site(mdt);
if (ss->ss_node_id != 0 &&
mdt->mdt_enable_remote_dir == 0) {
CERROR("%s: remote dir is only permitted on"
" MDT0 or set_param"
" mdt.*.enable_remote_dir=1\n",
mdt2obd_dev(mdt)->obd_name);
GOTO(out_put_child, rc = -EPERM);
}
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
/* capa for cross-ref will be stored here */
ma->ma_capa = req_capsule_server_get(info->mti_pill,
&RMF_CAPA1);
LASSERT(ma->ma_capa);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_CREATE_WRITE);
/* Version of child will be updated on disk. */
info->mti_mos = child;
rc = mdt_version_get_check_save(info, child, 2);
if (rc)
GOTO(out_put_child, rc);
/* Let lower layer know current lock mode. */
info->mti_spec.sp_cr_mode =
mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
/*
* Do not perform lookup sanity check. We know that name does
* not exist.
*/
info->mti_spec.sp_cr_lookup = 0;
info->mti_spec.sp_feat = &dt_directory_features;
rc = mdo_create(info->mti_env, next, lname,
mdt_object_child(child),
&info->mti_spec, ma);
if (rc == 0)
rc = mdt_attr_get_complex(info, child, ma);
if (rc == 0) {
/* Return fid & attr to client. */
if (ma->ma_valid & MA_INODE)
mdt_pack_attr2body(info, repbody, &ma->ma_attr,
mdt_object_fid(child));
}
out_put_child:
mdt_object_put(info->mti_env, child);
} else {
rc = PTR_ERR(child);
}
mdt_create_pack_capa(info, rc, child, repbody);
out_put_parent:
mdt_object_unlock_put(info, parent, lh, rc);
RETURN(rc);
}
/*
* Client's incoming reply callback
*/
void reply_in_callback(lnet_event_t *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_request *req = cbid->cbid_arg;
DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
LASSERT(ev->type == LNET_EVENT_PUT || ev->type == LNET_EVENT_UNLINK);
LASSERT(ev->md.start == req->rq_repbuf);
LASSERT(ev->offset + ev->mlength <= req->rq_repbuf_len);
/* We've set LNET_MD_MANAGE_REMOTE for all outgoing requests
for adaptive timeouts' early reply. */
LASSERT((ev->md.options & LNET_MD_MANAGE_REMOTE) != 0);
spin_lock(&req->rq_lock);
req->rq_receiving_reply = 0;
req->rq_early = 0;
if (ev->unlinked)
req->rq_must_unlink = 0;
if (ev->status)
goto out_wake;
if (ev->type == LNET_EVENT_UNLINK) {
LASSERT(ev->unlinked);
DEBUG_REQ(D_NET, req, "unlink");
goto out_wake;
}
if (ev->mlength < ev->rlength) {
CDEBUG(D_RPCTRACE, "truncate req %p rpc %d - %d+%d\n", req,
req->rq_replen, ev->rlength, ev->offset);
req->rq_reply_truncate = 1;
req->rq_replied = 1;
req->rq_status = -EOVERFLOW;
req->rq_nob_received = ev->rlength + ev->offset;
goto out_wake;
}
if ((ev->offset == 0) &&
((lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT))) {
/* Early reply */
DEBUG_REQ(D_ADAPTTO, req,
"Early reply received: mlen=%u offset=%d replen=%d "
"replied=%d unlinked=%d", ev->mlength, ev->offset,
req->rq_replen, req->rq_replied, ev->unlinked);
req->rq_early_count++; /* number received, client side */
if (req->rq_replied) /* already got the real reply */
goto out_wake;
req->rq_early = 1;
req->rq_reply_off = ev->offset;
req->rq_nob_received = ev->mlength;
/* And we're still receiving */
req->rq_receiving_reply = 1;
} else {
/* Real reply */
req->rq_rep_swab_mask = 0;
req->rq_replied = 1;
req->rq_reply_off = ev->offset;
req->rq_nob_received = ev->mlength;
/* LNetMDUnlink can't be called under the LNET_LOCK,
so we must unlink in ptlrpc_unregister_reply */
DEBUG_REQ(D_INFO, req,
"reply in flags=%x mlen=%u offset=%d replen=%d",
lustre_msg_get_flags(req->rq_reqmsg),
ev->mlength, ev->offset, req->rq_replen);
}
req->rq_import->imp_last_reply_time = cfs_time_current_sec();
out_wake:
/* NB don't unlock till after wakeup; req can disappear under us
* since we don't have our own ref */
ptlrpc_client_wake_req(req);
spin_unlock(&req->rq_lock);
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
int no_name = 0;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
RETURN(-EPERM);
/*
* step 1: Found the parent.
*/
mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
GOTO(out, rc);
}
parent_lh = &info->mti_lh[MDT_LH_PARENT];
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
if (mdt_object_remote(mp)) {
mdt_lock_reg_init(parent_lh, LCK_EX);
rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
parent_lh->mlh_rreg_mode,
MDS_INODELOCK_UPDATE);
if (rc != ELDLM_OK)
GOTO(put_parent, rc);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc)
GOTO(put_parent, rc);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(unlock_parent, rc);
}
/* step 2: find & lock the child */
/* lookup child object along with version checking */
fid_zero(child_fid);
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0) {
/* Name might not be able to find during resend of
* remote unlink, considering following case.
* dir_A is a remote directory, the name entry of
* dir_A is on MDT0, the directory is on MDT1,
*
* 1. client sends unlink req to MDT1.
* 2. MDT1 sends name delete update to MDT0.
* 3. name entry is being deleted in MDT0 synchronously.
* 4. MDT1 is restarted.
* 5. client resends unlink req to MDT1. So it can not
* find the name entry on MDT0 anymore.
* In this case, MDT1 only needs to destory the local
* directory.
* */
if (mdt_object_remote(mp) && rc == -ENOENT &&
!fid_is_zero(rr->rr_fid2) &&
lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
no_name = 1;
*child_fid = *rr->rr_fid2;
} else {
GOTO(unlock_parent, rc);
}
}
if (fid_is_obf(child_fid) || fid_is_dot_lustre(child_fid))
GOTO(unlock_parent, rc = -EPERM);
mdt_reint_init_ma(info, ma);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
if (IS_ERR(mc))
GOTO(unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
if (mdt_object_remote(mc)) {
struct mdt_body *repbody;
if (!fid_is_zero(rr->rr_fid2)) {
CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
GOTO(put_child, rc = -ENOENT);
}
CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
if (!mdt_is_dne_client(req->rq_export))
/* Return -EIO for old client */
GOTO(put_child, rc = -EIO);
if (info->mti_spec.sp_rm_entry) {
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: unlink remote entry is only "
"permitted for administrator: rc = %d\n",
mdt_obd_name(info->mti_mdt),
-EPERM);
GOTO(put_child, rc = -EPERM);
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
NULL, lname, ma, no_name);
GOTO(put_child, rc);
}
/* Revoke the LOOKUP lock of the remote object granted by
* this MDT. Since the unlink will happen on another MDT,
* it will release the LOOKUP lock right away. Then What
* would happen if another client try to grab the LOOKUP
* lock at the same time with unlink XXX */
mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
repbody->fid1 = *mdt_object_fid(mc);
repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
GOTO(unlock_child, rc = -EREMOTE);
} else if (info->mti_spec.sp_rm_entry) {
rc = -EPERM;
CDEBUG(D_INFO, "%s: no rm_entry on local dir '%s': rc = %d\n",
mdt_obd_name(info->mti_mdt), (char *)rr->rr_name, rc);
GOTO(put_child, rc);
}
/* We used to acquire MDS_INODELOCK_FULL here but we can't do
* this now because a running HSM restore on the child (unlink
* victim) will hold the layout lock. See LU-4002. */
rc = mdt_object_lock(info, mc, child_lh,
MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
MDT_CROSS_LOCK);
if (rc != 0)
GOTO(put_child, rc);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
mutex_lock(&mc->mot_lov_mutex);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma, no_name);
mutex_unlock(&mc->mot_lov_mutex);
if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
rc = mdt_attr_get_complex(info, mc, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
unlock_child:
mdt_object_unlock(info, mc, child_lh, rc);
put_child:
mdt_object_put(info->mti_env, mc);
unlock_parent:
mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
mdt_object_put(info->mti_env, mp);
out:
return rc;
}
/**
* rsx_bulkserver - Setup a source or sink for a server.
* @rq: RPC request associated with GET.
* @type: GET_SINK receive from client or PUT_SOURCE to push to a client.
* @ptl: portal to issue bulk xfer across.
* @iov: iovec array of receive buffer.
* @n: #iovecs.
* Returns: 0 or negative errno on error.
*/
int
rsx_bulkserver(struct pscrpc_request *rq, int type, int ptl,
struct iovec *iov, int n)
{
int sum, i, rc, comms_error;
struct pscrpc_bulk_desc *desc;
struct l_wait_info lwi;
uint64_t *v8;
uint8_t *v1;
psc_assert(type == BULK_GET_SINK || type == BULK_PUT_SOURCE);
desc = pscrpc_prep_bulk_exp(rq, n, type, ptl);
if (desc == NULL) {
psclog_warnx("pscrpc_prep_bulk_exp returned a null desc");
return (-ENOMEM); // XXX errno
}
desc->bd_nob = 0;
desc->bd_iov_count = n;
memcpy(desc->bd_iov, iov, n * sizeof(*iov));
for (i = 0; i < n; i++)
desc->bd_nob += iov[i].iov_len;
/* Check for client eviction during previous I/O before proceeding. */
if (desc->bd_export->exp_failed)
rc = -ENOTCONN;
else
rc = pscrpc_start_bulk_transfer(desc);
if (rc == 0) {
lwi = LWI_TIMEOUT_INTERVAL(OBD_TIMEOUT / 2,
100, pfl_rsx_timeout, desc);
rc = pscrpc_svr_wait_event(&desc->bd_waitq,
(!pscrpc_bulk_active(desc) || desc->bd_export->exp_failed),
&lwi, NULL);
LASSERT(rc == 0 || rc == -ETIMEDOUT);
if (rc == -ETIMEDOUT) {
DEBUG_REQ(PLL_ERROR, rq, "timeout on bulk GET");
pscrpc_abort_bulk(desc);
} else if (desc->bd_export->exp_failed) {
DEBUG_REQ(PLL_ERROR, rq, "eviction on bulk GET");
rc = -ENOTCONN;
pscrpc_abort_bulk(desc);
} else if (!desc->bd_success ||
desc->bd_nob_transferred != desc->bd_nob) {
DEBUG_REQ(PLL_ERROR, rq, "%s bulk GET %d(%d)",
desc->bd_success ? "truncated" : "network error on",
desc->bd_nob_transferred, desc->bd_nob);
rc = -ETIMEDOUT;
}
} else {
DEBUG_REQ(PLL_ERROR, rq, "pscrpc I/O bulk get failed: "
"rc=%d", rc);
}
comms_error = (rc != 0);
/* count the number of bytes received, and hold for later... */
if (rc == 0) {
v1 = desc->bd_iov[0].iov_base;
v8 = desc->bd_iov[0].iov_base;
if (v1 == NULL) {
DEBUG_REQ(PLL_ERROR, rq,
"desc->bd_iov[0].iov_base is NULL");
rc = -ENXIO;
goto out;
}
DEBUG_REQ(PLL_DIAG, rq,
"got %u bytes of bulk data across %d IOVs: "
"first byte is %#x (%"PRIx64")",
desc->bd_nob, desc->bd_iov_count, *v1, *v8);
sum = 0;
for (i = 0; i < desc->bd_iov_count; i++)
sum += desc->bd_iov[i].iov_len;
if (sum != desc->bd_nob)
DEBUG_REQ(PLL_WARN, rq,
"sum (%d) does not match bd_nob (%d)",
sum, desc->bd_nob);
//rc = pscrpc_reply(rq);
}
out:
if (rc == 0)
;
else if (!comms_error) {
/* Only reply if there were no comm problems with bulk. */
rq->rq_status = rc;
pscrpc_error(rq);
} else {
#if 0
// For now let's not free the reply state..
if (rq->rq_reply_state != NULL) {
/* reply out callback would free */
pscrpc_rs_decref(rq->rq_reply_state);
rq->rq_reply_state = NULL;
rq->rq_repmsg = NULL;
}
#endif
DEBUG_REQ(PLL_WARN, rq,
"ignoring bulk I/O comm error; "
"id %s - client will retry",
libcfs_id2str(rq->rq_peer));
}
pscrpc_free_bulk(desc);
return (rc);
}
/*
* VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
* name.
*/
static int mdt_reint_link(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct mdt_object *ms;
struct mdt_object *mp;
struct mdt_lock_handle *lhs;
struct mdt_lock_handle *lhp;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
RETURN(err_serious(-ENOENT));
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
/* Invalid case so return error immediately instead of
* processing it */
if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
RETURN(-EPERM);
/* step 1: find & lock the target parent dir */
lhp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lhp, LCK_PW, rr->rr_name,
rr->rr_namelen);
mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
MDS_INODELOCK_UPDATE);
if (IS_ERR(mp))
RETURN(PTR_ERR(mp));
if (mdt_object_obf(mp))
GOTO(out_unlock_parent, rc = -EPERM);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(out_unlock_parent, rc);
/* step 2: find & lock the source */
lhs = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(lhs, LCK_EX);
ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(ms))
GOTO(out_unlock_parent, rc = PTR_ERR(ms));
if (mdt_object_remote(ms)) {
mdt_object_put(info->mti_env, ms);
CERROR("Target directory "DFID" is on another MDT\n",
PFID(rr->rr_fid1));
GOTO(out_unlock_parent, rc = -EXDEV);
}
rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, ms);
GOTO(out_unlock_parent, rc);
}
/* step 3: link it */
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_LINK_WRITE);
info->mti_mos = ms;
rc = mdt_version_get_check_save(info, ms, 1);
if (rc)
GOTO(out_unlock_child, rc);
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
/** check target version by name during replay */
rc = mdt_lookup_version_check(info, mp, lname, &info->mti_tmp_fid1, 2);
if (rc != 0 && rc != -ENOENT)
GOTO(out_unlock_child, rc);
/* save version of file name for replay, it must be ENOENT here */
if (!req_is_replay(mdt_info_req(info))) {
info->mti_ver[2] = ENOENT_VERSION;
mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
}
rc = mdo_link(info->mti_env, mdt_object_child(mp),
mdt_object_child(ms), lname, ma);
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_LINK);
EXIT;
out_unlock_child:
mdt_object_unlock_put(info, ms, lhs, rc);
out_unlock_parent:
mdt_object_unlock_put(info, mp, lhp, rc);
return rc;
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
/*
* step 1: Found the parent.
*/
mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
GOTO(out, rc);
}
if (mdt_object_obf(mp))
GOTO(put_parent, rc = -EPERM);
parent_lh = &info->mti_lh[MDT_LH_PARENT];
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
if (mdt_object_remote(mp)) {
mdt_lock_reg_init(parent_lh, LCK_EX);
rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
parent_lh->mlh_rreg_mode,
MDS_INODELOCK_UPDATE);
if (rc != ELDLM_OK)
GOTO(put_parent, rc);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc)
GOTO(put_parent, rc);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(unlock_parent, rc);
}
/* step 2: find & lock the child */
/* lookup child object along with version checking */
fid_zero(child_fid);
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0)
GOTO(unlock_parent, rc);
mdt_reint_init_ma(info, ma);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
if (IS_ERR(mc))
GOTO(unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
if (mdt_object_remote(mc)) {
struct mdt_body *repbody;
if (!fid_is_zero(rr->rr_fid2)) {
CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
GOTO(unlock_parent, rc = -ENOENT);
}
CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
if (info->mti_spec.sp_rm_entry) {
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: unlink remote entry is only "
"permitted for administrator: rc = %d\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
-EPERM);
GOTO(unlock_parent, rc = -EPERM);
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
NULL, lname, ma);
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc);
}
/* Revoke the LOOKUP lock of the remote object granted by
* this MDT. Since the unlink will happen on another MDT,
* it will release the LOOKUP lock right away. Then What
* would happen if another client try to grab the LOOKUP
* lock at the same time with unlink XXX */
mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
repbody->fid1 = *mdt_object_fid(mc);
repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
mdt_object_unlock_put(info, mc, child_lh, rc);
GOTO(unlock_parent, rc = -EREMOTE);
} else if (info->mti_spec.sp_rm_entry) {
CERROR("%s: lfs rmdir should not be used on local dir %s\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name);
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc = -EPERM);
}
rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc);
}
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma);
if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
rc = mdt_attr_get_complex(info, mc, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
mdt_object_unlock_put(info, mc, child_lh, rc);
unlock_parent:
mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
mdt_object_put(info->mti_env, mp);
out:
return rc;
}
static int mdt_reint_setattr(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct md_attr *ma = &info->mti_attr;
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
struct mdt_file_data *mfd;
struct mdt_object *mo;
struct mdt_body *repbody;
int som_au, rc, rc2;
ENTRY;
DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
(unsigned int)ma->ma_attr.la_valid);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mo))
GOTO(out, rc = PTR_ERR(mo));
if (mdt_object_obf(mo))
GOTO(out_put, rc = -EPERM);
/* start a log jounal handle if needed */
if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
if ((ma->ma_attr.la_valid & LA_SIZE) ||
(rr->rr_flags & MRF_OPEN_TRUNC)) {
/* Check write access for the O_TRUNC case */
if (mdt_write_read(mo) < 0)
GOTO(out_put, rc = -ETXTBSY);
}
} else if (info->mti_ioepoch &&
(info->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
/* Truncate case. IOEpoch is opened. */
rc = mdt_write_get(mo);
if (rc)
GOTO(out_put, rc);
mfd = mdt_mfd_new();
if (mfd == NULL) {
mdt_write_put(mo);
GOTO(out_put, rc = -ENOMEM);
}
mdt_ioepoch_open(info, mo, 0);
repbody->ioepoch = mo->mot_ioepoch;
mdt_object_get(info->mti_env, mo);
mdt_mfd_set_mode(mfd, MDS_FMODE_TRUNC);
mfd->mfd_object = mo;
mfd->mfd_xid = req->rq_xid;
spin_lock(&med->med_open_lock);
cfs_list_add(&mfd->mfd_list, &med->med_open_head);
spin_unlock(&med->med_open_lock);
repbody->handle.cookie = mfd->mfd_handle.h_cookie;
}
som_au = info->mti_ioepoch && info->mti_ioepoch->flags & MF_SOM_CHANGE;
if (som_au) {
/* SOM Attribute update case. Find the proper mfd and update
* SOM attributes on the proper object. */
LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
LASSERT(info->mti_ioepoch);
spin_lock(&med->med_open_lock);
mfd = mdt_handle2mfd(info, &info->mti_ioepoch->handle);
if (mfd == NULL) {
spin_unlock(&med->med_open_lock);
CDEBUG(D_INODE, "no handle for file close: "
"fid = "DFID": cookie = "LPX64"\n",
PFID(info->mti_rr.rr_fid1),
info->mti_ioepoch->handle.cookie);
GOTO(out_put, rc = -ESTALE);
}
LASSERT(mfd->mfd_mode == MDS_FMODE_SOM);
LASSERT(!(info->mti_ioepoch->flags & MF_EPOCH_CLOSE));
class_handle_unhash(&mfd->mfd_handle);
cfs_list_del_init(&mfd->mfd_list);
spin_unlock(&med->med_open_lock);
mdt_mfd_close(info, mfd);
} else if ((ma->ma_valid & MA_INODE) && ma->ma_attr.la_valid) {
LASSERT((ma->ma_valid & MA_LOV) == 0);
rc = mdt_attr_set(info, mo, ma, rr->rr_flags);
if (rc)
GOTO(out_put, rc);
} else if ((ma->ma_valid & MA_LOV) && (ma->ma_valid & MA_INODE)) {
struct lu_buf *buf = &info->mti_buf;
LASSERT(ma->ma_attr.la_valid == 0);
buf->lb_buf = ma->ma_lmm;
buf->lb_len = ma->ma_lmm_size;
rc = mo_xattr_set(info->mti_env, mdt_object_child(mo),
buf, XATTR_NAME_LOV, 0);
if (rc)
GOTO(out_put, rc);
} else
LBUG();
/* If file data is modified, add the dirty flag */
if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
rc = mdt_add_dirty_flag(info, mo, ma);
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
rc = mdt_attr_get_complex(info, mo, ma);
if (rc != 0)
GOTO(out_put, rc);
mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
if (info->mti_mdt->mdt_opts.mo_oss_capa &&
exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
S_ISREG(lu_object_attr(&mo->mot_obj.mo_lu)) &&
(ma->ma_attr.la_valid & LA_SIZE) && !som_au) {
struct lustre_capa *capa;
capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
LASSERT(capa);
capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
if (rc)
GOTO(out_put, rc);
repbody->valid |= OBD_MD_FLOSSCAPA;
}
EXIT;
out_put:
mdt_object_put(info->mti_env, mo);
out:
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_SETATTR);
mdt_client_compatibility(info);
rc2 = mdt_fix_reply(info);
if (rc == 0)
rc = rc2;
return rc;
}
/* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
* should take the normal semaphore and go to the normal portal.
*
* If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
* magic open-path setattr that should take the setattr semaphore and
* go to the setattr portal. */
int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
void *ea, int ealen, void *ea2, int ea2len,
struct ptlrpc_request **request, struct md_open_data **mod)
{
CFS_LIST_HEAD(cancels);
struct ptlrpc_request *req;
struct mdc_rpc_lock *rpc_lock;
struct obd_device *obd = exp->exp_obd;
int count = 0, rc;
__u64 bits;
ENTRY;
LASSERT(op_data != NULL);
bits = MDS_INODELOCK_UPDATE;
if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
bits |= MDS_INODELOCK_LOOKUP;
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX, bits);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_SETATTR);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
0);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
ea2len);
rc = mdc_prep_elc_req(exp, req, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
req->rq_request_portal = MDS_SETATTR_PORTAL;
ptlrpc_at_set_req_timeout(req);
rpc_lock = obd->u.cli.cl_setattr_lock;
} else {
rpc_lock = obd->u.cli.cl_rpc_lock;
}
if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
", ctime "CFS_TIME_T"\n",
LTIME_S(op_data->op_attr.ia_mtime),
LTIME_S(op_data->op_attr.ia_ctime));
mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
ptlrpc_request_set_replen(req);
if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
req->rq_import->imp_replayable)
{
LASSERT(*mod == NULL);
*mod = obd_mod_alloc();
if (*mod == NULL) {
DEBUG_REQ(D_ERROR, req, "Can't allocate "
"md_open_data");
} else {
req->rq_replay = 1;
req->rq_cb_data = *mod;
(*mod)->mod_open_req = req;
req->rq_commit_cb = mdc_commit_open;
/**
* Take an extra reference on \var mod, it protects \var
* mod from being freed on eviction (commit callback is
* called despite rq_replay flag).
* Will be put on mdc_done_writing().
*/
obd_mod_get(*mod);
}
}
rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
/* Save the obtained info in the original RPC for the replay case. */
if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
struct mdt_ioepoch *epoch;
struct mdt_body *body;
epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
LASSERT(epoch != NULL);
LASSERT(body != NULL);
epoch->handle = body->handle;
epoch->ioepoch = body->ioepoch;
req->rq_replay_cb = mdc_replay_open;
/** bug 3633, open may be committed and estale answer is not error */
} else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
rc = 0;
} else if (rc == -ERESTARTSYS) {
rc = 0;
}
*request = req;
if (rc && req->rq_commit_cb) {
/* Put an extra reference on \var mod on error case. */
obd_mod_put(*mod);
req->rq_commit_cb(req);
}
RETURN(rc);
}
/*
* VBR: rename versions in reply: 0 - src parent; 1 - tgt parent;
* 2 - src child; 3 - tgt child.
* Update on disk version of src child.
*/
static int mdt_reint_rename(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct md_attr *ma = &info->mti_attr;
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_object *msrcdir;
struct mdt_object *mtgtdir;
struct mdt_object *mold;
struct mdt_object *mnew = NULL;
struct mdt_lock_handle *lh_srcdirp;
struct mdt_lock_handle *lh_tgtdirp;
struct mdt_lock_handle *lh_oldp;
struct mdt_lock_handle *lh_newp;
struct lu_fid *old_fid = &info->mti_tmp_fid1;
struct lu_fid *new_fid = &info->mti_tmp_fid2;
struct lustre_handle rename_lh = { 0 };
struct lu_name slname = { 0 };
struct lu_name *lname;
int rc;
ENTRY;
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
PFID(rr->rr_fid1), rr->rr_name,
PFID(rr->rr_fid2), rr->rr_tgt);
rc = mdt_rename_lock(info, &rename_lh);
if (rc) {
CERROR("Can't lock FS for rename, rc %d\n", rc);
RETURN(rc);
}
lh_newp = &info->mti_lh[MDT_LH_NEW];
/* step 1: lock the source dir. */
lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
rr->rr_namelen);
msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
MDS_INODELOCK_UPDATE);
if (IS_ERR(msrcdir))
GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
if (mdt_object_obf(msrcdir))
GOTO(out_unlock_source, rc = -EPERM);
rc = mdt_version_get_check_save(info, msrcdir, 0);
if (rc)
GOTO(out_unlock_source, rc);
/* step 2: find & lock the target dir. */
lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
rr->rr_tgtlen);
if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
mdt_object_get(info->mti_env, msrcdir);
mtgtdir = msrcdir;
if (lh_tgtdirp->mlh_pdo_hash != lh_srcdirp->mlh_pdo_hash) {
rc = mdt_pdir_hash_lock(info, lh_tgtdirp, mtgtdir,
MDS_INODELOCK_UPDATE);
if (rc)
GOTO(out_unlock_source, rc);
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
}
} else {
mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
rr->rr_fid2);
if (IS_ERR(mtgtdir))
GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
if (mdt_object_obf(mtgtdir))
GOTO(out_put_target, rc = -EPERM);
/* check early, the real version will be saved after locking */
rc = mdt_version_get_check(info, mtgtdir, 1);
if (rc)
GOTO(out_put_target, rc);
if (unlikely(mdt_object_remote(mtgtdir))) {
CDEBUG(D_INFO, "Source dir "DFID" target dir "DFID
"on different MDTs\n", PFID(rr->rr_fid1),
PFID(rr->rr_fid2));
GOTO(out_put_target, rc = -EXDEV);
} else {
if (likely(mdt_object_exists(mtgtdir))) {
/* we lock the target dir if it is local */
rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc != 0)
GOTO(out_put_target, rc);
/* get and save correct version after locking */
mdt_version_get_save(info, mtgtdir, 1);
} else {
GOTO(out_put_target, rc = -ESTALE);
}
}
}
/* step 3: find & lock the old object. */
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
mdt_name_copy(&slname, lname);
fid_zero(old_fid);
rc = mdt_lookup_version_check(info, msrcdir, &slname, old_fid, 2);
if (rc != 0)
GOTO(out_unlock_target, rc);
if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
GOTO(out_unlock_target, rc = -EINVAL);
mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
if (IS_ERR(mold))
GOTO(out_unlock_target, rc = PTR_ERR(mold));
if (mdt_object_remote(mold)) {
mdt_object_put(info->mti_env, mold);
CDEBUG(D_INFO, "Source child "DFID" is on another MDT\n",
PFID(old_fid));
GOTO(out_unlock_target, rc = -EXDEV);
}
if (mdt_object_obf(mold)) {
mdt_object_put(info->mti_env, mold);
GOTO(out_unlock_target, rc = -EPERM);
}
lh_oldp = &info->mti_lh[MDT_LH_OLD];
mdt_lock_reg_init(lh_oldp, LCK_EX);
rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mold);
GOTO(out_unlock_target, rc);
}
info->mti_mos = mold;
/* save version after locking */
mdt_version_get_save(info, mold, 2);
mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
/* step 4: find & lock the new object. */
/* new target object may not exist now */
lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
/* lookup with version checking */
fid_zero(new_fid);
rc = mdt_lookup_version_check(info, mtgtdir, lname, new_fid, 3);
if (rc == 0) {
/* the new_fid should have been filled at this moment */
if (lu_fid_eq(old_fid, new_fid))
GOTO(out_unlock_old, rc);
if (lu_fid_eq(new_fid, rr->rr_fid1) ||
lu_fid_eq(new_fid, rr->rr_fid2))
GOTO(out_unlock_old, rc = -EINVAL);
mdt_lock_reg_init(lh_newp, LCK_EX);
mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
if (IS_ERR(mnew))
GOTO(out_unlock_old, rc = PTR_ERR(mnew));
if (mdt_object_obf(mnew)) {
mdt_object_put(info->mti_env, mnew);
GOTO(out_unlock_old, rc = -EPERM);
}
if (mdt_object_remote(mnew)) {
mdt_object_put(info->mti_env, mnew);
CDEBUG(D_INFO, "src child "DFID" is on another MDT\n",
PFID(new_fid));
GOTO(out_unlock_old, rc = -EXDEV);
}
rc = mdt_object_lock(info, mnew, lh_newp,
MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mnew);
GOTO(out_unlock_old, rc);
}
/* get and save version after locking */
mdt_version_get_save(info, mnew, 3);
mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
} else if (rc != -EREMOTE && rc != -ENOENT) {
GOTO(out_unlock_old, rc);
} else {
mdt_enoent_version_save(info, 3);
}
/* step 5: rename it */
mdt_reint_init_ma(info, ma);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_RENAME_WRITE);
/* Check if @dst is subdir of @src. */
rc = mdt_rename_sanity(info, old_fid);
if (rc)
GOTO(out_unlock_new, rc);
rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
mdt_object_child(mtgtdir), old_fid, &slname,
(mnew ? mdt_object_child(mnew) : NULL),
lname, ma);
/* handle last link of tgt object */
if (rc == 0) {
mdt_counter_incr(req, LPROC_MDT_RENAME);
if (mnew)
mdt_handle_last_unlink(info, mnew, ma);
mdt_rename_counter_tally(info, info->mti_mdt, req,
msrcdir, mtgtdir);
}
EXIT;
out_unlock_new:
if (mnew)
mdt_object_unlock_put(info, mnew, lh_newp, rc);
out_unlock_old:
mdt_object_unlock_put(info, mold, lh_oldp, rc);
out_unlock_target:
mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
out_put_target:
mdt_object_put(info->mti_env, mtgtdir);
out_unlock_source:
mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
out_rename_lock:
if (lustre_handle_is_used(&rename_lh))
mdt_rename_unlock(&rename_lh);
return rc;
}
int llog_origin_handle_cancel(struct ptlrpc_request *req)
{
struct obd_device *obd = req->rq_export->exp_obd;
int num_cookies, rc = 0, err, i, failed = 0;
struct obd_device *disk_obd;
struct llog_cookie *logcookies;
struct llog_ctxt *ctxt = NULL;
struct lvfs_run_ctxt saved;
struct llog_handle *cathandle;
struct inode *inode;
void *handle;
ENTRY;
logcookies = req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES);
num_cookies = req_capsule_get_size(&req->rq_pill, &RMF_LOGCOOKIES,
RCL_CLIENT) / sizeof(*logcookies);
if (logcookies == NULL || num_cookies == 0) {
DEBUG_REQ(D_HA, req, "No llog cookies sent");
RETURN(-EFAULT);
}
ctxt = llog_get_context(obd, logcookies->lgc_subsys);
if (ctxt == NULL)
RETURN(-ENODEV);
disk_obd = ctxt->loc_exp->exp_obd;
push_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
for (i = 0; i < num_cookies; i++, logcookies++) {
cathandle = ctxt->loc_handle;
LASSERT(cathandle != NULL);
inode = cathandle->lgh_file->f_dentry->d_inode;
handle = fsfilt_start_log(disk_obd, inode,
FSFILT_OP_CANCEL_UNLINK, NULL, 1);
if (IS_ERR(handle)) {
CERROR("fsfilt_start_log() failed: %ld\n",
PTR_ERR(handle));
GOTO(pop_ctxt, rc = PTR_ERR(handle));
}
rc = llog_cat_cancel_records(cathandle, 1, logcookies);
/*
* Do not raise -ENOENT errors for resent rpcs. This rec already
* might be killed.
*/
if (rc == -ENOENT &&
(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)) {
/*
* Do not change this message, reply-single.sh test_59b
* expects to find this in log.
*/
CDEBUG(D_RPCTRACE, "RESENT cancel req %p - ignored\n",
req);
rc = 0;
} else if (rc == 0) {
CDEBUG(D_RPCTRACE, "Canceled %d llog-records\n",
num_cookies);
}
err = fsfilt_commit(disk_obd, inode, handle, 0);
if (err) {
CERROR("Error committing transaction: %d\n", err);
if (!rc)
rc = err;
failed++;
GOTO(pop_ctxt, rc);
} else if (rc)
failed++;
}
GOTO(pop_ctxt, rc);
pop_ctxt:
pop_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
if (rc)
CERROR("Cancel %d of %d llog-records failed: %d\n",
failed, num_cookies, rc);
llog_ctxt_put(ctxt);
return rc;
}
/* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
int mgs_handle(struct ptlrpc_request *req)
{
int fail = OBD_FAIL_MGS_ALL_REPLY_NET;
int opc, rc = 0;
ENTRY;
req_capsule_init(&req->rq_pill, req, RCL_SERVER);
CFS_FAIL_TIMEOUT_MS(OBD_FAIL_MGS_PAUSE_REQ, cfs_fail_val);
if (CFS_FAIL_CHECK(OBD_FAIL_MGS_ALL_REQUEST_NET))
RETURN(0);
LASSERT(current->journal_info == NULL);
opc = lustre_msg_get_opc(req->rq_reqmsg);
if (opc == SEC_CTX_INIT ||
opc == SEC_CTX_INIT_CONT ||
opc == SEC_CTX_FINI)
GOTO(out, rc = 0);
if (opc != MGS_CONNECT) {
if (!class_connected_export(req->rq_export)) {
DEBUG_REQ(D_MGS, req, "operation on unconnected MGS\n");
req->rq_status = -ENOTCONN;
GOTO(out, rc = -ENOTCONN);
}
}
switch (opc) {
case MGS_CONNECT:
DEBUG_REQ(D_MGS, req, "connect");
/* MGS and MDS have same request format for connect */
req_capsule_set(&req->rq_pill, &RQF_MDS_CONNECT);
rc = target_handle_connect(req);
if (rc == 0)
rc = mgs_connect_check_sptlrpc(req);
if (!rc && (lustre_msg_get_conn_cnt(req->rq_reqmsg) > 1))
/* Make clients trying to reconnect after a MGS restart
happy; also requires obd_replayable */
lustre_msg_add_op_flags(req->rq_repmsg,
MSG_CONNECT_RECONNECT);
break;
case MGS_DISCONNECT:
DEBUG_REQ(D_MGS, req, "disconnect");
/* MGS and MDS have same request format for disconnect */
req_capsule_set(&req->rq_pill, &RQF_MDS_DISCONNECT);
rc = target_handle_disconnect(req);
req->rq_status = rc; /* superfluous? */
break;
case MGS_EXCEPTION:
DEBUG_REQ(D_MGS, req, "exception");
rc = mgs_handle_exception(req);
break;
case MGS_TARGET_REG:
DEBUG_REQ(D_MGS, req, "target add");
req_capsule_set(&req->rq_pill, &RQF_MGS_TARGET_REG);
rc = mgs_handle_target_reg(req);
break;
case MGS_TARGET_DEL:
DEBUG_REQ(D_MGS, req, "target del");
rc = mgs_handle_target_del(req);
break;
case MGS_SET_INFO:
DEBUG_REQ(D_MGS, req, "set_info");
req_capsule_set(&req->rq_pill, &RQF_MGS_SET_INFO);
rc = mgs_set_info_rpc(req);
break;
case MGS_CONFIG_READ:
DEBUG_REQ(D_MGS, req, "read config");
req_capsule_set(&req->rq_pill, &RQF_MGS_CONFIG_READ);
rc = mgs_config_read(req);
break;
case LDLM_ENQUEUE:
DEBUG_REQ(D_MGS, req, "enqueue");
req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
ldlm_server_blocking_ast, NULL);
break;
case LDLM_BL_CALLBACK:
case LDLM_CP_CALLBACK:
DEBUG_REQ(D_MGS, req, "callback");
CERROR("callbacks should not happen on MGS\n");
LBUG();
break;
case OBD_PING:
DEBUG_REQ(D_INFO, req, "ping");
req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
rc = target_handle_ping(req);
break;
case OBD_LOG_CANCEL:
DEBUG_REQ(D_MGS, req, "log cancel");
rc = -ENOTSUPP; /* la la la */
break;
case LLOG_ORIGIN_HANDLE_CREATE:
DEBUG_REQ(D_MGS, req, "llog_init");
req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
rc = llog_origin_handle_create(req);
if (rc == 0)
(void)mgs_handle_fslog_hack(req);
break;
case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
DEBUG_REQ(D_MGS, req, "llog next block");
req_capsule_set(&req->rq_pill,
&RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
rc = llog_origin_handle_next_block(req);
break;
case LLOG_ORIGIN_HANDLE_READ_HEADER:
DEBUG_REQ(D_MGS, req, "llog read header");
req_capsule_set(&req->rq_pill,
&RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
rc = llog_origin_handle_read_header(req);
break;
case LLOG_ORIGIN_HANDLE_CLOSE:
DEBUG_REQ(D_MGS, req, "llog close");
rc = llog_origin_handle_close(req);
break;
case LLOG_CATINFO:
DEBUG_REQ(D_MGS, req, "llog catinfo");
req_capsule_set(&req->rq_pill, &RQF_LLOG_CATINFO);
rc = llog_catinfo(req);
break;
default:
req->rq_status = -ENOTSUPP;
rc = ptlrpc_error(req);
RETURN(rc);
}
LASSERT(current->journal_info == NULL);
if (rc)
CERROR("MGS handle cmd=%d rc=%d\n", opc, rc);
out:
target_send_reply(req, rc, fail);
RETURN(0);
}
/**
* Send request \a request.
* if \a noreply is set, don't expect any reply back and don't set up
* reply buffers.
* Returns 0 on success or error code.
*/
int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
{
int rc;
int rc2;
int mpflag = 0;
struct ptlrpc_connection *connection;
lnet_handle_me_t reply_me_h;
lnet_md_t reply_md;
struct obd_device *obd = request->rq_import->imp_obd;
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_RPC))
return 0;
LASSERT(request->rq_type == PTL_RPC_MSG_REQUEST);
LASSERT(request->rq_wait_ctx == 0);
/* If this is a re-transmit, we're required to have disengaged
* cleanly from the previous attempt */
LASSERT(!request->rq_receiving_reply);
LASSERT(!((lustre_msg_get_flags(request->rq_reqmsg) & MSG_REPLAY) &&
(request->rq_import->imp_state == LUSTRE_IMP_FULL)));
if (unlikely(obd != NULL && obd->obd_fail)) {
CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",
obd->obd_name);
/* this prevents us from waiting in ptlrpc_queue_wait */
spin_lock(&request->rq_lock);
request->rq_err = 1;
spin_unlock(&request->rq_lock);
request->rq_status = -ENODEV;
return -ENODEV;
}
connection = request->rq_import->imp_connection;
lustre_msg_set_handle(request->rq_reqmsg,
&request->rq_import->imp_remote_handle);
lustre_msg_set_type(request->rq_reqmsg, PTL_RPC_MSG_REQUEST);
lustre_msg_set_conn_cnt(request->rq_reqmsg,
request->rq_import->imp_conn_cnt);
lustre_msghdr_set_flags(request->rq_reqmsg,
request->rq_import->imp_msghdr_flags);
if (request->rq_resend)
lustre_msg_add_flags(request->rq_reqmsg, MSG_RESENT);
if (request->rq_memalloc)
mpflag = cfs_memory_pressure_get_and_set();
rc = sptlrpc_cli_wrap_request(request);
if (rc)
goto out;
/* bulk register should be done after wrap_request() */
if (request->rq_bulk != NULL) {
rc = ptlrpc_register_bulk(request);
if (rc != 0)
goto out;
}
if (!noreply) {
LASSERT(request->rq_replen != 0);
if (request->rq_repbuf == NULL) {
LASSERT(request->rq_repdata == NULL);
LASSERT(request->rq_repmsg == NULL);
rc = sptlrpc_cli_alloc_repbuf(request,
request->rq_replen);
if (rc) {
/* this prevents us from looping in
* ptlrpc_queue_wait */
spin_lock(&request->rq_lock);
request->rq_err = 1;
spin_unlock(&request->rq_lock);
request->rq_status = rc;
goto cleanup_bulk;
}
} else {
request->rq_repdata = NULL;
request->rq_repmsg = NULL;
}
rc = LNetMEAttach(request->rq_reply_portal,/*XXX FIXME bug 249*/
connection->c_peer, request->rq_xid, 0,
LNET_UNLINK, LNET_INS_AFTER, &reply_me_h);
if (rc != 0) {
CERROR("LNetMEAttach failed: %d\n", rc);
LASSERT(rc == -ENOMEM);
rc = -ENOMEM;
goto cleanup_bulk;
}
}
spin_lock(&request->rq_lock);
/* If the MD attach succeeds, there _will_ be a reply_in callback */
request->rq_receiving_reply = !noreply;
request->rq_req_unlink = 1;
/* We are responsible for unlinking the reply buffer */
request->rq_reply_unlink = !noreply;
/* Clear any flags that may be present from previous sends. */
request->rq_replied = 0;
request->rq_err = 0;
request->rq_timedout = 0;
request->rq_net_err = 0;
request->rq_resend = 0;
request->rq_restart = 0;
request->rq_reply_truncate = 0;
spin_unlock(&request->rq_lock);
if (!noreply) {
reply_md.start = request->rq_repbuf;
reply_md.length = request->rq_repbuf_len;
/* Allow multiple early replies */
reply_md.threshold = LNET_MD_THRESH_INF;
/* Manage remote for early replies */
reply_md.options = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT |
LNET_MD_MANAGE_REMOTE |
LNET_MD_TRUNCATE; /* allow to make EOVERFLOW error */
reply_md.user_ptr = &request->rq_reply_cbid;
reply_md.eq_handle = ptlrpc_eq_h;
/* We must see the unlink callback to unset rq_reply_unlink,
so we can't auto-unlink */
rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
&request->rq_reply_md_h);
if (rc != 0) {
CERROR("LNetMDAttach failed: %d\n", rc);
LASSERT(rc == -ENOMEM);
spin_lock(&request->rq_lock);
/* ...but the MD attach didn't succeed... */
request->rq_receiving_reply = 0;
spin_unlock(&request->rq_lock);
rc = -ENOMEM;
goto cleanup_me;
}
CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid %llu, portal %u\n",
request->rq_repbuf_len, request->rq_xid,
request->rq_reply_portal);
}
/* add references on request for request_out_callback */
ptlrpc_request_addref(request);
if (obd != NULL && obd->obd_svc_stats != NULL)
lprocfs_counter_add(obd->obd_svc_stats, PTLRPC_REQACTIVE_CNTR,
atomic_read(&request->rq_import->imp_inflight));
OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_SEND, request->rq_timeout + 5);
ktime_get_real_ts64(&request->rq_arrival_time);
request->rq_sent = ktime_get_real_seconds();
/* We give the server rq_timeout secs to process the req, and
add the network latency for our local timeout. */
request->rq_deadline = request->rq_sent + request->rq_timeout +
ptlrpc_at_get_net_latency(request);
ptlrpc_pinger_sending_on_import(request->rq_import);
DEBUG_REQ(D_INFO, request, "send flg=%x",
lustre_msg_get_flags(request->rq_reqmsg));
rc = ptl_send_buf(&request->rq_req_md_h,
request->rq_reqbuf, request->rq_reqdata_len,
LNET_NOACK_REQ, &request->rq_req_cbid,
connection,
request->rq_request_portal,
request->rq_xid, 0);
if (rc == 0)
goto out;
ptlrpc_req_finished(request);
if (noreply)
goto out;
cleanup_me:
/* MEUnlink is safe; the PUT didn't even get off the ground, and
* nobody apart from the PUT's target has the right nid+XID to
* access the reply buffer. */
rc2 = LNetMEUnlink(reply_me_h);
LASSERT(rc2 == 0);
/* UNLINKED callback called synchronously */
LASSERT(!request->rq_receiving_reply);
cleanup_bulk:
/* We do sync unlink here as there was no real transfer here so
* the chance to have long unlink to sluggish net is smaller here. */
ptlrpc_unregister_bulk(request, 0);
out:
if (request->rq_memalloc)
cfs_memory_pressure_restore(mpflag);
return rc;
}
/*
* VBR: we save three versions in reply:
* 0 - parent. Check that parent version is the same during replay.
* 1 - name. Version of 'name' if file exists with the same name or
* ENOENT_VERSION, it is needed because file may appear due to missed replays.
* 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
* check.
*/
static int mdt_md_create(struct mdt_thread_info *info)
{
struct mdt_device *mdt = info->mti_mdt;
struct mdt_object *parent;
struct mdt_object *child;
struct mdt_lock_handle *lh;
struct mdt_body *repbody;
struct md_attr *ma = &info->mti_attr;
struct mdt_reint_record *rr = &info->mti_rr;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
lh = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
MDS_INODELOCK_UPDATE, MDT_OBJ_MUST_EXIST);
if (IS_ERR(parent))
RETURN(PTR_ERR(parent));
rc = mdt_version_get_check_save(info, parent, 0);
if (rc)
GOTO(out_put_parent, rc);
/*
* Check child name version during replay.
* During create replay a file may exist with same name.
*/
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
rc = mdt_lookup_version_check(info, parent, lname,
&info->mti_tmp_fid1, 1);
/* -ENOENT is expected here */
if (rc != 0 && rc != -ENOENT)
GOTO(out_put_parent, rc);
/* save version of file name for replay, it must be ENOENT here */
mdt_enoent_version_save(info, 1);
child = mdt_object_find(info->mti_env, mdt, rr->rr_fid2,
MDT_OBJ_MAY_NOT_EXIST);
if (likely(!IS_ERR(child))) {
struct md_object *next = mdt_object_child(parent);
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
/* capa for cross-ref will be stored here */
ma->ma_capa = req_capsule_server_get(info->mti_pill,
&RMF_CAPA1);
LASSERT(ma->ma_capa);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_CREATE_WRITE);
/* Version of child will be updated on disk. */
info->mti_mos = child;
rc = mdt_version_get_check_save(info, child, 2);
if (rc)
GOTO(out_put_child, rc);
/* Let lower layer know current lock mode. */
info->mti_spec.sp_cr_mode =
mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
/*
* Do perform lookup sanity check. We do not know if name exists
* or not.
*/
info->mti_spec.sp_cr_lookup = 1;
info->mti_spec.sp_feat = &dt_directory_features;
rc = mdo_create(info->mti_env, next, lname,
mdt_object_child(child),
&info->mti_spec, ma);
if (rc == 0) {
/* Return fid & attr to client. */
if (ma->ma_valid & MA_INODE)
mdt_pack_attr2body(info, repbody, &ma->ma_attr,
mdt_object_fid(child));
}
out_put_child:
mdt_object_put(info->mti_env, child);
} else {
rc = PTR_ERR(child);
}
mdt_create_pack_capa(info, rc, child, repbody);
out_put_parent:
mdt_object_unlock_put(info, parent, lh, rc);
RETURN(rc);
}
