int mds_quota_ctl(struct obd_device *obd, struct obd_export *unused,
struct obd_quotactl *oqctl)
{
struct obd_device_target *obt = &obd->u.obt;
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
int rc = 0;
ENTRY;
cfs_gettimeofday(&work_start);
switch (oqctl->qc_cmd) {
case Q_QUOTAON:
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
rc = mds_quota_on(obd, oqctl);
break;
case Q_QUOTAOFF:
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
rc = mds_quota_off(obd, oqctl);
break;
case Q_SETINFO:
rc = mds_set_dqinfo(obd, oqctl);
break;
case Q_GETINFO:
rc = mds_get_dqinfo(obd, oqctl);
break;
case Q_SETQUOTA:
rc = mds_set_dqblk(obd, oqctl);
break;
case Q_GETQUOTA:
rc = mds_get_dqblk(obd, oqctl);
break;
case Q_GETOINFO:
case Q_GETOQUOTA:
rc = mds_get_obd_quota(obd, oqctl);
break;
case LUSTRE_Q_INVALIDATE:
rc = mds_quota_invalidate(obd, oqctl);
break;
case LUSTRE_Q_FINVALIDATE:
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
rc = mds_quota_finvalidate(obd, oqctl);
break;
default:
CERROR("%s: unsupported mds_quotactl command: %d\n",
obd->obd_name, oqctl->qc_cmd);
RETURN(-EFAULT);
}
if (rc)
CDEBUG(D_INFO, "mds_quotactl admin quota command %d, id %u, "
"type %d, failed: rc = %d\n",
oqctl->qc_cmd, oqctl->qc_id, oqctl->qc_type, rc);
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats, LQUOTA_QUOTA_CTL, timediff);
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_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;
}
static int osd_bufs_get_write(const struct lu_env *env, struct osd_object *obj,
loff_t off, ssize_t len, struct niobuf_local *lnb)
{
struct osd_device *osd = osd_obj2dev(obj);
int plen, off_in_block, sz_in_block;
int rc, i = 0, npages = 0;
arc_buf_t *abuf;
uint32_t bs;
uint64_t dummy;
ENTRY;
dmu_object_size_from_db(obj->oo_db, &bs, &dummy);
/*
* currently only full blocks are subject to zerocopy approach:
* so that we're sure nobody is trying to update the same block
*/
while (len > 0) {
LASSERT(npages < PTLRPC_MAX_BRW_PAGES);
off_in_block = off & (bs - 1);
sz_in_block = min_t(int, bs - off_in_block, len);
if (sz_in_block == bs) {
/* full block, try to use zerocopy */
abuf = dmu_request_arcbuf(obj->oo_db, bs);
if (unlikely(abuf == NULL))
GOTO(out_err, rc = -ENOMEM);
atomic_inc(&osd->od_zerocopy_loan);
/* go over pages arcbuf contains, put them as
* local niobufs for ptlrpc's bulks */
while (sz_in_block > 0) {
plen = min_t(int, sz_in_block, PAGE_CACHE_SIZE);
lnb[i].lnb_file_offset = off;
lnb[i].lnb_page_offset = 0;
lnb[i].lnb_len = plen;
lnb[i].lnb_rc = 0;
if (sz_in_block == bs)
lnb[i].lnb_data = abuf;
else
lnb[i].lnb_data = NULL;
/* this one is not supposed to fail */
lnb[i].lnb_page = kmem_to_page(abuf->b_data +
off_in_block);
LASSERT(lnb[i].lnb_page);
lprocfs_counter_add(osd->od_stats,
LPROC_OSD_ZEROCOPY_IO, 1);
sz_in_block -= plen;
len -= plen;
off += plen;
off_in_block += plen;
i++;
npages++;
}
} else {
if (off_in_block == 0 && len < bs &&
int filter_quota_ctl(struct obd_device *unused, struct obd_export *exp,
struct obd_quotactl *oqctl)
{
struct obd_device *obd = exp->exp_obd;
struct obd_device_target *obt = &obd->u.obt;
struct lvfs_run_ctxt saved;
struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
struct lustre_qunit_size *lqs;
void *handle = NULL;
struct timeval work_start;
struct timeval work_end;
long timediff;
int rc = 0;
ENTRY;
cfs_gettimeofday(&work_start);
switch (oqctl->qc_cmd) {
case Q_QUOTAON:
oqctl->qc_id = obt->obt_qfmt;
rc = generic_quota_on(obd, oqctl, 0);
break;
case Q_FINVALIDATE:
case Q_QUOTAOFF:
cfs_down(&obt->obt_quotachecking);
if (oqctl->qc_cmd == Q_FINVALIDATE &&
(obt->obt_qctxt.lqc_flags & UGQUOTA2LQC(oqctl->qc_type))) {
CWARN("quota[%u] is on yet\n", oqctl->qc_type);
cfs_up(&obt->obt_quotachecking);
rc = -EBUSY;
break;
}
oqctl->qc_id = obt->obt_qfmt; /* override qfmt version */
case Q_GETOINFO:
case Q_GETOQUOTA:
case Q_GETQUOTA:
/* In recovery scenario, this pending dqacq/dqrel might have
* been processed by master successfully before it's dquot
* on master enter recovery mode. We must wait for this
* dqacq/dqrel done then return the correct limits to master */
if (oqctl->qc_stat == QUOTA_RECOVERING)
handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
if (oqctl->qc_stat == QUOTA_RECOVERING)
quota_unbarrier(handle);
if (oqctl->qc_cmd == Q_QUOTAOFF ||
oqctl->qc_cmd == Q_FINVALIDATE) {
if (oqctl->qc_cmd == Q_QUOTAOFF) {
if (!rc)
obt->obt_qctxt.lqc_flags &=
~UGQUOTA2LQC(oqctl->qc_type);
else if (quota_is_off(qctxt, oqctl))
rc = -EALREADY;
CDEBUG(D_QUOTA, "%s: quotaoff type:flags:rc "
"%u:%lu:%d\n", obd->obd_name,
oqctl->qc_type, qctxt->lqc_flags, rc);
}
cfs_up(&obt->obt_quotachecking);
}
break;
case Q_SETQUOTA:
/* currently, it is only used for nullifying the quota */
handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
if (!rc) {
oqctl->qc_cmd = Q_SYNC;
fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
oqctl->qc_cmd = Q_SETQUOTA;
}
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
quota_unbarrier(handle);
lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
qctxt, 0);
if (lqs == NULL || IS_ERR(lqs)){
CERROR("fail to create lqs during setquota operation "
"for %sid %u\n", oqctl->qc_type ? "g" : "u",
oqctl->qc_id);
} else {
lqs->lqs_flags &= ~QB_SET;
lqs_putref(lqs);
}
break;
case Q_INITQUOTA:
{
unsigned int id[MAXQUOTAS] = { 0, 0 };
/* Initialize quota limit to MIN_QLIMIT */
LASSERT(oqctl->qc_dqblk.dqb_valid == QIF_BLIMITS);
LASSERT(oqctl->qc_dqblk.dqb_bsoftlimit == 0);
if (!oqctl->qc_dqblk.dqb_bhardlimit)
goto adjust;
/* There might be a pending dqacq/dqrel (which is going to
* clear stale limits on slave). we should wait for it's
* completion then initialize limits */
handle = quota_barrier(&obd->u.obt.obt_qctxt, oqctl, 1);
LASSERT(oqctl->qc_dqblk.dqb_bhardlimit == MIN_QLIMIT);
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
/* Update on-disk quota, in case of lose the changed limits
* (MIN_QLIMIT) on crash, which cannot be recovered.*/
if (!rc) {
oqctl->qc_cmd = Q_SYNC;
fsfilt_quotactl(obd, obd->u.obt.obt_sb, oqctl);
oqctl->qc_cmd = Q_INITQUOTA;
}
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
quota_unbarrier(handle);
if (rc)
RETURN(rc);
adjust:
lqs = quota_search_lqs(LQS_KEY(oqctl->qc_type, oqctl->qc_id),
qctxt, 1);
if (lqs == NULL || IS_ERR(lqs)){
CERROR("fail to create lqs during setquota operation "
"for %sid %u\n", oqctl->qc_type ? "g" : "u",
oqctl->qc_id);
break;
} else {
lqs->lqs_flags |= QB_SET;
lqs_putref(lqs);
}
/* Trigger qunit pre-acquire */
if (oqctl->qc_type == USRQUOTA)
id[USRQUOTA] = oqctl->qc_id;
else
id[GRPQUOTA] = oqctl->qc_id;
rc = qctxt_adjust_qunit(obd, &obd->u.obt.obt_qctxt,
id, 1, 0, NULL);
if (rc == -EDQUOT || rc == -EBUSY) {
CDEBUG(D_QUOTA, "rc: %d.\n", rc);
rc = 0;
}
break;
}
default:
CERROR("%s: unsupported filter_quotactl command: %d\n",
obd->obd_name, oqctl->qc_cmd);
RETURN(-EFAULT);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats, LQUOTA_QUOTA_CTL, timediff);
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;
}
/**
* when a block_write or inode_create rpc is finished, adjust the record for
* pending blocks and inodes
*/
static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
int pending[], int isblk)
{
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
int i;
struct qunit_data qdata[MAXQUOTAS];
ENTRY;
CDEBUG(D_QUOTA, "commit pending quota for %s\n", obd->obd_name);
CLASSERT(MAXQUOTAS < 4);
if (!ll_sb_any_quota_active(qctxt->lqc_sb))
RETURN(0);
cfs_gettimeofday(&work_start);
for (i = 0; i < MAXQUOTAS; i++) {
struct lustre_qunit_size *lqs = NULL;
LASSERT(pending[i] >= 0);
if (pending[i] == 0)
continue;
qdata[i].qd_id = id[i];
qdata[i].qd_flags = i;
if (isblk)
QDATA_SET_BLK(&qdata[i]);
qdata[i].qd_count = 0;
if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
continue;
lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
if (lqs == NULL || IS_ERR(lqs)) {
CERROR("can not find lqs for pending_commit: "
"[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
"maybe cause unexpected lqs refcount error!\n",
id[i], i ? 'g': 'u', pending[i], isblk,
lqs ? PTR_ERR(lqs) : -1);
continue;
}
cfs_spin_lock(&lqs->lqs_lock);
if (isblk) {
LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
"there are too many blocks! [id %u] [%c] "
"[bwrite_pending %lu] [pending %u]\n",
id[i], i % 2 ? 'g' : 'u',
lqs->lqs_bwrite_pending, pending[i]);
lqs->lqs_bwrite_pending -= pending[i];
} else {
LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
"there are too many files! [id %u] [%c] "
"[iwrite_pending %lu] [pending %u]\n",
id[i], i % 2 ? 'g' : 'u',
lqs->lqs_iwrite_pending, pending[i]);
lqs->lqs_iwrite_pending -= pending[i];
}
CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
obd->obd_name,
isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
i, pending[i], isblk);
cfs_spin_unlock(&lqs->lqs_lock);
/* for quota_search_lqs in pending_commit */
lqs_putref(lqs);
/* for quota_search_lqs in quota_check */
lqs_putref(lqs);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats,
isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
LQUOTA_WAIT_FOR_COMMIT_INO,
timediff);
RETURN(0);
}
static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
const unsigned int id[], int pending[],
int count, quota_acquire acquire,
struct obd_trans_info *oti, int isblk,
struct inode *inode, int frags)
{
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
struct l_wait_info lwi = { 0 };
int rc = 0, cycle = 0, count_err = 1;
ENTRY;
if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
RETURN(0);
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* If the client has been evicted or if it
* timed out and tried to reconnect already,
* abort the request immediately */
RETURN(-ENOTCONN);
CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
pending[USRQUOTA] = pending[GRPQUOTA] = 0;
/* Unfortunately, if quota master is too busy to handle the
* pre-dqacq in time and quota hash on ost is used up, we
* have to wait for the completion of in flight dqacq/dqrel,
* in order to get enough quota for write b=12588 */
cfs_gettimeofday(&work_start);
while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
inode, frags)) &
QUOTA_RET_ACQUOTA) {
cfs_spin_lock(&qctxt->lqc_lock);
if (!qctxt->lqc_import && oti) {
cfs_spin_unlock(&qctxt->lqc_lock);
LASSERT(oti && oti->oti_thread &&
oti->oti_thread->t_watchdog);
lc_watchdog_disable(oti->oti_thread->t_watchdog);
CDEBUG(D_QUOTA, "sleep for quota master\n");
l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
&lwi);
CDEBUG(D_QUOTA, "wake up when quota master is back\n");
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
} else {
cfs_spin_unlock(&qctxt->lqc_lock);
}
cycle++;
if (isblk)
OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
/* after acquire(), we should run quota_check_common again
* so that we confirm there are enough quota to finish write */
rc = acquire(obd, id, oti, isblk);
/* please reference to dqacq_completion for the below */
/* a new request is finished, try again */
if (rc == QUOTA_REQ_RETURNED) {
CDEBUG(D_QUOTA, "finish a quota req, try again\n");
continue;
}
/* it is out of quota already */
if (rc == -EDQUOT) {
CDEBUG(D_QUOTA, "out of quota, return -EDQUOT\n");
break;
}
/* Related quota has been disabled by master, but enabled by
* slave, do not try again. */
if (unlikely(rc == -ESRCH)) {
CERROR("mismatched quota configuration, stop try.\n");
break;
}
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* The client has been evicted or tried to
* to reconnect already, abort the request */
RETURN(-ENOTCONN);
/* -EBUSY and others, wait a second and try again */
if (rc < 0) {
cfs_waitq_t waitq;
struct l_wait_info lwi;
if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
count_err++);
cfs_waitq_init(&waitq);
lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
NULL);
l_wait_event(waitq, 0, &lwi);
}
if (rc < 0 || cycle % 10 == 0) {
cfs_spin_lock(&last_print_lock);
if (last_print == 0 ||
cfs_time_before((last_print + cfs_time_seconds(30)),
cfs_time_current())) {
last_print = cfs_time_current();
cfs_spin_unlock(&last_print_lock);
CWARN("still haven't managed to acquire quota "
"space from the quota master after %d "
"retries (err=%d, rc=%d)\n",
cycle, count_err - 1, rc);
} else {
cfs_spin_unlock(&last_print_lock);
}
}
CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
cycle);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats,
isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
LQUOTA_WAIT_FOR_CHK_INO,
timediff);
if (rc > 0)
rc = 0;
RETURN(rc);
}
