* returned and may be compared against out object.
*/
/* In the function below, .hs_keycmp resolves to
- * conn_keycmp() */
+ * conn_keycmp()
+ */
/* coverity[overrun-buffer-val] */
conn2 = cfs_hash_findadd_unique(conn_hash, &peer, &conn->c_hash);
if (conn != conn2) {
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... */
+ * like failing sends in client.c does currently...
+ */
req->rq_net_err = 1;
ptlrpc_client_wake_req(req);
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. */
+ * for adaptive timeouts' early reply.
+ */
LASSERT((ev->md.options & LNET_MD_MANAGE_REMOTE) != 0);
spin_lock(&req->rq_lock);
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 */
+ * 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),
out_wake:
/* NB don't unlock till after wakeup; req can disappear under us
- * since we don't have our own ref */
+ * since we don't have our own ref
+ */
ptlrpc_client_wake_req(req);
spin_unlock(&req->rq_lock);
}
desc->bd_failure = 1;
/* NB don't unlock till after wakeup; desc can disappear under us
- * otherwise */
+ * otherwise
+ */
if (desc->bd_md_count == 0)
ptlrpc_client_wake_req(desc->bd_req);
__u64 new_seq;
/* set sequence ID for request and add it to history list,
- * it must be called with hold svcpt::scp_lock */
+ * it must be called with hold svcpt::scp_lock
+ */
new_seq = (sec << REQS_SEC_SHIFT) |
(usec << REQS_USEC_SHIFT) |
if (new_seq > svcpt->scp_hist_seq) {
/* This handles the initial case of scp_hist_seq == 0 or
- * we just jumped into a new time window */
+ * we just jumped into a new time window
+ */
svcpt->scp_hist_seq = new_seq;
} else {
LASSERT(REQS_SEQ_SHIFT(svcpt) < REQS_USEC_SHIFT);
* however, it's possible that we used up all bits for
* sequence and jumped into the next usec bucket (future time),
* then we hope there will be less RPCs per bucket at some
- * point, and sequence will catch up again */
+ * point, and sequence will catch up again
+ */
svcpt->scp_hist_seq += (1U << REQS_SEQ_SHIFT(svcpt));
new_seq = svcpt->scp_hist_seq;
}
* request buffer we can use the request object embedded in
* rqbd. Note that if we failed to allocate a request,
* we'd have to re-post the rqbd, which we can't do in this
- * context. */
+ * context.
+ */
req = &rqbd->rqbd_req;
memset(req, 0, sizeof(*req));
} else {
/* NB we ABSOLUTELY RELY on req being zeroed, so pointers are NULL,
* flags are reset and scalars are zero. We only set the message
- * size to non-zero if this was a successful receive. */
+ * size to non-zero if this was a successful receive.
+ */
req->rq_xid = ev->match_bits;
req->rq_reqbuf = ev->md.start + ev->offset;
if (ev->type == LNET_EVENT_PUT && ev->status == 0)
svcpt->scp_nrqbds_posted);
/* Normally, don't complain about 0 buffers posted; LNET won't
- * drop incoming reqs since we set the portal lazy */
+ * drop incoming reqs since we set the portal lazy
+ */
if (test_req_buffer_pressure &&
ev->type != LNET_EVENT_UNLINK &&
svcpt->scp_nrqbds_posted == 0)
svcpt->scp_nreqs_incoming++;
/* NB everything can disappear under us once the request
- * has been queued and we unlock, so do the wake now... */
+ * has been queued and we unlock, so do the wake now...
+ */
wake_up(&svcpt->scp_waitq);
spin_unlock(&svcpt->scp_lock);
if (!rs->rs_difficult) {
/* 'Easy' replies have no further processing so I drop the
- * net's ref on 'rs' */
+ * net's ref on 'rs'
+ */
LASSERT(ev->unlinked);
ptlrpc_rs_decref(rs);
return;
if (ev->unlinked) {
/* Last network callback. The net's ref on 'rs' stays put
- * until ptlrpc_handle_rs() is done with it */
+ * until ptlrpc_handle_rs() is done with it
+ */
spin_lock(&svcpt->scp_rep_lock);
spin_lock(&rs->rs_lock);
/* Wait for the event queue to become idle since there may still be
* messages in flight with pending events (i.e. the fire-and-forget
* messages == client requests and "non-difficult" server
- * replies */
+ * replies
+ */
for (retries = 0;; retries++) {
rc = LNetEQFree(ptlrpc_eq_h);
}
/* CAVEAT EMPTOR: how we process portals events is _radically_
- * different depending on... */
+ * different depending on...
+ */
/* kernel LNet calls our master callback when there are new event,
* because we are guaranteed to get every event via callback,
* so we just set EQ size to 0 to avoid overhead of serializing
- * enqueue/dequeue operations in LNet. */
+ * enqueue/dequeue operations in LNet.
+ */
rc = LNetEQAlloc(0, ptlrpc_master_callback, &ptlrpc_eq_h);
if (rc == 0)
return 0;
* CLOSED. I would rather refcount the import and free it after
* disconnection like we do with exports. To do that, the client_obd
* will need to save the peer info somewhere other than in the import,
- * though. */
+ * though.
+ */
int ptlrpc_init_import(struct obd_import *imp)
{
spin_lock(&imp->imp_lock);
/* Wait forever until inflight == 0. We really can't do it another
* way because in some cases we need to wait for very long reply
* unlink. We can't do anything before that because there is really
- * no guarantee that some rdma transfer is not in progress right now. */
+ * no guarantee that some rdma transfer is not in progress right now.
+ */
do {
/* Calculate max timeout for waiting on rpcs to error
* out. Use obd_timeout if calculated value is smaller
- * than it. */
+ * than it.
+ */
if (!OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK)) {
timeout = ptlrpc_inflight_timeout(imp);
timeout += timeout / 3;
/* Wait for all requests to error out and call completion
* callbacks. Cap it at obd_timeout -- these should all
- * have been locally cancelled by ptlrpc_abort_inflight. */
+ * have been locally cancelled by ptlrpc_abort_inflight.
+ */
lwi = LWI_TIMEOUT_INTERVAL(
cfs_timeout_cap(cfs_time_seconds(timeout)),
(timeout > 1)?cfs_time_seconds(1):cfs_time_seconds(1)/2,
* maybe waiting for long reply unlink in
* sluggish nets). Let's check this. If there
* is no inflight and unregistering != 0, this
- * is bug. */
+ * is bug.
+ */
LASSERTF(count == 0, "Some RPCs are still unregistering: %d\n",
count);
/* Let's save one loop as soon as inflight have
* dropped to zero. No new inflights possible at
- * this point. */
+ * this point.
+ */
rc = 0;
} else {
list_for_each_safe(tmp, n,
conn->oic_last_attempt);
/* If we have not tried this connection since
- the last successful attempt, go with this one */
+ * the last successful attempt, go with this one
+ */
if ((conn->oic_last_attempt == 0) ||
cfs_time_beforeq_64(conn->oic_last_attempt,
imp->imp_last_success_conn)) {
}
/* If all of the connections have already been tried
- since the last successful connection; just choose the
- least recently used */
+ * since the last successful connection; just choose the
+ * least recently used
+ */
if (!imp_conn)
imp_conn = conn;
else if (cfs_time_before_64(conn->oic_last_attempt,
LASSERT(imp_conn->oic_conn);
/* If we've tried everything, and we're back to the beginning of the
- list, increase our timeout and try again. It will be reset when
- we do finally connect. (FIXME: really we should wait for all network
- state associated with the last connection attempt to drain before
- trying to reconnect on it.) */
+ * list, increase our timeout and try again. It will be reset when
+ * we do finally connect. (FIXME: really we should wait for all network
+ * state associated with the last connection attempt to drain before
+ * trying to reconnect on it.)
+ */
if (tried_all && (imp->imp_conn_list.next == &imp_conn->oic_item)) {
struct adaptive_timeout *at = &imp->imp_at.iat_net_latency;
struct list_head *tmp;
/* The requests in committed_list always have smaller transnos than
- * the requests in replay_list */
+ * the requests in replay_list
+ */
if (!list_empty(&imp->imp_committed_list)) {
tmp = imp->imp_committed_list.next;
req = list_entry(tmp, struct ptlrpc_request, rq_replay_list);
goto out;
/* Reset connect flags to the originally requested flags, in case
- * the server is updated on-the-fly we will get the new features. */
+ * the server is updated on-the-fly we will get the new features.
+ */
imp->imp_connect_data.ocd_connect_flags = imp->imp_connect_flags_orig;
/* Reset ocd_version each time so the server knows the exact versions */
imp->imp_connect_data.ocd_version = LUSTRE_VERSION_CODE;
}
/* Report the rpc service time to the server so that it knows how long
- * to wait for clients to join recovery */
+ * to wait for clients to join recovery
+ */
lustre_msg_set_service_time(request->rq_reqmsg,
at_timeout2est(request->rq_timeout));
* import_select_connection will increase the net latency on
* repeated reconnect attempts to cover slow networks.
* We override/ignore the server rpc completion estimate here,
- * which may be large if this is a reconnect attempt */
+ * which may be large if this is a reconnect attempt
+ */
request->rq_timeout = INITIAL_CONNECT_TIMEOUT;
lustre_msg_set_timeout(request->rq_reqmsg, request->rq_timeout);
if (rc) {
/* if this reconnect to busy export - not need select new target
- * for connecting*/
+ * for connecting
+ */
imp->imp_force_reconnect = ptlrpc_busy_reconnect(rc);
spin_unlock(&imp->imp_lock);
ptlrpc_maybe_ping_import_soon(imp);
if (!exp) {
/* This could happen if export is cleaned during the
- connect attempt */
+ * connect attempt
+ */
CERROR("%s: missing export after connect\n",
imp->imp_obd->obd_name);
rc = -ENODEV;
}
/* if applies, adjust the imp->imp_msg_magic here
- * according to reply flags */
+ * according to reply flags
+ */
imp->imp_remote_handle =
*lustre_msg_get_handle(request->rq_repmsg);
/* Initial connects are allowed for clients with non-random
* uuids when servers are in recovery. Simply signal the
- * servers replay is complete and wait in REPLAY_WAIT. */
+ * servers replay is complete and wait in REPLAY_WAIT.
+ */
if (msg_flags & MSG_CONNECT_RECOVERING) {
CDEBUG(D_HA, "connect to %s during recovery\n",
obd2cli_tgt(imp->imp_obd));
* already erased all of our state because of previous
* eviction. If it is in recovery - we are safe to
* participate since we can reestablish all of our state
- * with server again */
+ * with server again
+ */
if ((msg_flags & MSG_CONNECT_RECOVERING)) {
CDEBUG(level, "%s@%s changed server handle from %#llx to %#llx but is still in recovery\n",
obd2cli_tgt(imp->imp_obd),
ocd->ocd_version < LUSTRE_VERSION_CODE -
LUSTRE_VERSION_OFFSET_WARN)) {
/* Sigh, some compilers do not like #ifdef in the middle
- of macro arguments */
+ * of macro arguments
+ */
const char *older = "older. Consider upgrading server or downgrading client"
;
const char *newer = "newer than client version. Consider upgrading client"
* fixup is version-limited, because we don't want to carry the
* OBD_CONNECT_MNE_SWAB flag around forever, just so long as we
* need interop with unpatched 2.2 servers. For newer servers,
- * the client will do MNE swabbing only as needed. LU-1644 */
+ * the client will do MNE swabbing only as needed. LU-1644
+ */
if (unlikely((ocd->ocd_connect_flags & OBD_CONNECT_VERSION) &&
!(ocd->ocd_connect_flags & OBD_CONNECT_MNE_SWAB) &&
OBD_OCD_VERSION_MAJOR(ocd->ocd_version) == 2 &&
if (ocd->ocd_connect_flags & OBD_CONNECT_CKSUM) {
/* We sent to the server ocd_cksum_types with bits set
* for algorithms we understand. The server masked off
- * the checksum types it doesn't support */
+ * the checksum types it doesn't support
+ */
if ((ocd->ocd_cksum_types &
cksum_types_supported_client()) == 0) {
LCONSOLE_WARN("The negotiation of the checksum algorithm to use with server %s failed (%x/%x), disabling checksums\n",
}
} else {
/* The server does not support OBD_CONNECT_CKSUM.
- * Enforce ADLER for backward compatibility*/
+ * Enforce ADLER for backward compatibility
+ */
cli->cl_supp_cksum_types = OBD_CKSUM_ADLER;
}
cli->cl_cksum_type = cksum_type_select(cli->cl_supp_cksum_types);
/* Reset ns_connect_flags only for initial connect. It might be
* changed in while using FS and if we reset it in reconnect
* this leads to losing user settings done before such as
- * disable lru_resize, etc. */
+ * disable lru_resize, etc.
+ */
if (old_connect_flags != exp_connect_flags(exp) ||
aa->pcaa_initial_connect) {
CDEBUG(D_HA, "%s: Resetting ns_connect_flags to server flags: %#llx\n",
if ((ocd->ocd_connect_flags & OBD_CONNECT_AT) &&
(imp->imp_msg_magic == LUSTRE_MSG_MAGIC_V2))
/* We need a per-message support flag, because
- a. we don't know if the incoming connect reply
- supports AT or not (in reply_in_callback)
- until we unpack it.
- b. failovered server means export and flags are gone
- (in ptlrpc_send_reply).
- Can only be set when we know AT is supported at
- both ends */
+ * a. we don't know if the incoming connect reply
+ * supports AT or not (in reply_in_callback)
+ * until we unpack it.
+ * b. failovered server means export and flags are gone
+ * (in ptlrpc_send_reply).
+ * Can only be set when we know AT is supported at
+ * both ends
+ */
imp->imp_msghdr_flags |= MSGHDR_AT_SUPPORT;
else
imp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT;
/* bug 17802: XXX client_disconnect_export vs connect request
* race. if client will evicted at this time, we start
* invalidate thread without reference to import and import can
- * be freed at same time. */
+ * be freed at same time.
+ */
class_import_get(imp);
task = kthread_run(ptlrpc_invalidate_import_thread, imp,
"ll_imp_inval");
if (req) {
/* We are disconnecting, do not retry a failed DISCONNECT rpc if
* it fails. We can get through the above with a down server
- * if the client doesn't know the server is gone yet. */
+ * if the client doesn't know the server is gone yet.
+ */
req->rq_no_resend = 1;
/* We want client umounts to happen quickly, no matter the
- server state... */
+ * server state...
+ */
req->rq_timeout = min_t(int, req->rq_timeout,
INITIAL_CONNECT_TIMEOUT);
extern unsigned int at_min, at_max, at_history;
/* Bin into timeslices using AT_BINS bins.
- This gives us a max of the last binlimit*AT_BINS secs without the storage,
- but still smoothing out a return to normalcy from a slow response.
- (E.g. remember the maximum latency in each minute of the last 4 minutes.) */
+ * This gives us a max of the last binlimit*AT_BINS secs without the storage,
+ * but still smoothing out a return to normalcy from a slow response.
+ * (E.g. remember the maximum latency in each minute of the last 4 minutes.)
+ */
int at_measured(struct adaptive_timeout *at, unsigned int val)
{
unsigned int old = at->at_current;
if (val == 0)
/* 0's don't count, because we never want our timeout to
- drop to 0, and because 0 could mean an error */
+ * drop to 0, and because 0 could mean an error
+ */
return 0;
spin_lock(&at->at_lock);
if (at->at_flags & AT_FLG_NOHIST)
/* Only keep last reported val; keeping the rest of the history
- for proc only */
+ * for debugfs only
+ */
at->at_current = val;
if (at_max > 0)
} while (0)
/* This is a callback from the llog_* functions.
- * Assumes caller has already pushed us into the kernel context. */
+ * Assumes caller has already pushed us into the kernel context.
+ */
static int llog_client_open(const struct lu_env *env,
struct llog_handle *lgh, struct llog_logid *logid,
char *name, enum llog_open_param open_param)
struct llog_handle *handle)
{
/* this doesn't call LLOG_ORIGIN_HANDLE_CLOSE because
- the servers all close the file at the end of every
- other LLOG_ RPC. */
+ * the servers all close the file at the end of every
+ * other LLOG_ RPC.
+ */
return 0;
}
/* This sanity check is more of an insanity check; we can still
* hose a kernel by allowing the request history to grow too
- * far. */
+ * far.
+ */
bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
if (val > totalram_pages / (2 * bufpages))
return -ERANGE;
* recent), search from it onwards.
* Since the service history is LRU (i.e. culled reqs will
* be near the head), we shouldn't have to do long
- * re-scans */
+ * re-scans
+ */
LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
"%s:%d: seek seq %llu, request seq %llu\n",
svcpt->scp_service->srv_name, svcpt->scp_cpt,
* here. The request could contain any old crap, so you
* must be just as careful as the service's request
* parser. Currently I only print stuff here I know is OK
- * to look at coz it was set up in request_in_callback()!!! */
+ * to look at coz it was set up in request_in_callback()!!!
+ */
seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld:%lds(%+lds) ",
req->rq_history_seq, nidstr,
libcfs_id2str(req->rq_peer), req->rq_xid,
int rc2;
/* We're going to get an UNLINK event when I unlink below,
* which will complete just like any other failed send, so
- * I fall through and return success here! */
+ * I fall through and return success here!
+ */
CERROR("LNetPut(%s, %d, %lld) failed: %d\n",
libcfs_id2str(conn->c_peer), portal, xid, rc);
rc2 = LNetMDUnlink(*mdh);
* using the same RDMA match bits after an error.
*
* For multi-bulk RPCs, rq_xid is the last XID needed for bulks. The
- * first bulk XID is power-of-two aligned before rq_xid. LU-1431 */
+ * first bulk XID is power-of-two aligned before rq_xid. LU-1431
+ */
xid = req->rq_xid & ~((__u64)desc->bd_md_max_brw - 1);
LASSERTF(!(desc->bd_registered &&
req->rq_send_state != LUSTRE_IMP_REPLAY) ||
}
/* Set rq_xid to matchbits of the final bulk so that server can
- * infer the number of bulks that were prepared */
+ * infer the number of bulks that were prepared
+ */
req->rq_xid = --xid;
LASSERTF(desc->bd_last_xid == (req->rq_xid & PTLRPC_BULK_OPS_MASK),
"bd_last_xid = x%llu, rq_xid = x%llu\n",
/* 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() */
+ * 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 */
for (;;) {
/* Network access will complete in finite time but the HUGE
- * timeout lets us CWARN for visibility of sluggish NALs */
+ * timeout lets us CWARN for visibility of sluggish LNDs
+ */
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);
(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 */
+ * toward our service time estimate
+ */
int oldse = at_measured(&svcpt->scp_at_estimate, service_time);
if (oldse != 0) {
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) */
+ * (bz15815)
+ */
if (req->rq_type == PTL_RPC_MSG_ERR && !req->rq_export)
lustre_msg_set_timeout(req->rq_repmsg, 0);
else
LASSERT(request->rq_wait_ctx == 0);
/* If this is a re-transmit, we're required to have disengaged
- * cleanly from the previous attempt */
+ * 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)));
request->rq_replen);
if (rc) {
/* this prevents us from looping in
- * ptlrpc_queue_wait */
+ * ptlrpc_queue_wait
+ */
spin_lock(&request->rq_lock);
request->rq_err = 1;
spin_unlock(&request->rq_lock);
reply_md.eq_handle = ptlrpc_eq_h;
/* We must see the unlink callback to unset rq_reply_unlink,
- so we can't auto-unlink */
+ * so we can't auto-unlink
+ */
rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
&request->rq_reply_md_h);
if (rc != 0) {
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. */
+ * add the network latency for our local timeout.
+ */
request->rq_deadline = request->rq_sent + request->rq_timeout +
ptlrpc_at_get_net_latency(request);
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. */
+ * access the reply buffer.
+ */
rc2 = LNetMEUnlink(reply_me_h);
LASSERT(rc2 == 0);
/* UNLINKED callback called synchronously */
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. */
+ * the chance to have long unlink to sluggish net is smaller here.
+ */
ptlrpc_unregister_bulk(request, 0);
out:
if (request->rq_memalloc)
/* NB: CPT affinity service should use new LNet flag LNET_INS_LOCAL,
* which means buffer can only be attached on local CPT, and LND
- * threads can find it by grabbing a local lock */
+ * threads can find it by grabbing a local lock
+ */
rc = LNetMEAttach(service->srv_req_portal,
match_id, 0, ~0, LNET_UNLINK,
rqbd->rqbd_svcpt->scp_cpt >= 0 ?
#include "../../include/linux/libcfs/libcfs.h"
#include "ptlrpc_internal.h"
-/* XXX: This is just for liblustre. Remove the #if defined directive when the
- * "cfs_" prefix is dropped from cfs_list_head. */
-
/**
* NRS core object.
*/
if (req->rq_nrq.nr_initialized) {
nrs_resource_put_safe(req->rq_nrq.nr_res_ptrs);
/* no protection on bit nr_initialized because no
- * contention at this late stage */
+ * contention at this late stage
+ */
req->rq_nrq.nr_finalized = 1;
}
}
* NOTE: this should only be used for NEW requests, and should always be
* in the form of a v2 request. If this is a connection to a v1
* target then the first buffer will be stripped because the ptlrpc
- * data is part of the lustre_msg_v1 header. b=14043 */
+ * data is part of the lustre_msg_v1 header. b=14043
+ */
int lustre_msg_size(__u32 magic, int count, __u32 *lens)
{
__u32 size[] = { sizeof(struct ptlrpc_body) };
EXPORT_SYMBOL(lustre_msg_size);
/* This is used to determine the size of a buffer that was already packed
- * and will correctly handle the different message formats. */
+ * and will correctly handle the different message formats.
+ */
int lustre_packed_msg_size(struct lustre_msg *msg)
{
switch (msg->lm_magic) {
spin_unlock(&svcpt->scp_rep_lock);
/* If we cannot get anything for some long time, we better
- * bail out instead of waiting infinitely */
+ * bail out instead of waiting infinitely
+ */
lwi = LWI_TIMEOUT(cfs_time_seconds(10), NULL, NULL);
rc = l_wait_event(svcpt->scp_rep_waitq,
!list_empty(&svcpt->scp_rep_idle), &lwi);
EXPORT_SYMBOL(lustre_msg_buflen);
/* NB return the bufcount for lustre_msg_v2 format, so if message is packed
- * in V1 format, the result is one bigger. (add struct ptlrpc_body). */
+ * in V1 format, the result is one bigger. (add struct ptlrpc_body).
+ */
int lustre_msg_bufcount(struct lustre_msg *m)
{
switch (m->lm_magic) {
/* no break */
default:
/* flags might be printed in debug code while message
- * uninitialized */
+ * uninitialized
+ */
return 0;
}
}
/* no break */
default:
/* status might be printed in debug code while message
- * uninitialized */
+ * uninitialized
+ */
return -EINVAL;
}
}
struct ptlrpc_body *pb;
/* Don't set jobid for ldlm ast RPCs, they've been shrunk.
- * See the comment in ptlrpc_request_pack(). */
+ * See the comment in ptlrpc_request_pack().
+ */
if (!opc || opc == LDLM_BL_CALLBACK ||
opc == LDLM_CP_CALLBACK || opc == LDLM_GL_CALLBACK)
return;
* clients and servers without ptlrpc_body_v2 (< 2.3)
* do not swab any fields beyond pb_jobid, as we are
* using this swab function for both ptlrpc_body
- * and ptlrpc_body_v2. */
+ * and ptlrpc_body_v2.
+ */
CLASSERT(offsetof(typeof(*b), pb_jobid) != 0);
}
EXPORT_SYMBOL(lustre_swab_ptlrpc_body);
__swab32s(&ocd->ocd_index);
__swab32s(&ocd->ocd_brw_size);
/* ocd_blocksize and ocd_inodespace don't need to be swabbed because
- * they are 8-byte values */
+ * they are 8-byte values
+ */
__swab16s(&ocd->ocd_grant_extent);
__swab32s(&ocd->ocd_unused);
__swab64s(&ocd->ocd_transno);
/* Fields after ocd_cksum_types are only accessible by the receiver
* if the corresponding flag in ocd_connect_flags is set. Accessing
* any field after ocd_maxbytes on the receiver without a valid flag
- * may result in out-of-bound memory access and kernel oops. */
+ * may result in out-of-bound memory access and kernel oops.
+ */
if (ocd->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)
__swab32s(&ocd->ocd_max_easize);
if (ocd->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
{
/* the lock data is a union and the first two fields are always an
* extent so it's ok to process an LDLM_EXTENT and LDLM_FLOCK lock
- * data the same way. */
+ * data the same way.
+ */
__swab64s(&d->l_extent.start);
__swab64s(&d->l_extent.end);
__swab64s(&d->l_extent.gid);
/* Wait until the next ping time, or until we're stopped. */
time_to_next_wake = pinger_check_timeout(this_ping);
/* The ping sent by ptlrpc_send_rpc may get sent out
- say .01 second after this.
- ptlrpc_pinger_sending_on_import will then set the
- next ping time to next_ping + .01 sec, which means
- we will SKIP the next ping at next_ping, and the
- ping will get sent 2 timeouts from now! Beware. */
+ * say .01 second after this.
+ * ptlrpc_pinger_sending_on_import will then set the
+ * next ping time to next_ping + .01 sec, which means
+ * we will SKIP the next ping at next_ping, and the
+ * ping will get sent 2 timeouts from now! Beware.
+ */
CDEBUG(D_INFO, "next wakeup in " CFS_DURATION_T " (%ld)\n",
time_to_next_wake,
cfs_time_add(this_ping,
* registration/unregistration, and NRS core lprocfs operations.
*/
struct mutex nrs_mutex;
- /* XXX: This is just for liblustre. Remove the #if defined directive
- * when the * "cfs_" prefix is dropped from cfs_list_head. */
/**
* List of all policy descriptors registered with NRS core; protected
* by nrs_core::nrs_mutex.
l_wait_event(req->rq_set_waitq, !req->rq_set, &lwi);
} else if (req->rq_set) {
/* If we have a valid "rq_set", just reuse it to avoid double
- * linked. */
+ * linked.
+ */
LASSERT(req->rq_phase == RQ_PHASE_NEW);
LASSERT(req->rq_send_state == LUSTRE_IMP_REPLAY);
rc |= ptlrpc_check_set(env, set);
/* NB: ptlrpc_check_set has already moved completed request at the
- * head of seq::set_requests */
+ * head of seq::set_requests
+ */
list_for_each_safe(pos, tmp, &set->set_requests) {
req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
if (req->rq_phase != RQ_PHASE_COMPLETE)
rc = atomic_read(&set->set_new_count);
/* If we have nothing to do, check whether we can take some
- * work from our partner threads. */
+ * work from our partner threads.
+ */
if (rc == 0 && pc->pc_npartners > 0) {
struct ptlrpcd_ctl *partner;
struct ptlrpc_request_set *ps;
if (req->rq_transno > last_transno) {
/* Since the imp_committed_list is immutable before
* all of it's requests being replayed, it's safe to
- * use a cursor to accelerate the search */
+ * use a cursor to accelerate the search
+ */
imp->imp_replay_cursor = imp->imp_replay_cursor->next;
while (imp->imp_replay_cursor !=
}
/* All the requests in committed list have been replayed, let's replay
- * the imp_replay_list */
+ * the imp_replay_list
+ */
if (!req) {
list_for_each_safe(tmp, pos, &imp->imp_replay_list) {
req = list_entry(tmp, struct ptlrpc_request,
/* If need to resend the last sent transno (because a reconnect
* has occurred), then stop on the matching req and send it again.
* If, however, the last sent transno has been committed then we
- * continue replay from the next request. */
+ * continue replay from the next request.
+ */
if (req && imp->imp_resend_replay)
lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
}
/* Wait for recovery to complete and resend. If evicted, then
- this request will be errored out later.*/
+ * this request will be errored out later.
+ */
spin_lock(&failed_req->rq_lock);
if (!failed_req->rq_no_resend)
failed_req->rq_resend = 1;
LASSERT(obd);
/* When deactivating, mark import invalid, and abort in-flight
- * requests. */
+ * requests.
+ */
if (!active) {
LCONSOLE_WARN("setting import %s INACTIVE by administrator request\n",
obd2cli_tgt(imp->imp_obd));
/* set before invalidate to avoid messages about imp_inval
- * set without imp_deactive in ptlrpc_import_delay_req */
+ * set without imp_deactive in ptlrpc_import_delay_req
+ */
spin_lock(&imp->imp_lock);
imp->imp_deactive = 1;
spin_unlock(&imp->imp_lock);
/* alloc new request buffer
* we don't need to alloc reply buffer here, leave it to the
- * rest procedure of ptlrpc */
+ * rest procedure of ptlrpc
+ */
if (reqmsg_size != 0) {
rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
if (!rc) {
spin_unlock(&sec->ps_lock);
/* force SVC_NULL for context initiation rpc, SVC_INTG for context
- * destruction rpc */
+ * destruction rpc
+ */
if (unlikely(req->rq_ctx_init))
flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
else if (unlikely(req->rq_ctx_fini))
return 0;
/* client side export has no imp_reverse, skip
- * FIXME maybe we should check flavor this as well??? */
+ * FIXME maybe we should check flavor this as well???
+ */
if (!exp->exp_imp_reverse)
return 0;
* the first req with the new flavor, then treat it as current flavor,
* adapt reverse sec according to it.
* note the first rpc with new flavor might not be with root ctx, in
- * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
+ * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
+ */
if (unlikely(exp->exp_flvr_changed) &&
flavor_allowed(&exp->exp_flvr_old[1], req)) {
/* make the new flavor as "current", and old ones as
- * about-to-expire */
+ * about-to-expire
+ */
CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
flavor = exp->exp_flvr_old[1];
}
/* if it equals to the current flavor, we accept it, but need to
- * dealing with reverse sec/ctx */
+ * dealing with reverse sec/ctx
+ */
if (likely(flavor_allowed(&exp->exp_flvr, req))) {
/* most cases should return here, we only interested in
- * gss root ctx init */
+ * gss root ctx init
+ */
if (!req->rq_auth_gss || !req->rq_ctx_init ||
(!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
!req->rq_auth_usr_ost)) {
/* if flavor just changed, we should not proceed, just leave
* it and current flavor will be discovered and replaced
- * shortly, and let _this_ rpc pass through */
+ * shortly, and let _this_ rpc pass through
+ */
if (exp->exp_flvr_changed) {
LASSERT(exp->exp_flvr_adapt);
spin_unlock(&exp->exp_lock);
}
/* now it doesn't match the current flavor, the only chance we can
- * accept it is match the old flavors which is not expired. */
+ * accept it is match the old flavors which is not expired.
+ */
if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
* is not optimal. we perhaps want to use balanced binary tree
* to trace each sec as order of expiry time.
* another issue here is we wakeup as fixed interval instead of
- * according to each sec's expiry time */
+ * according to each sec's expiry time
+ */
mutex_lock(&sec_gc_mutex);
list_for_each_entry(sec, &sec_gc_list, ps_gc_list) {
/* if someone is waiting to be deleted, let it
- * proceed as soon as possible. */
+ * proceed as soon as possible.
+ */
if (atomic_read(&sec_gc_wait_del)) {
CDEBUG(D_SEC, "deletion pending, start over\n");
mutex_unlock(&sec_gc_mutex);
* imp_replay_list traversing threads. See LU-3333
* This is a bandaid at best, we really need to deal with this
* in request enlarging code before unpacking that's already
- * there */
+ * there
+ */
if (req->rq_import)
spin_lock(&req->rq_import->imp_lock);
memcpy(newbuf, req->rq_reqbuf, req->rq_reqlen);
}
} else {
/* whether we sent with bulk or not, we expect the same
- * in reply, except for early reply */
+ * in reply, except for early reply
+ */
if (!req->rq_early &&
!equi(req->rq_pack_bulk == 1,
phdr->ph_flags & PLAIN_FL_BULK)) {
* imp_replay_list traversing threads. See LU-3333
* This is a bandaid at best, we really need to deal with this
* in request enlarging code before unpacking that's already
- * there */
+ * there
+ */
if (req->rq_import)
spin_lock(&req->rq_import->imp_lock);
for (i = 0; i < svc->srv_nbuf_per_group; i++) {
/* NB: another thread might have recycled enough rqbds, we
- * need to make sure it wouldn't over-allocate, see LU-1212. */
+ * need to make sure it wouldn't over-allocate, see LU-1212.
+ */
if (svcpt->scp_nrqbds_posted >= svc->srv_nbuf_per_group)
break;
list_add_tail(&rqbd->rqbd_list, &svcpt->scp_rqbd_idle);
/* Don't complain if no request buffers are posted right now; LNET
- * won't drop requests because we set the portal lazy! */
+ * won't drop requests because we set the portal lazy!
+ */
spin_unlock(&svcpt->scp_lock);
init = max_t(int, init, tc->tc_nthrs_init);
/* NB: please see comments in lustre_lnet.h for definition
- * details of these members */
+ * details of these members
+ */
LASSERT(tc->tc_nthrs_max != 0);
if (tc->tc_nthrs_user != 0) {
/* In case there is a reason to test a service with many
* threads, we give a less strict check here, it can
- * be up to 8 * nthrs_max */
+ * be up to 8 * nthrs_max
+ */
total = min(tc->tc_nthrs_max * 8, tc->tc_nthrs_user);
nthrs = total / svc->srv_ncpts;
init = max(init, nthrs);
total = tc->tc_nthrs_max;
if (tc->tc_nthrs_base == 0) {
/* don't care about base threads number per partition,
- * this is most for non-affinity service */
+ * this is most for non-affinity service
+ */
nthrs = total / svc->srv_ncpts;
goto out;
}
/* NB: Increase the base number if it's single partition
* and total number of cores/HTs is larger or equal to 4.
- * result will always < 2 * nthrs_base */
+ * result will always < 2 * nthrs_base
+ */
weight = cfs_cpt_weight(svc->srv_cptable, CFS_CPT_ANY);
for (i = 1; (weight >> (i + 1)) != 0 && /* >= 4 cores/HTs */
(tc->tc_nthrs_base >> i) != 0; i++)
(unsigned long)svcpt);
/* At SOW, service time should be quick; 10s seems generous. If client
- * timeout is less than this, we'll be sending an early reply. */
+ * timeout is less than this, we'll be sending an early reply.
+ */
at_init(&svcpt->scp_at_estimate, 10, 0);
/* assign this before call ptlrpc_grow_req_bufs */
/* Now allocate the request buffers, but don't post them now */
rc = ptlrpc_grow_req_bufs(svcpt, 0);
/* We shouldn't be under memory pressure at startup, so
- * fail if we can't allocate all our buffers at this time. */
+ * fail if we can't allocate all our buffers at this time.
+ */
if (rc != 0)
goto free_reqs_count;
LASSERT(list_empty(&req->rq_timed_list));
/* DEBUG_REQ() assumes the reply state of a request with a valid
- * ref will not be destroyed until that reference is dropped. */
+ * ref will not be destroyed until that reference is dropped.
+ */
ptlrpc_req_drop_rs(req);
sptlrpc_svc_ctx_decref(req);
if (req != &req->rq_rqbd->rqbd_req) {
/* NB request buffers use an embedded
* req if the incoming req unlinked the
- * MD; this isn't one of them! */
+ * MD; this isn't one of them!
+ */
ptlrpc_request_cache_free(req);
}
}
if (req->rq_at_linked) {
spin_lock(&svcpt->scp_at_lock);
/* recheck with lock, in case it's unlinked by
- * ptlrpc_at_check_timed() */
+ * ptlrpc_at_check_timed()
+ */
if (likely(req->rq_at_linked))
ptlrpc_at_remove_timed(req);
spin_unlock(&svcpt->scp_at_lock);
svcpt->scp_hist_nrqbds++;
/* cull some history?
- * I expect only about 1 or 2 rqbds need to be recycled here */
+ * I expect only about 1 or 2 rqbds need to be recycled here
+ */
while (svcpt->scp_hist_nrqbds > svc->srv_hist_nrqbds_cpt_max) {
rqbd = list_entry(svcpt->scp_hist_rqbds.next,
struct ptlrpc_request_buffer_desc,
svcpt->scp_hist_nrqbds--;
/* remove rqbd's reqs from svc's req history while
- * I've got the service lock */
+ * I've got the service lock
+ */
list_for_each(tmp, &rqbd->rqbd_reqs) {
req = list_entry(tmp, struct ptlrpc_request,
rq_list);
div_u64_rem(req->rq_deadline, array->paa_size, &index);
if (array->paa_reqs_count[index] > 0) {
/* latest rpcs will have the latest deadlines in the list,
- * so search backward. */
+ * so search backward.
+ */
list_for_each_entry_reverse(rq,
&array->paa_reqs_array[index],
rq_timed_list) {
int rc;
/* deadline is when the client expects us to reply, margin is the
- difference between clients' and servers' expectations */
+ * difference between clients' and servers' expectations
+ */
DEBUG_REQ(D_ADAPTTO, req,
"%ssending early reply (deadline %+lds, margin %+lds) for %d+%d",
AT_OFF ? "AT off - not " : "",
}
/* Fake our processing time into the future to ask the clients
- * for some extra amount of time */
+ * for some extra amount of time
+ */
at_measured(&svcpt->scp_at_estimate, at_extra +
ktime_get_real_seconds() - req->rq_arrival_time.tv_sec);
/* Check to see if we've actually increased the deadline -
- * we may be past adaptive_max */
+ * we may be past adaptive_max
+ */
if (req->rq_deadline >= req->rq_arrival_time.tv_sec +
at_get(&svcpt->scp_at_estimate)) {
DEBUG_REQ(D_WARNING, req, "Couldn't add any time (%ld/%lld), not sending early reply\n",
}
/* Free the (early) reply state from lustre_pack_reply.
- (ptlrpc_send_reply takes it's own rs ref, so this is safe here) */
+ * (ptlrpc_send_reply takes it's own rs ref, so this is safe here)
+ */
ptlrpc_req_drop_rs(reqcopy);
out_put:
}
/* Send early replies to everybody expiring within at_early_margin
- asking for at_extra time */
+ * asking for at_extra time
+ */
static int ptlrpc_at_check_timed(struct ptlrpc_service_part *svcpt)
{
struct ptlrpc_at_array *array = &svcpt->scp_at_array;
}
/* We're close to a timeout, and we don't know how much longer the
- server will take. Send early replies to everyone expiring soon. */
+ * server will take. Send early replies to everyone expiring soon.
+ */
INIT_LIST_HEAD(&work_list);
deadline = -1;
div_u64_rem(array->paa_deadline, array->paa_size, &index);
first, at_extra, counter);
if (first < 0) {
/* We're already past request deadlines before we even get a
- chance to send early replies */
+ * chance to send early replies
+ */
LCONSOLE_WARN("%s: This server is not able to keep up with request traffic (cpu-bound).\n",
svcpt->scp_service->srv_name);
CWARN("earlyQ=%d reqQ=%d recA=%d, svcEst=%d, delay=%ld(jiff)\n",
}
/* we took additional refcount so entries can't be deleted from list, no
- * locking is needed */
+ * locking is needed
+ */
while (!list_empty(&work_list)) {
rq = list_entry(work_list.next, struct ptlrpc_request,
rq_timed_list);
if (req->rq_export && req->rq_ops) {
/* Perform request specific check. We should do this check
* before the request is added into exp_hp_rpcs list otherwise
- * it may hit swab race at LU-1044. */
+ * it may hit swab race at LU-1044.
+ */
if (req->rq_ops->hpreq_check) {
rc = req->rq_ops->hpreq_check(req);
/**
{
if (req->rq_export && req->rq_ops) {
/* refresh lock timeout again so that client has more
- * room to send lock cancel RPC. */
+ * room to send lock cancel RPC.
+ */
if (req->rq_ops->hpreq_fini)
req->rq_ops->hpreq_fini(req);
list_del_init(&req->rq_list);
svcpt->scp_nreqs_incoming--;
/* Consider this still a "queued" request as far as stats are
- * concerned */
+ * concerned
+ */
spin_unlock(&svcpt->scp_lock);
/* go through security check/transform */
}
/* Discard requests queued for longer than the deadline.
- The deadline is increased if we send an early reply. */
+ * The deadline is increased if we send an early reply.
+ */
if (ktime_get_real_seconds() > request->rq_deadline) {
DEBUG_REQ(D_ERROR, request, "Dropping timed-out request from %s: deadline " CFS_DURATION_T ":" CFS_DURATION_T "s ago\n",
libcfs_id2str(request->rq_peer),
if (nlocks == 0 && !been_handled) {
/* If we see this, we should already have seen the warning
- * in mds_steal_ack_locks() */
+ * in mds_steal_ack_locks()
+ */
CDEBUG(D_HA, "All locks stolen from rs %p x%lld.t%lld o%d NID %s\n",
rs,
rs->rs_xid, rs->rs_transno, rs->rs_opc,
/* CAVEAT EMPTOR: We might be allocating buffers here because we've
* allowed the request history to grow out of control. We could put a
* sanity check on that here and cull some history if we need the
- * space. */
+ * space.
+ */
if (avail <= low_water)
ptlrpc_grow_req_bufs(svcpt, 1);
/* NB: we will call cfs_cpt_bind() for all threads, because we
* might want to run lustre server only on a subset of system CPUs,
- * in that case ->scp_cpt is CFS_CPT_ANY */
+ * in that case ->scp_cpt is CFS_CPT_ANY
+ */
rc = cfs_cpt_bind(svc->srv_cptable, svcpt->scp_cpt);
if (rc != 0) {
CWARN("%s: failed to bind %s on CPT %d\n",
/* SVC_STOPPING may already be set here if someone else is trying
* to stop the service while this new thread has been dynamically
* forked. We still set SVC_RUNNING to let our creator know that
- * we are now running, however we will exit as soon as possible */
+ * we are now running, however we will exit as soon as possible
+ */
thread_add_flags(thread, SVC_RUNNING);
svcpt->scp_nthrs_running++;
spin_unlock(&svcpt->scp_lock);
ptlrpc_server_post_idle_rqbds(svcpt) < 0) {
/* I just failed to repost request buffers.
* Wait for a timeout (unless something else
- * happens) before I try again */
+ * happens) before I try again
+ */
svcpt->scp_rqbd_timeout = cfs_time_seconds(1) / 10;
CDEBUG(D_RPCTRACE, "Posted buffers: %d\n",
svcpt->scp_nrqbds_posted);
if (svcpt->scp_nthrs_starting != 0) {
/* serialize starting because some modules (obdfilter)
- * might require unique and contiguous t_id */
+ * might require unique and contiguous t_id
+ */
LASSERT(svcpt->scp_nthrs_starting == 1);
spin_unlock(&svcpt->scp_lock);
kfree(thread);
int i;
/* All history will be culled when the next request buffer is
- * freed in ptlrpc_service_purge_all() */
+ * freed in ptlrpc_service_purge_all()
+ */
svc->srv_hist_nrqbds_cpt_max = 0;
rc = LNetClearLazyPortal(svc->srv_req_portal);
break;
/* Unlink all the request buffers. This forces a 'final'
- * event with its 'unlink' flag set for each posted rqbd */
+ * event with its 'unlink' flag set for each posted rqbd
+ */
list_for_each_entry(rqbd, &svcpt->scp_rqbd_posted,
rqbd_list) {
rc = LNetMDUnlink(rqbd->rqbd_md_h);
break;
/* Wait for the network to release any buffers
- * it's currently filling */
+ * it's currently filling
+ */
spin_lock(&svcpt->scp_lock);
while (svcpt->scp_nrqbds_posted != 0) {
spin_unlock(&svcpt->scp_lock);
/* Network access will complete in finite time but
* the HUGE timeout lets us CWARN for visibility
- * of sluggish NALs */
+ * of sluggish LNDs
+ */
lwi = LWI_TIMEOUT_INTERVAL(
cfs_time_seconds(LONG_UNLINK),
cfs_time_seconds(1), NULL, NULL);
/* purge the request queue. NB No new replies (rqbds
* all unlinked) and no service threads, so I'm the only
- * thread noodling the request queue now */
+ * thread noodling the request queue now
+ */
while (!list_empty(&svcpt->scp_req_incoming)) {
req = list_entry(svcpt->scp_req_incoming.next,
struct ptlrpc_request, rq_list);
LASSERT(svcpt->scp_nreqs_incoming == 0);
LASSERT(svcpt->scp_nreqs_active == 0);
/* history should have been culled by
- * ptlrpc_server_finish_request */
+ * ptlrpc_server_finish_request
+ */
LASSERT(svcpt->scp_hist_nrqbds == 0);
/* Now free all the request buffers since nothing
- * references them any more... */
+ * references them any more...
+ */
while (!list_empty(&svcpt->scp_rqbd_idle)) {
rqbd = list_entry(svcpt->scp_rqbd_idle.next,
projects / openwrt / staging / blogic.git / commitdiff