2 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2015, Intel Corporation.
6 * This file is part of Lustre, https://wiki.hpdd.intel.com/
8 * Portals is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Portals is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Portals; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #define DEBUG_SUBSYSTEM S_LNET
24 #include <libcfs/libcfs.h>
25 #include <lnet/lib-lnet.h>
27 /* This is really lnet_proc.c. You might need to update sanity test 215
28 * if any file format is changed. */
30 static struct ctl_table_header *lnet_table_header = NULL;
32 #define LNET_LOFFT_BITS (sizeof(loff_t) * 8)
34 * NB: max allowed LNET_CPT_BITS is 8 on 64-bit system and 2 on 32-bit system
36 #define LNET_PROC_CPT_BITS (LNET_CPT_BITS + 1)
37 /* change version, 16 bits or 8 bits */
38 #define LNET_PROC_VER_BITS MAX(((MIN(LNET_LOFFT_BITS, 64)) / 4), 8)
40 #define LNET_PROC_HASH_BITS LNET_PEER_HASH_BITS
42 * bits for peer hash offset
43 * NB: we don't use the highest bit of *ppos because it's signed
45 #define LNET_PROC_HOFF_BITS (LNET_LOFFT_BITS - \
46 LNET_PROC_CPT_BITS - \
47 LNET_PROC_VER_BITS - \
48 LNET_PROC_HASH_BITS - 1)
49 /* bits for hash index + position */
50 #define LNET_PROC_HPOS_BITS (LNET_PROC_HASH_BITS + LNET_PROC_HOFF_BITS)
51 /* bits for peer hash table + hash version */
52 #define LNET_PROC_VPOS_BITS (LNET_PROC_HPOS_BITS + LNET_PROC_VER_BITS)
54 #define LNET_PROC_CPT_MASK ((1ULL << LNET_PROC_CPT_BITS) - 1)
55 #define LNET_PROC_VER_MASK ((1ULL << LNET_PROC_VER_BITS) - 1)
56 #define LNET_PROC_HASH_MASK ((1ULL << LNET_PROC_HASH_BITS) - 1)
57 #define LNET_PROC_HOFF_MASK ((1ULL << LNET_PROC_HOFF_BITS) - 1)
59 #define LNET_PROC_CPT_GET(pos) \
60 (int)(((pos) >> LNET_PROC_VPOS_BITS) & LNET_PROC_CPT_MASK)
62 #define LNET_PROC_VER_GET(pos) \
63 (int)(((pos) >> LNET_PROC_HPOS_BITS) & LNET_PROC_VER_MASK)
65 #define LNET_PROC_HASH_GET(pos) \
66 (int)(((pos) >> LNET_PROC_HOFF_BITS) & LNET_PROC_HASH_MASK)
68 #define LNET_PROC_HOFF_GET(pos) \
69 (int)((pos) & LNET_PROC_HOFF_MASK)
71 #define LNET_PROC_POS_MAKE(cpt, ver, hash, off) \
72 (((((loff_t)(cpt)) & LNET_PROC_CPT_MASK) << LNET_PROC_VPOS_BITS) | \
73 ((((loff_t)(ver)) & LNET_PROC_VER_MASK) << LNET_PROC_HPOS_BITS) | \
74 ((((loff_t)(hash)) & LNET_PROC_HASH_MASK) << LNET_PROC_HOFF_BITS) | \
75 ((off) & LNET_PROC_HOFF_MASK))
77 #define LNET_PROC_VERSION(v) ((unsigned int)((v) & LNET_PROC_VER_MASK))
79 static int __proc_lnet_stats(void *data, int write,
80 loff_t pos, void __user *buffer, int nob)
83 struct lnet_counters *ctrs;
86 const int tmpsiz = 256; /* 7 %u and 4 __u64 */
89 lnet_counters_reset();
95 LIBCFS_ALLOC(ctrs, sizeof(*ctrs));
99 LIBCFS_ALLOC(tmpstr, tmpsiz);
100 if (tmpstr == NULL) {
101 LIBCFS_FREE(ctrs, sizeof(*ctrs));
105 lnet_counters_get(ctrs);
107 len = snprintf(tmpstr, tmpsiz,
108 "%u %u %u %u %u %u %u %llu %llu "
110 ctrs->msgs_alloc, ctrs->msgs_max,
112 ctrs->send_count, ctrs->recv_count,
113 ctrs->route_count, ctrs->drop_count,
114 ctrs->send_length, ctrs->recv_length,
115 ctrs->route_length, ctrs->drop_length);
117 if (pos >= min_t(int, len, strlen(tmpstr)))
120 rc = cfs_trace_copyout_string(buffer, nob,
123 LIBCFS_FREE(tmpstr, tmpsiz);
124 LIBCFS_FREE(ctrs, sizeof(*ctrs));
129 proc_lnet_stats(struct ctl_table *table, int write, void __user *buffer,
130 size_t *lenp, loff_t *ppos)
132 return lprocfs_call_handler(table->data, write, ppos, buffer, lenp,
137 proc_lnet_routes(struct ctl_table *table, int write, void __user *buffer,
138 size_t *lenp, loff_t *ppos)
140 const int tmpsiz = 256;
148 CLASSERT(sizeof(loff_t) >= 4);
150 off = LNET_PROC_HOFF_GET(*ppos);
151 ver = LNET_PROC_VER_GET(*ppos);
158 LIBCFS_ALLOC(tmpstr, tmpsiz);
162 s = tmpstr; /* points to current position in tmpstr[] */
165 s += snprintf(s, tmpstr + tmpsiz - s, "Routing %s\n",
166 the_lnet.ln_routing ? "enabled" : "disabled");
167 LASSERT(tmpstr + tmpsiz - s > 0);
169 s += snprintf(s, tmpstr + tmpsiz - s, "%-8s %4s %8s %7s %s\n",
170 "net", "hops", "priority", "state", "router");
171 LASSERT(tmpstr + tmpsiz - s > 0);
174 ver = (unsigned int)the_lnet.ln_remote_nets_version;
176 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
180 struct lnet_route *route = NULL;
181 struct lnet_remotenet *rnet = NULL;
183 struct list_head *rn_list;
188 if (ver != LNET_PROC_VERSION(the_lnet.ln_remote_nets_version)) {
190 LIBCFS_FREE(tmpstr, tmpsiz);
194 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE && route == NULL;
196 rn_list = &the_lnet.ln_remote_nets_hash[i];
200 while (n != rn_list && route == NULL) {
201 rnet = list_entry(n, struct lnet_remotenet,
204 r = rnet->lrn_routes.next;
206 while (r != &rnet->lrn_routes) {
207 struct lnet_route *re =
208 list_entry(r, struct lnet_route,
224 __u32 net = rnet->lrn_net;
225 __u32 hops = route->lr_hops;
226 unsigned int priority = route->lr_priority;
227 lnet_nid_t nid = route->lr_gateway->lpni_nid;
228 int alive = lnet_is_route_alive(route);
230 s += snprintf(s, tmpstr + tmpsiz - s,
231 "%-8s %4d %8u %7s %s\n",
232 libcfs_net2str(net), hops,
234 alive ? "up" : "down",
235 libcfs_nid2str(nid));
236 LASSERT(tmpstr + tmpsiz - s > 0);
242 len = s - tmpstr; /* how many bytes was written */
244 if (len > *lenp) { /* linux-supplied buffer is too small */
246 } else if (len > 0) { /* wrote something */
247 if (copy_to_user(buffer, tmpstr, len))
251 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
255 LIBCFS_FREE(tmpstr, tmpsiz);
264 proc_lnet_routers(struct ctl_table *table, int write, void __user *buffer,
265 size_t *lenp, loff_t *ppos)
270 const int tmpsiz = 256;
275 off = LNET_PROC_HOFF_GET(*ppos);
276 ver = LNET_PROC_VER_GET(*ppos);
283 LIBCFS_ALLOC(tmpstr, tmpsiz);
287 s = tmpstr; /* points to current position in tmpstr[] */
290 s += snprintf(s, tmpstr + tmpsiz - s,
291 "%-4s %7s %9s %6s %12s %9s %8s %7s %s\n",
292 "ref", "rtr_ref", "alive_cnt", "state",
293 "last_ping", "ping_sent", "deadline",
294 "down_ni", "router");
295 LASSERT(tmpstr + tmpsiz - s > 0);
298 ver = (unsigned int)the_lnet.ln_routers_version;
300 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
303 struct lnet_peer_ni *peer = NULL;
308 if (ver != LNET_PROC_VERSION(the_lnet.ln_routers_version)) {
311 LIBCFS_FREE(tmpstr, tmpsiz);
315 r = the_lnet.ln_routers.next;
317 while (r != &the_lnet.ln_routers) {
318 struct lnet_peer_ni *lp =
319 list_entry(r, struct lnet_peer_ni,
332 lnet_nid_t nid = peer->lpni_nid;
333 cfs_time_t now = cfs_time_current();
334 cfs_time_t deadline = peer->lpni_ping_deadline;
335 int nrefs = atomic_read(&peer->lpni_refcount);
336 int nrtrrefs = peer->lpni_rtr_refcount;
337 int alive_cnt = peer->lpni_alive_count;
338 int alive = peer->lpni_alive;
339 int pingsent = !peer->lpni_ping_notsent;
340 int last_ping = cfs_duration_sec(cfs_time_sub(now,
341 peer->lpni_ping_timestamp));
343 struct lnet_route *rtr;
345 if ((peer->lpni_ping_feats &
346 LNET_PING_FEAT_NI_STATUS) != 0) {
347 list_for_each_entry(rtr, &peer->lpni_routes,
349 /* downis on any route should be the
350 * number of downis on the gateway */
351 if (rtr->lr_downis != 0) {
352 down_ni = rtr->lr_downis;
359 s += snprintf(s, tmpstr + tmpsiz - s,
360 "%-4d %7d %9d %6s %12d %9d %8s %7d %s\n",
361 nrefs, nrtrrefs, alive_cnt,
362 alive ? "up" : "down", last_ping,
363 pingsent, "NA", down_ni,
364 libcfs_nid2str(nid));
366 s += snprintf(s, tmpstr + tmpsiz - s,
367 "%-4d %7d %9d %6s %12d %9d %8lu %7d %s\n",
368 nrefs, nrtrrefs, alive_cnt,
369 alive ? "up" : "down", last_ping,
371 cfs_duration_sec(cfs_time_sub(deadline, now)),
372 down_ni, libcfs_nid2str(nid));
373 LASSERT(tmpstr + tmpsiz - s > 0);
379 len = s - tmpstr; /* how many bytes was written */
381 if (len > *lenp) { /* linux-supplied buffer is too small */
383 } else if (len > 0) { /* wrote something */
384 if (copy_to_user(buffer, tmpstr, len))
388 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
392 LIBCFS_FREE(tmpstr, tmpsiz);
400 /* TODO: there should be no direct access to ptable. We should add a set
401 * of APIs that give access to the ptable and its members */
403 proc_lnet_peers(struct ctl_table *table, int write, void __user *buffer,
404 size_t *lenp, loff_t *ppos)
406 const int tmpsiz = 256;
407 struct lnet_peer_table *ptable;
410 int cpt = LNET_PROC_CPT_GET(*ppos);
411 int ver = LNET_PROC_VER_GET(*ppos);
412 int hash = LNET_PROC_HASH_GET(*ppos);
413 int hoff = LNET_PROC_HOFF_GET(*ppos);
419 struct lnet_peer_ni *peer;
421 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
423 for (hash = 0; hash < LNET_PEER_HASH_SIZE; hash++) {
424 list_for_each_entry(peer,
425 &ptable->pt_hash[hash],
427 peer->lpni_mintxcredits =
428 peer->lpni_txcredits;
429 peer->lpni_minrtrcredits =
430 peer->lpni_rtrcredits;
442 CLASSERT(LNET_PROC_HASH_BITS >= LNET_PEER_HASH_BITS);
444 if (cpt >= LNET_CPT_NUMBER) {
449 LIBCFS_ALLOC(tmpstr, tmpsiz);
453 s = tmpstr; /* points to current position in tmpstr[] */
456 s += snprintf(s, tmpstr + tmpsiz - s,
457 "%-24s %4s %5s %5s %5s %5s %5s %5s %5s %s\n",
458 "nid", "refs", "state", "last", "max",
459 "rtr", "min", "tx", "min", "queue");
460 LASSERT(tmpstr + tmpsiz - s > 0);
464 struct lnet_peer_ni *peer;
474 ptable = the_lnet.ln_peer_tables[cpt];
476 ver = LNET_PROC_VERSION(ptable->pt_version);
478 if (ver != LNET_PROC_VERSION(ptable->pt_version)) {
479 lnet_net_unlock(cpt);
480 LIBCFS_FREE(tmpstr, tmpsiz);
484 while (hash < LNET_PEER_HASH_SIZE) {
486 p = ptable->pt_hash[hash].next;
488 while (p != &ptable->pt_hash[hash]) {
489 struct lnet_peer_ni *lp =
490 list_entry(p, struct lnet_peer_ni,
495 /* minor optimization: start from idx+1
496 * on next iteration if we've just
497 * drained lpni_hashlist */
498 if (lp->lpni_hashlist.next ==
499 &ptable->pt_hash[hash]) {
510 p = lp->lpni_hashlist.next;
522 lnet_nid_t nid = peer->lpni_nid;
523 int nrefs = atomic_read(&peer->lpni_refcount);
525 char *aliveness = "NA";
526 int maxcr = (peer->lpni_net) ?
527 peer->lpni_net->net_tunables.lct_peer_tx_credits : 0;
528 int txcr = peer->lpni_txcredits;
529 int mintxcr = peer->lpni_mintxcredits;
530 int rtrcr = peer->lpni_rtrcredits;
531 int minrtrcr = peer->lpni_minrtrcredits;
532 int txqnob = peer->lpni_txqnob;
534 if (lnet_isrouter(peer) ||
535 lnet_peer_aliveness_enabled(peer))
536 aliveness = peer->lpni_alive ? "up" : "down";
538 if (lnet_peer_aliveness_enabled(peer)) {
539 cfs_time_t now = cfs_time_current();
540 cfs_duration_t delta;
542 delta = cfs_time_sub(now, peer->lpni_last_alive);
543 lastalive = cfs_duration_sec(delta);
545 /* No need to mess up peers contents with
546 * arbitrarily long integers - it suffices to
547 * know that lastalive is more than 10000s old
549 if (lastalive >= 10000)
553 lnet_net_unlock(cpt);
555 s += snprintf(s, tmpstr + tmpsiz - s,
556 "%-24s %4d %5s %5d %5d %5d %5d %5d %5d %d\n",
557 libcfs_nid2str(nid), nrefs, aliveness,
558 lastalive, maxcr, rtrcr, minrtrcr, txcr,
560 LASSERT(tmpstr + tmpsiz - s > 0);
562 } else { /* peer is NULL */
563 lnet_net_unlock(cpt);
566 if (hash == LNET_PEER_HASH_SIZE) {
570 if (peer == NULL && cpt < LNET_CPT_NUMBER)
575 len = s - tmpstr; /* how many bytes was written */
577 if (len > *lenp) { /* linux-supplied buffer is too small */
579 } else if (len > 0) { /* wrote something */
580 if (copy_to_user(buffer, tmpstr, len))
583 *ppos = LNET_PROC_POS_MAKE(cpt, ver, hash, hoff);
586 LIBCFS_FREE(tmpstr, tmpsiz);
594 static int __proc_lnet_buffers(void *data, int write,
595 loff_t pos, void __user *buffer, int nob)
607 /* (4 %d) * 4 * LNET_CPT_NUMBER */
608 tmpsiz = 64 * (LNET_NRBPOOLS + 1) * LNET_CPT_NUMBER;
609 LIBCFS_ALLOC(tmpstr, tmpsiz);
613 s = tmpstr; /* points to current position in tmpstr[] */
615 s += snprintf(s, tmpstr + tmpsiz - s,
617 "pages", "count", "credits", "min");
618 LASSERT(tmpstr + tmpsiz - s > 0);
620 if (the_lnet.ln_rtrpools == NULL)
621 goto out; /* I'm not a router */
623 for (idx = 0; idx < LNET_NRBPOOLS; idx++) {
624 struct lnet_rtrbufpool *rbp;
626 lnet_net_lock(LNET_LOCK_EX);
627 cfs_percpt_for_each(rbp, i, the_lnet.ln_rtrpools) {
628 s += snprintf(s, tmpstr + tmpsiz - s,
631 rbp[idx].rbp_nbuffers,
632 rbp[idx].rbp_credits,
633 rbp[idx].rbp_mincredits);
634 LASSERT(tmpstr + tmpsiz - s > 0);
636 lnet_net_unlock(LNET_LOCK_EX);
642 if (pos >= min_t(int, len, strlen(tmpstr)))
645 rc = cfs_trace_copyout_string(buffer, nob,
648 LIBCFS_FREE(tmpstr, tmpsiz);
653 proc_lnet_buffers(struct ctl_table *table, int write, void __user *buffer,
654 size_t *lenp, loff_t *ppos)
656 return lprocfs_call_handler(table->data, write, ppos, buffer, lenp,
657 __proc_lnet_buffers);
661 proc_lnet_nis(struct ctl_table *table, int write, void __user *buffer,
662 size_t *lenp, loff_t *ppos)
664 int tmpsiz = 128 * LNET_CPT_NUMBER;
674 /* Just reset the min stat. */
676 struct lnet_net *net;
680 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
681 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
682 struct lnet_tx_queue *tq;
686 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
687 for (j = 0; ni->ni_cpts != NULL &&
688 j < ni->ni_ncpts; j++) {
689 if (i == ni->ni_cpts[j])
693 if (j == ni->ni_ncpts)
698 tq->tq_credits_min = tq->tq_credits;
709 LIBCFS_ALLOC(tmpstr, tmpsiz);
713 s = tmpstr; /* points to current position in tmpstr[] */
716 s += snprintf(s, tmpstr + tmpsiz - s,
717 "%-24s %6s %5s %4s %4s %4s %5s %5s %5s\n",
718 "nid", "status", "alive", "refs", "peer",
719 "rtr", "max", "tx", "min");
720 LASSERT (tmpstr + tmpsiz - s > 0);
722 struct lnet_ni *ni = NULL;
723 int skip = *ppos - 1;
727 ni = lnet_get_ni_idx_locked(skip);
730 struct lnet_tx_queue *tq;
732 time64_t now = ktime_get_real_seconds();
737 if (the_lnet.ln_routing)
738 last_alive = now - ni->ni_last_alive;
740 /* @lo forever alive */
741 if (ni->ni_net->net_lnd->lnd_type == LOLND)
745 LASSERT(ni->ni_status != NULL);
746 stat = (ni->ni_status->ns_status ==
747 LNET_NI_STATUS_UP) ? "up" : "down";
750 /* we actually output credits information for
751 * TX queue of each partition */
752 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
753 for (j = 0; ni->ni_cpts != NULL &&
754 j < ni->ni_ncpts; j++) {
755 if (i == ni->ni_cpts[j])
759 if (j == ni->ni_ncpts)
765 s += snprintf(s, tmpstr + tmpsiz - s,
766 "%-24s %6s %5d %4d %4d %4d %5d %5d %5d\n",
767 libcfs_nid2str(ni->ni_nid), stat,
768 last_alive, *ni->ni_refs[i],
769 ni->ni_net->net_tunables.lct_peer_tx_credits,
770 ni->ni_net->net_tunables.lct_peer_rtr_credits,
772 tq->tq_credits, tq->tq_credits_min);
776 LASSERT(tmpstr + tmpsiz - s > 0);
782 len = s - tmpstr; /* how many bytes was written */
784 if (len > *lenp) { /* linux-supplied buffer is too small */
786 } else if (len > 0) { /* wrote something */
787 if (copy_to_user(buffer, tmpstr, len))
793 LIBCFS_FREE(tmpstr, tmpsiz);
801 struct lnet_portal_rotors {
807 static struct lnet_portal_rotors portal_rotors[] = {
809 .pr_value = LNET_PTL_ROTOR_OFF,
811 .pr_desc = "Turn off message rotor for wildcard portals"
814 .pr_value = LNET_PTL_ROTOR_ON,
816 .pr_desc = "round-robin dispatch all PUT messages for "
820 .pr_value = LNET_PTL_ROTOR_RR_RT,
822 .pr_desc = "round-robin dispatch routed PUT message for "
826 .pr_value = LNET_PTL_ROTOR_HASH_RT,
827 .pr_name = "HASH_RT",
828 .pr_desc = "dispatch routed PUT message by hashing source "
829 "NID for wildcard portals"
838 static int __proc_lnet_portal_rotor(void *data, int write,
839 loff_t pos, void __user *buffer, int nob)
841 const int buf_len = 128;
847 LIBCFS_ALLOC(buf, buf_len);
854 for (i = 0; portal_rotors[i].pr_value >= 0; i++) {
855 if (portal_rotors[i].pr_value == portal_rotor)
859 LASSERT(portal_rotors[i].pr_value == portal_rotor);
862 rc = snprintf(buf, buf_len,
863 "{\n\tportals: all\n"
864 "\trotor: %s\n\tdescription: %s\n}",
865 portal_rotors[i].pr_name,
866 portal_rotors[i].pr_desc);
868 if (pos >= min_t(int, rc, buf_len)) {
871 rc = cfs_trace_copyout_string(buffer, nob,
877 rc = cfs_trace_copyin_string(buf, buf_len, buffer, nob);
881 tmp = cfs_trimwhite(buf);
885 for (i = 0; portal_rotors[i].pr_name != NULL; i++) {
886 if (strncasecmp(portal_rotors[i].pr_name, tmp,
887 strlen(portal_rotors[i].pr_name)) == 0) {
888 portal_rotor = portal_rotors[i].pr_value;
895 LIBCFS_FREE(buf, buf_len);
900 proc_lnet_portal_rotor(struct ctl_table *table, int write, void __user *buffer,
901 size_t *lenp, loff_t *ppos)
903 return lprocfs_call_handler(table->data, write, ppos, buffer, lenp,
904 __proc_lnet_portal_rotor);
908 static struct ctl_table lnet_table[] = {
910 * NB No .strategy entries have been provided since sysctl(8) prefers
911 * to go via /proc for portability.
917 .proc_handler = &proc_lnet_stats,
921 .procname = "routes",
923 .proc_handler = &proc_lnet_routes,
927 .procname = "routers",
929 .proc_handler = &proc_lnet_routers,
935 .proc_handler = &proc_lnet_peers,
939 .procname = "buffers",
941 .proc_handler = &proc_lnet_buffers,
947 .proc_handler = &proc_lnet_nis,
951 .procname = "portal_rotor",
953 .proc_handler = &proc_lnet_portal_rotor,
958 static struct ctl_table top_table[] = {
974 if (lnet_table_header == NULL)
975 lnet_table_header = register_sysctl_table(top_table);
983 if (lnet_table_header != NULL)
984 unregister_sysctl_table(lnet_table_header);
986 lnet_table_header = NULL;