2 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2017, Intel Corporation.
6 * This file is part of Lustre, https://wiki.whamcloud.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
25 #include <linux/uaccess.h>
27 #include <libcfs/libcfs.h>
28 #include <lnet/lib-lnet.h>
30 /* This is really lnet_proc.c. You might need to update sanity test 215
31 * if any file format is changed. */
33 #define LNET_LOFFT_BITS (sizeof(loff_t) * 8)
35 * NB: max allowed LNET_CPT_BITS is 8 on 64-bit system and 2 on 32-bit system
37 #define LNET_PROC_CPT_BITS (LNET_CPT_BITS + 1)
38 /* change version, 16 bits or 8 bits */
39 #define LNET_PROC_VER_BITS MAX(((MIN(LNET_LOFFT_BITS, 64)) / 4), 8)
41 #define LNET_PROC_HASH_BITS LNET_PEER_HASH_BITS
43 * bits for peer hash offset
44 * NB: we don't use the highest bit of *ppos because it's signed
46 #define LNET_PROC_HOFF_BITS (LNET_LOFFT_BITS - \
47 LNET_PROC_CPT_BITS - \
48 LNET_PROC_VER_BITS - \
49 LNET_PROC_HASH_BITS - 1)
50 /* bits for hash index + position */
51 #define LNET_PROC_HPOS_BITS (LNET_PROC_HASH_BITS + LNET_PROC_HOFF_BITS)
52 /* bits for peer hash table + hash version */
53 #define LNET_PROC_VPOS_BITS (LNET_PROC_HPOS_BITS + LNET_PROC_VER_BITS)
55 #define LNET_PROC_CPT_MASK ((1ULL << LNET_PROC_CPT_BITS) - 1)
56 #define LNET_PROC_VER_MASK ((1ULL << LNET_PROC_VER_BITS) - 1)
57 #define LNET_PROC_HASH_MASK ((1ULL << LNET_PROC_HASH_BITS) - 1)
58 #define LNET_PROC_HOFF_MASK ((1ULL << LNET_PROC_HOFF_BITS) - 1)
60 #define LNET_PROC_CPT_GET(pos) \
61 (int)(((pos) >> LNET_PROC_VPOS_BITS) & LNET_PROC_CPT_MASK)
63 #define LNET_PROC_VER_GET(pos) \
64 (int)(((pos) >> LNET_PROC_HPOS_BITS) & LNET_PROC_VER_MASK)
66 #define LNET_PROC_HASH_GET(pos) \
67 (int)(((pos) >> LNET_PROC_HOFF_BITS) & LNET_PROC_HASH_MASK)
69 #define LNET_PROC_HOFF_GET(pos) \
70 (int)((pos) & LNET_PROC_HOFF_MASK)
72 #define LNET_PROC_POS_MAKE(cpt, ver, hash, off) \
73 (((((loff_t)(cpt)) & LNET_PROC_CPT_MASK) << LNET_PROC_VPOS_BITS) | \
74 ((((loff_t)(ver)) & LNET_PROC_VER_MASK) << LNET_PROC_HPOS_BITS) | \
75 ((((loff_t)(hash)) & LNET_PROC_HASH_MASK) << LNET_PROC_HOFF_BITS) | \
76 ((off) & LNET_PROC_HOFF_MASK))
78 #define LNET_PROC_VERSION(v) ((unsigned int)((v) & LNET_PROC_VER_MASK))
80 static int __proc_lnet_stats(void *data, int write,
81 loff_t pos, void __user *buffer, int nob)
84 struct lnet_counters *ctrs;
85 struct lnet_counters_common common;
88 const int tmpsiz = 256; /* 7 %u and 4 __u64 */
91 lnet_counters_reset();
97 LIBCFS_ALLOC(ctrs, sizeof(*ctrs));
101 LIBCFS_ALLOC(tmpstr, tmpsiz);
102 if (tmpstr == NULL) {
103 LIBCFS_FREE(ctrs, sizeof(*ctrs));
107 lnet_counters_get(ctrs);
108 common = ctrs->lct_common;
110 len = snprintf(tmpstr, tmpsiz,
111 "%u %u %u %u %u %u %u %llu %llu "
113 common.lcc_msgs_alloc, common.lcc_msgs_max,
115 common.lcc_send_count, common.lcc_recv_count,
116 common.lcc_route_count, common.lcc_drop_count,
117 common.lcc_send_length, common.lcc_recv_length,
118 common.lcc_route_length, common.lcc_drop_length);
120 if (pos >= min_t(int, len, strlen(tmpstr)))
123 rc = cfs_trace_copyout_string(buffer, nob,
126 LIBCFS_FREE(tmpstr, tmpsiz);
127 LIBCFS_FREE(ctrs, sizeof(*ctrs));
132 proc_lnet_stats(struct ctl_table *table, int write, void __user *buffer,
133 size_t *lenp, loff_t *ppos)
135 return lprocfs_call_handler(table->data, write, ppos, buffer, lenp,
140 proc_lnet_routes(struct ctl_table *table, int write, void __user *buffer,
141 size_t *lenp, loff_t *ppos)
143 const int tmpsiz = 256;
151 CLASSERT(sizeof(loff_t) >= 4);
153 off = LNET_PROC_HOFF_GET(*ppos);
154 ver = LNET_PROC_VER_GET(*ppos);
161 LIBCFS_ALLOC(tmpstr, tmpsiz);
165 s = tmpstr; /* points to current position in tmpstr[] */
168 s += snprintf(s, tmpstr + tmpsiz - s, "Routing %s\n",
169 the_lnet.ln_routing ? "enabled" : "disabled");
170 LASSERT(tmpstr + tmpsiz - s > 0);
172 s += snprintf(s, tmpstr + tmpsiz - s, "%-8s %4s %8s %7s %s\n",
173 "net", "hops", "priority", "state", "router");
174 LASSERT(tmpstr + tmpsiz - s > 0);
177 ver = (unsigned int)the_lnet.ln_remote_nets_version;
179 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
183 struct lnet_route *route = NULL;
184 struct lnet_remotenet *rnet = NULL;
186 struct list_head *rn_list;
191 if (ver != LNET_PROC_VERSION(the_lnet.ln_remote_nets_version)) {
193 LIBCFS_FREE(tmpstr, tmpsiz);
197 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE && route == NULL;
199 rn_list = &the_lnet.ln_remote_nets_hash[i];
203 while (n != rn_list && route == NULL) {
204 rnet = list_entry(n, struct lnet_remotenet,
207 r = rnet->lrn_routes.next;
209 while (r != &rnet->lrn_routes) {
210 struct lnet_route *re =
211 list_entry(r, struct lnet_route,
227 __u32 net = rnet->lrn_net;
228 __u32 hops = route->lr_hops;
229 unsigned int priority = route->lr_priority;
230 lnet_nid_t nid = route->lr_gateway->lpni_nid;
231 int alive = lnet_is_route_alive(route);
233 s += snprintf(s, tmpstr + tmpsiz - s,
234 "%-8s %4d %8u %7s %s\n",
235 libcfs_net2str(net), hops,
237 alive ? "up" : "down",
238 libcfs_nid2str(nid));
239 LASSERT(tmpstr + tmpsiz - s > 0);
245 len = s - tmpstr; /* how many bytes was written */
247 if (len > *lenp) { /* linux-supplied buffer is too small */
249 } else if (len > 0) { /* wrote something */
250 if (copy_to_user(buffer, tmpstr, len))
254 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
258 LIBCFS_FREE(tmpstr, tmpsiz);
267 proc_lnet_routers(struct ctl_table *table, int write, void __user *buffer,
268 size_t *lenp, loff_t *ppos)
273 const int tmpsiz = 256;
278 off = LNET_PROC_HOFF_GET(*ppos);
279 ver = LNET_PROC_VER_GET(*ppos);
286 LIBCFS_ALLOC(tmpstr, tmpsiz);
290 s = tmpstr; /* points to current position in tmpstr[] */
293 s += snprintf(s, tmpstr + tmpsiz - s,
294 "%-4s %7s %9s %6s %12s %9s %8s %7s %s\n",
295 "ref", "rtr_ref", "alive_cnt", "state",
296 "last_ping", "ping_sent", "deadline",
297 "down_ni", "router");
298 LASSERT(tmpstr + tmpsiz - s > 0);
301 ver = (unsigned int)the_lnet.ln_routers_version;
303 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
306 struct lnet_peer_ni *peer = NULL;
311 if (ver != LNET_PROC_VERSION(the_lnet.ln_routers_version)) {
314 LIBCFS_FREE(tmpstr, tmpsiz);
318 r = the_lnet.ln_routers.next;
320 while (r != &the_lnet.ln_routers) {
321 struct lnet_peer_ni *lp =
322 list_entry(r, struct lnet_peer_ni,
335 lnet_nid_t nid = peer->lpni_nid;
336 time64_t now = ktime_get_seconds();
337 time64_t deadline = peer->lpni_ping_deadline;
338 int nrefs = atomic_read(&peer->lpni_refcount);
339 int nrtrrefs = peer->lpni_rtr_refcount;
340 int alive_cnt = peer->lpni_alive_count;
341 int alive = peer->lpni_alive;
342 int pingsent = !peer->lpni_ping_notsent;
343 time64_t last_ping = now - peer->lpni_ping_timestamp;
345 struct lnet_route *rtr;
347 if ((peer->lpni_ping_feats &
348 LNET_PING_FEAT_NI_STATUS) != 0) {
349 list_for_each_entry(rtr, &peer->lpni_routes,
351 /* downis on any route should be the
352 * number of downis on the gateway */
353 if (rtr->lr_downis != 0) {
354 down_ni = rtr->lr_downis;
361 s += snprintf(s, tmpstr + tmpsiz - s,
362 "%-4d %7d %9d %6s %12llu %9d %8s %7d %s\n",
363 nrefs, nrtrrefs, alive_cnt,
364 alive ? "up" : "down", last_ping,
365 pingsent, "NA", down_ni,
366 libcfs_nid2str(nid));
368 s += snprintf(s, tmpstr + tmpsiz - s,
369 "%-4d %7d %9d %6s %12llu %9d %8llu %7d %s\n",
370 nrefs, nrtrrefs, alive_cnt,
371 alive ? "up" : "down", last_ping,
374 down_ni, libcfs_nid2str(nid));
375 LASSERT(tmpstr + tmpsiz - s > 0);
381 len = s - tmpstr; /* how many bytes was written */
383 if (len > *lenp) { /* linux-supplied buffer is too small */
385 } else if (len > 0) { /* wrote something */
386 if (copy_to_user(buffer, tmpstr, len))
390 *ppos = LNET_PROC_POS_MAKE(0, ver, 0, off);
394 LIBCFS_FREE(tmpstr, tmpsiz);
402 /* TODO: there should be no direct access to ptable. We should add a set
403 * of APIs that give access to the ptable and its members */
405 proc_lnet_peers(struct ctl_table *table, int write, void __user *buffer,
406 size_t *lenp, loff_t *ppos)
408 const int tmpsiz = 256;
409 struct lnet_peer_table *ptable;
412 int cpt = LNET_PROC_CPT_GET(*ppos);
413 int ver = LNET_PROC_VER_GET(*ppos);
414 int hash = LNET_PROC_HASH_GET(*ppos);
415 int hoff = LNET_PROC_HOFF_GET(*ppos);
421 struct lnet_peer_ni *peer;
423 cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
425 for (hash = 0; hash < LNET_PEER_HASH_SIZE; hash++) {
426 list_for_each_entry(peer,
427 &ptable->pt_hash[hash],
429 peer->lpni_mintxcredits =
430 peer->lpni_txcredits;
431 peer->lpni_minrtrcredits =
432 peer->lpni_rtrcredits;
444 CLASSERT(LNET_PROC_HASH_BITS >= LNET_PEER_HASH_BITS);
446 if (cpt >= LNET_CPT_NUMBER) {
451 LIBCFS_ALLOC(tmpstr, tmpsiz);
455 s = tmpstr; /* points to current position in tmpstr[] */
458 s += snprintf(s, tmpstr + tmpsiz - s,
459 "%-24s %4s %5s %5s %5s %5s %5s %5s %5s %s\n",
460 "nid", "refs", "state", "last", "max",
461 "rtr", "min", "tx", "min", "queue");
462 LASSERT(tmpstr + tmpsiz - s > 0);
466 struct lnet_peer_ni *peer;
476 ptable = the_lnet.ln_peer_tables[cpt];
478 ver = LNET_PROC_VERSION(ptable->pt_version);
480 if (ver != LNET_PROC_VERSION(ptable->pt_version)) {
481 lnet_net_unlock(cpt);
482 LIBCFS_FREE(tmpstr, tmpsiz);
486 while (hash < LNET_PEER_HASH_SIZE) {
488 p = ptable->pt_hash[hash].next;
490 while (p != &ptable->pt_hash[hash]) {
491 struct lnet_peer_ni *lp =
492 list_entry(p, struct lnet_peer_ni,
497 /* minor optimization: start from idx+1
498 * on next iteration if we've just
499 * drained lpni_hashlist */
500 if (lp->lpni_hashlist.next ==
501 &ptable->pt_hash[hash]) {
512 p = lp->lpni_hashlist.next;
524 lnet_nid_t nid = peer->lpni_nid;
525 int nrefs = atomic_read(&peer->lpni_refcount);
526 time64_t lastalive = -1;
527 char *aliveness = "NA";
528 int maxcr = (peer->lpni_net) ?
529 peer->lpni_net->net_tunables.lct_peer_tx_credits : 0;
530 int txcr = peer->lpni_txcredits;
531 int mintxcr = peer->lpni_mintxcredits;
532 int rtrcr = peer->lpni_rtrcredits;
533 int minrtrcr = peer->lpni_minrtrcredits;
534 int txqnob = peer->lpni_txqnob;
536 if (lnet_isrouter(peer) ||
537 lnet_peer_aliveness_enabled(peer))
538 aliveness = peer->lpni_alive ? "up" : "down";
540 if (lnet_peer_aliveness_enabled(peer)) {
541 time64_t now = ktime_get_seconds();
543 lastalive = now - peer->lpni_last_alive;
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 %5lld %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();
733 time64_t last_alive = -1;
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 %5lld %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 void lnet_router_debugfs_init(void)
960 lnet_insert_debugfs(lnet_table, NULL);