Whamcloud - gitweb
LU-3679 lnet: reflect down routes in /proc/sys/lnet/routes
[fs/lustre-release.git] / lnet / lnet / router_proc.c
index 5f85792..5bbcbb7 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  *
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
  *
  *   This file is part of Portals
  *   http://sourceforge.net/projects/sandiaportals/
@@ -30,7 +30,7 @@
 /* This is really lnet_proc.c. You might need to update sanity test 215
  * if any file format is changed. */
 
-static cfs_sysctl_table_header_t *lnet_table_header = NULL;
+static struct ctl_table_header *lnet_table_header = NULL;
 
 #ifndef HAVE_SYSCTL_UNNUMBERED
 #define CTL_LNET         (0x100)
@@ -41,6 +41,7 @@ enum {
         PSDEV_LNET_PEERS,
         PSDEV_LNET_BUFFERS,
         PSDEV_LNET_NIS,
+       PSDEV_LNET_PTL_ROTOR,
 };
 #else
 #define CTL_LNET           CTL_UNNUMBERED
@@ -50,13 +51,14 @@ enum {
 #define PSDEV_LNET_PEERS   CTL_UNNUMBERED
 #define PSDEV_LNET_BUFFERS CTL_UNNUMBERED
 #define PSDEV_LNET_NIS     CTL_UNNUMBERED
+#define PSDEV_LNET_PTL_ROTOR   CTL_UNNUMBERED
 #endif
 
 #define LNET_LOFFT_BITS                (sizeof(loff_t) * 8)
 /*
  * NB: max allowed LNET_CPT_BITS is 8 on 64-bit system and 2 on 32-bit system
  */
-#define LNET_PROC_CPT_BITS     LNET_CPT_BITS
+#define LNET_PROC_CPT_BITS     (LNET_CPT_BITS + 1)
 /* change version, 16 bits or 8 bits */
 #define LNET_PROC_VER_BITS     MAX(((MIN(LNET_LOFFT_BITS, 64)) / 4), 8)
 
@@ -160,8 +162,6 @@ int LL_PROC_PROTO(proc_lnet_routes)
        int             ver;
        int             off;
 
-       DECLARE_LL_PROC_PPOS_DECL;
-
        CLASSERT(sizeof(loff_t) >= 4);
 
        off = LNET_PROC_HOFF_GET(*ppos);
@@ -183,8 +183,8 @@ int LL_PROC_PROTO(proc_lnet_routes)
                               the_lnet.ln_routing ? "enabled" : "disabled");
                 LASSERT (tmpstr + tmpsiz - s > 0);
 
-                s += snprintf(s, tmpstr + tmpsiz - s, "%-8s %4s %7s %s\n",
-                              "net", "hops", "state", "router");
+               s += snprintf(s, tmpstr + tmpsiz - s, "%-8s %4s %8s %7s %s\n",
+                             "net", "hops", "priority", "state", "router");
                 LASSERT (tmpstr + tmpsiz - s > 0);
 
                lnet_net_lock(0);
@@ -197,47 +197,57 @@ int LL_PROC_PROTO(proc_lnet_routes)
                lnet_route_t            *route = NULL;
                lnet_remotenet_t        *rnet  = NULL;
                int                     skip  = off - 1;
+               cfs_list_t              *rn_list;
+               int                     i;
 
                lnet_net_lock(0);
 
                if (ver != LNET_PROC_VERSION(the_lnet.ln_remote_nets_version)) {
                        lnet_net_unlock(0);
-                        LIBCFS_FREE(tmpstr, tmpsiz);
-                        return -ESTALE;
-                }
+                       LIBCFS_FREE(tmpstr, tmpsiz);
+                       return -ESTALE;
+               }
 
-                n = the_lnet.ln_remote_nets.next;
+               for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE && route == NULL;
+                    i++) {
+                       rn_list = &the_lnet.ln_remote_nets_hash[i];
 
-                while (n != &the_lnet.ln_remote_nets && route == NULL) {
-                        rnet = cfs_list_entry(n, lnet_remotenet_t, lrn_list);
+                       n = rn_list->next;
 
-                        r = rnet->lrn_routes.next;
+                       while (n != rn_list && route == NULL) {
+                               rnet = cfs_list_entry(n, lnet_remotenet_t,
+                                                     lrn_list);
 
-                        while (r != &rnet->lrn_routes) {
-                                lnet_route_t *re =
-                                        cfs_list_entry(r, lnet_route_t,
-                                                       lr_list);
-                                if (skip == 0) {
-                                        route = re;
-                                        break;
-                                }
+                               r = rnet->lrn_routes.next;
 
-                                skip--;
-                                r = r->next;
-                        }
+                               while (r != &rnet->lrn_routes) {
+                                       lnet_route_t *re =
+                                               cfs_list_entry(r, lnet_route_t,
+                                                              lr_list);
+                                       if (skip == 0) {
+                                               route = re;
+                                               break;
+                                       }
 
-                        n = n->next;
-                }
+                                       skip--;
+                                       r = r->next;
+                               }
+
+                               n = n->next;
+                       }
+               }
 
                 if (route != NULL) {
-                        __u32        net   = rnet->lrn_net;
-                        unsigned int hops  = route->lr_hops;
-                        lnet_nid_t   nid   = route->lr_gateway->lp_nid;
-                        int          alive = route->lr_gateway->lp_alive;
+                       __u32        net        = rnet->lrn_net;
+                       unsigned int hops       = route->lr_hops;
+                       unsigned int priority   = route->lr_priority;
+                       lnet_nid_t   nid        = route->lr_gateway->lp_nid;
+                       int          alive      = lnet_is_route_alive(route);
 
                        s += snprintf(s, tmpstr + tmpsiz - s,
-                                     "%-8s %4u %7s %s\n",
+                                     "%-8s %4u %8u %7s %s\n",
                                      libcfs_net2str(net), hops,
+                                     priority,
                                      alive ? "up" : "down",
                                      libcfs_nid2str(nid));
                        LASSERT(tmpstr + tmpsiz - s > 0);
@@ -251,7 +261,7 @@ int LL_PROC_PROTO(proc_lnet_routes)
         if (len > *lenp) {    /* linux-supplied buffer is too small */
                 rc = -EINVAL;
         } else if (len > 0) { /* wrote something */
-                if (cfs_copy_to_user(buffer, tmpstr, len))
+               if (copy_to_user(buffer, tmpstr, len))
                         rc = -EFAULT;
                 else {
                        off += 1;
@@ -277,8 +287,6 @@ int LL_PROC_PROTO(proc_lnet_routers)
         int        ver;
        int        off;
 
-       DECLARE_LL_PROC_PPOS_DECL;
-
        off = LNET_PROC_HOFF_GET(*ppos);
        ver = LNET_PROC_VER_GET(*ppos);
 
@@ -348,7 +356,8 @@ int LL_PROC_PROTO(proc_lnet_routers)
                        int down_ni   = 0;
                        lnet_route_t *rtr;
 
-                       if (peer->lp_ping_version == LNET_PROTO_PING_VERSION) {
+                       if ((peer->lp_ping_feats &
+                            LNET_PING_FEAT_NI_STATUS) != 0) {
                                cfs_list_for_each_entry(rtr, &peer->lp_routes,
                                                        lr_gwlist) {
                                        /* downis on any route should be the
@@ -386,7 +395,7 @@ int LL_PROC_PROTO(proc_lnet_routers)
         if (len > *lenp) {    /* linux-supplied buffer is too small */
                 rc = -EINVAL;
         } else if (len > 0) { /* wrote something */
-                if (cfs_copy_to_user(buffer, tmpstr, len))
+               if (copy_to_user(buffer, tmpstr, len))
                         rc = -EFAULT;
                 else {
                        off += 1;
@@ -410,8 +419,8 @@ int LL_PROC_PROTO(proc_lnet_peers)
        char                    *s;
        int                     cpt  = LNET_PROC_CPT_GET(*ppos);
        int                     ver  = LNET_PROC_VER_GET(*ppos);
-       int                     hoff = LNET_PROC_HOFF_GET(*ppos);
        int                     hash = LNET_PROC_HASH_GET(*ppos);
+       int                     hoff = LNET_PROC_HOFF_GET(*ppos);
        int                     rc = 0;
        int                     len;
 
@@ -421,8 +430,10 @@ int LL_PROC_PROTO(proc_lnet_peers)
        if (*lenp == 0)
                return 0;
 
-       if (cpt >= LNET_CPT_NUMBER)
+       if (cpt >= LNET_CPT_NUMBER) {
+               *lenp = 0;
                return 0;
+       }
 
         LIBCFS_ALLOC(tmpstr, tmpsiz);
         if (tmpstr == NULL)
@@ -439,11 +450,14 @@ int LL_PROC_PROTO(proc_lnet_peers)
 
                hoff++;
        } else {
-               struct lnet_peer        *peer   = NULL;
-               cfs_list_t              *p      = NULL;
-               int                     skip    = hoff - 1;
-
+               struct lnet_peer        *peer;
+               cfs_list_t              *p;
+               int                     skip;
  again:
+               p = NULL;
+               peer = NULL;
+               skip = hoff - 1;
+
                lnet_net_lock(cpt);
                ptable = the_lnet.ln_peer_tables[cpt];
                if (hoff == 1)
@@ -533,14 +547,14 @@ int LL_PROC_PROTO(proc_lnet_peers)
 
                } else { /* peer is NULL */
                        lnet_net_unlock(cpt);
+               }
 
-                       if (hash == LNET_PEER_HASH_SIZE &&
-                           cpt < LNET_CPT_NUMBER - 1) {
-                               cpt++;
-                               hash = 0;
-                               hoff = 1;
+               if (hash == LNET_PEER_HASH_SIZE) {
+                       cpt++;
+                       hash = 0;
+                       hoff = 1;
+                       if (peer == NULL && cpt < LNET_CPT_NUMBER)
                                goto again;
-                       }
                }
         }
 
@@ -549,7 +563,7 @@ int LL_PROC_PROTO(proc_lnet_peers)
         if (len > *lenp) {    /* linux-supplied buffer is too small */
                 rc = -EINVAL;
         } else if (len > 0) { /* wrote something */
-                if (cfs_copy_to_user(buffer, tmpstr, len))
+               if (copy_to_user(buffer, tmpstr, len))
                         rc = -EFAULT;
                 else
                        *ppos = LNET_PROC_POS_MAKE(cpt, ver, hash, hoff);
@@ -631,8 +645,6 @@ int LL_PROC_PROTO(proc_lnet_nis)
         char      *s;
         int        len;
 
-        DECLARE_LL_PROC_PPOS_DECL;
-
         LASSERT (!write);
 
         if (*lenp == 0)
@@ -672,11 +684,12 @@ int LL_PROC_PROTO(proc_lnet_nis)
                 }
 
                 if (ni != NULL) {
-                       char    *stat;
                        struct lnet_tx_queue    *tq;
+                       char    *stat;
                        long    now = cfs_time_current_sec();
                        int     last_alive = -1;
                        int     i;
+                       int     j;
 
                        if (the_lnet.ln_routing)
                                last_alive = now - ni->ni_last_alive;
@@ -694,6 +707,15 @@ int LL_PROC_PROTO(proc_lnet_nis)
                        /* we actually output credits information for
                         * TX queue of each partition */
                        cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
+                               for (j = 0; ni->ni_cpts != NULL &&
+                                    j < ni->ni_ncpts; j++) {
+                                       if (i == ni->ni_cpts[j])
+                                               break;
+                               }
+
+                               if (j == ni->ni_ncpts)
+                                       continue;
+
                                if (i != 0)
                                        lnet_net_lock(i);
 
@@ -719,7 +741,7 @@ int LL_PROC_PROTO(proc_lnet_nis)
         if (len > *lenp) {    /* linux-supplied buffer is too small */
                 rc = -EINVAL;
         } else if (len > 0) { /* wrote something */
-                if (cfs_copy_to_user(buffer, tmpstr, len))
+               if (copy_to_user(buffer, tmpstr, len))
                         rc = -EFAULT;
                 else
                         *ppos += 1;
@@ -733,7 +755,108 @@ int LL_PROC_PROTO(proc_lnet_nis)
         return rc;
 }
 
-static cfs_sysctl_table_t lnet_table[] = {
+struct lnet_portal_rotors {
+       int             pr_value;
+       const char      *pr_name;
+       const char      *pr_desc;
+};
+
+static struct lnet_portal_rotors       portal_rotors[] = {
+       {
+               .pr_value = LNET_PTL_ROTOR_OFF,
+               .pr_name  = "OFF",
+               .pr_desc  = "Turn off message rotor for wildcard portals"
+       },
+       {
+               .pr_value = LNET_PTL_ROTOR_ON,
+               .pr_name  = "ON",
+               .pr_desc  = "round-robin dispatch all PUT messages for "
+                           "wildcard portals"
+       },
+       {
+               .pr_value = LNET_PTL_ROTOR_RR_RT,
+               .pr_name  = "RR_RT",
+               .pr_desc  = "round-robin dispatch routed PUT message for "
+                           "wildcard portals"
+       },
+       {
+               .pr_value = LNET_PTL_ROTOR_HASH_RT,
+               .pr_name  = "HASH_RT",
+               .pr_desc  = "dispatch routed PUT message by hashing source "
+                           "NID for wildcard portals"
+       },
+       {
+               .pr_value = -1,
+               .pr_name  = NULL,
+               .pr_desc  = NULL
+       },
+};
+
+extern int portal_rotor;
+
+static int __proc_lnet_portal_rotor(void *data, int write,
+                                   loff_t pos, void *buffer, int nob)
+{
+       const int       buf_len = 128;
+       char            *buf;
+       char            *tmp;
+       int             rc;
+       int             i;
+
+       LIBCFS_ALLOC(buf, buf_len);
+       if (buf == NULL)
+               return -ENOMEM;
+
+       if (!write) {
+               lnet_res_lock(0);
+
+               for (i = 0; portal_rotors[i].pr_value >= 0; i++) {
+                       if (portal_rotors[i].pr_value == portal_rotor)
+                               break;
+               }
+
+               LASSERT(portal_rotors[i].pr_value == portal_rotor);
+               lnet_res_unlock(0);
+
+               rc = snprintf(buf, buf_len,
+                             "{\n\tportals: all\n"
+                             "\trotor: %s\n\tdescription: %s\n}",
+                             portal_rotors[i].pr_name,
+                             portal_rotors[i].pr_desc);
+
+               if (pos >= min_t(int, rc, buf_len)) {
+                       rc = 0;
+               } else {
+                       rc = cfs_trace_copyout_string(buffer, nob,
+                                       buf + pos, "\n");
+               }
+               goto out;
+       }
+
+       rc = cfs_trace_copyin_string(buf, buf_len, buffer, nob);
+       if (rc < 0)
+               goto out;
+
+       tmp = cfs_trimwhite(buf);
+
+       rc = -EINVAL;
+       lnet_res_lock(0);
+       for (i = 0; portal_rotors[i].pr_name != NULL; i++) {
+               if (cfs_strncasecmp(portal_rotors[i].pr_name, tmp,
+                                   strlen(portal_rotors[i].pr_name)) == 0) {
+                       portal_rotor = portal_rotors[i].pr_value;
+                       rc = 0;
+                       break;
+               }
+       }
+       lnet_res_unlock(0);
+out:
+       LIBCFS_FREE(buf, buf_len);
+       return rc;
+}
+DECLARE_PROC_HANDLER(proc_lnet_portal_rotor);
+
+static struct ctl_table lnet_table[] = {
         /*
          * NB No .strategy entries have been provided since sysctl(8) prefers
          * to go via /proc for portability.
@@ -775,11 +898,17 @@ static cfs_sysctl_table_t lnet_table[] = {
                 .proc_handler = &proc_lnet_nis,
         },
         {
-                INIT_CTL_NAME(0)
-        }
+               INIT_CTL_NAME(PSDEV_LNET_PTL_ROTOR)
+               .procname = "portal_rotor",
+               .mode     = 0644,
+               .proc_handler = &proc_lnet_portal_rotor,
+       },
+       {
+               INIT_CTL_NAME(0)
+       }
 };
 
-static cfs_sysctl_table_t top_table[] = {
+static struct ctl_table top_table[] = {
         {
                 INIT_CTL_NAME(CTL_LNET)
                 .procname = "lnet",
@@ -797,8 +926,8 @@ void
 lnet_proc_init(void)
 {
 #ifdef CONFIG_SYSCTL
-        if (lnet_table_header == NULL)
-                lnet_table_header = cfs_register_sysctl_table(top_table, 0);
+       if (lnet_table_header == NULL)
+               lnet_table_header = register_sysctl_table(top_table);
 #endif
 }
 
@@ -806,10 +935,10 @@ void
 lnet_proc_fini(void)
 {
 #ifdef CONFIG_SYSCTL
-        if (lnet_table_header != NULL)
-                cfs_unregister_sysctl_table(lnet_table_header);
+       if (lnet_table_header != NULL)
+               unregister_sysctl_table(lnet_table_header);
 
-        lnet_table_header = NULL;
+       lnet_table_header = NULL;
 #endif
 }