Whamcloud - gitweb
LU-12691 ldlm: obd_max_recoverable_clients is not atomic
[fs/lustre-release.git] / lnet / lnet / net_fault.c
index 04c98d5..4f1bc30 100644 (file)
@@ -36,6 +36,7 @@
 
 #define DEBUG_SUBSYSTEM S_LNET
 
+#include <linux/random.h>
 #include <lnet/lib-lnet.h>
 #include <uapi/linux/lnet/lnetctl.h>
 
@@ -79,10 +80,12 @@ lnet_fault_nid_match(lnet_nid_t nid, lnet_nid_t msg_nid)
 
 static bool
 lnet_fault_attr_match(struct lnet_fault_attr *attr, lnet_nid_t src,
-                     lnet_nid_t dst, unsigned int type, unsigned int portal)
+                     lnet_nid_t local_nid, lnet_nid_t dst,
+                     unsigned int type, unsigned int portal)
 {
        if (!lnet_fault_nid_match(attr->fa_src, src) ||
-           !lnet_fault_nid_match(attr->fa_dst, dst))
+           !lnet_fault_nid_match(attr->fa_dst, dst) ||
+           !lnet_fault_nid_match(attr->fa_local_nid, local_nid))
                return false;
 
        if (!(attr->fa_msg_mask & (1 << type)))
@@ -172,9 +175,9 @@ lnet_drop_rule_add(struct lnet_fault_attr *attr)
        if (attr->u.drop.da_interval != 0) {
                rule->dr_time_base = ktime_get_seconds() + attr->u.drop.da_interval;
                rule->dr_drop_time = ktime_get_seconds() +
-                                    cfs_rand() % attr->u.drop.da_interval;
+                                    prandom_u32_max(attr->u.drop.da_interval);
        } else {
-               rule->dr_drop_at = cfs_rand() % attr->u.drop.da_rate;
+               rule->dr_drop_at = prandom_u32_max(attr->u.drop.da_rate);
        }
 
        lnet_net_lock(LNET_LOCK_EX);
@@ -281,10 +284,10 @@ lnet_drop_rule_reset(void)
 
                memset(&rule->dr_stat, 0, sizeof(rule->dr_stat));
                if (attr->u.drop.da_rate != 0) {
-                       rule->dr_drop_at = cfs_rand() % attr->u.drop.da_rate;
+                       rule->dr_drop_at = prandom_u32_max(attr->u.drop.da_rate);
                } else {
                        rule->dr_drop_time = ktime_get_seconds() +
-                                            cfs_rand() % attr->u.drop.da_interval;
+                                            prandom_u32_max(attr->u.drop.da_interval);
                        rule->dr_time_base = ktime_get_seconds() + attr->u.drop.da_interval;
                }
                spin_unlock(&rule->dr_lock);
@@ -297,15 +300,13 @@ lnet_drop_rule_reset(void)
 static void
 lnet_fault_match_health(enum lnet_msg_hstatus *hstatus, __u32 mask)
 {
-       unsigned int random;
        int choice;
        int delta;
        int best_delta;
        int i;
 
        /* assign a random failure */
-       random = cfs_rand();
-       choice = random % (LNET_MSG_STATUS_END - LNET_MSG_STATUS_OK);
+       choice = prandom_u32_max(LNET_MSG_STATUS_END - LNET_MSG_STATUS_OK);
        if (choice == 0)
                choice++;
 
@@ -344,15 +345,22 @@ lnet_fault_match_health(enum lnet_msg_hstatus *hstatus, __u32 mask)
  */
 static bool
 drop_rule_match(struct lnet_drop_rule *rule, lnet_nid_t src,
-               lnet_nid_t dst, unsigned int type, unsigned int portal,
+               lnet_nid_t local_nid, lnet_nid_t dst,
+               unsigned int type, unsigned int portal,
                enum lnet_msg_hstatus *hstatus)
 {
        struct lnet_fault_attr  *attr = &rule->dr_attr;
        bool                     drop;
 
-       if (!lnet_fault_attr_match(attr, src, dst, type, portal))
+       if (!lnet_fault_attr_match(attr, src, local_nid, dst, type, portal))
                return false;
 
+       if (attr->u.drop.da_drop_all) {
+               CDEBUG(D_NET, "set to drop all messages\n");
+               drop = true;
+               goto drop_matched;
+       }
+
        /*
         * if we're trying to match a health status error but it hasn't
         * been set in the rule, then don't match
@@ -364,7 +372,7 @@ drop_rule_match(struct lnet_drop_rule *rule, lnet_nid_t src,
        /* match this rule, check drop rate now */
        spin_lock(&rule->dr_lock);
        if (attr->u.drop.da_random) {
-               int value = cfs_rand() % attr->u.drop.da_interval;
+               int value = prandom_u32_max(attr->u.drop.da_interval);
                if (value >= (attr->u.drop.da_interval / 2))
                        drop = true;
                else
@@ -379,7 +387,7 @@ drop_rule_match(struct lnet_drop_rule *rule, lnet_nid_t src,
                                rule->dr_time_base = now;
 
                        rule->dr_drop_time = rule->dr_time_base +
-                                            cfs_rand() % attr->u.drop.da_interval;
+                                            prandom_u32_max(attr->u.drop.da_interval);
                        rule->dr_time_base += attr->u.drop.da_interval;
 
                        CDEBUG(D_NET, "Drop Rule %s->%s: next drop : %lld\n",
@@ -395,13 +403,15 @@ drop_rule_match(struct lnet_drop_rule *rule, lnet_nid_t src,
                count = rule->dr_stat.fs_count;
                if (do_div(count, attr->u.drop.da_rate) == 0) {
                        rule->dr_drop_at = rule->dr_stat.fs_count +
-                                          cfs_rand() % attr->u.drop.da_rate;
+                                          prandom_u32_max(attr->u.drop.da_rate);
                        CDEBUG(D_NET, "Drop Rule %s->%s: next drop: %lu\n",
                               libcfs_nid2str(attr->fa_src),
                               libcfs_nid2str(attr->fa_dst), rule->dr_drop_at);
                }
        }
 
+drop_matched:
+
        if (drop) { /* drop this message, update counters */
                if (hstatus)
                        lnet_fault_match_health(hstatus,
@@ -418,7 +428,9 @@ drop_rule_match(struct lnet_drop_rule *rule, lnet_nid_t src,
  * Check if message from \a src to \a dst can match any existed drop rule
  */
 bool
-lnet_drop_rule_match(struct lnet_hdr *hdr, enum lnet_msg_hstatus *hstatus)
+lnet_drop_rule_match(struct lnet_hdr *hdr,
+                    lnet_nid_t local_nid,
+                    enum lnet_msg_hstatus *hstatus)
 {
        lnet_nid_t src = le64_to_cpu(hdr->src_nid);
        lnet_nid_t dst = le64_to_cpu(hdr->dest_nid);
@@ -437,7 +449,7 @@ lnet_drop_rule_match(struct lnet_hdr *hdr, enum lnet_msg_hstatus *hstatus)
 
        cpt = lnet_net_lock_current();
        list_for_each_entry(rule, &the_lnet.ln_drop_rules, dr_link) {
-               drop = drop_rule_match(rule, src, dst, typ, ptl,
+               drop = drop_rule_match(rule, src, local_nid, dst, typ, ptl,
                                       hstatus);
                if (drop)
                        break;
@@ -528,7 +540,8 @@ delay_rule_match(struct lnet_delay_rule *rule, lnet_nid_t src,
        struct lnet_fault_attr  *attr = &rule->dl_attr;
        bool                     delay;
 
-       if (!lnet_fault_attr_match(attr, src, dst, type, portal))
+       if (!lnet_fault_attr_match(attr, src, LNET_NID_ANY,
+                                  dst, type, portal))
                return false;
 
        /* match this rule, check delay rate now */
@@ -543,7 +556,7 @@ delay_rule_match(struct lnet_delay_rule *rule, lnet_nid_t src,
                                rule->dl_time_base = now;
 
                        rule->dl_delay_time = rule->dl_time_base +
-                                             cfs_rand() % attr->u.delay.la_interval;
+                                             prandom_u32_max(attr->u.delay.la_interval);
                        rule->dl_time_base += attr->u.delay.la_interval;
 
                        CDEBUG(D_NET, "Delay Rule %s->%s: next delay : %lld\n",
@@ -560,7 +573,7 @@ delay_rule_match(struct lnet_delay_rule *rule, lnet_nid_t src,
                count = rule->dl_stat.fs_count;
                if (do_div(count, attr->u.delay.la_rate) == 0) {
                        rule->dl_delay_at = rule->dl_stat.fs_count +
-                                           cfs_rand() % attr->u.delay.la_rate;
+                                           prandom_u32_max(attr->u.delay.la_rate);
                        CDEBUG(D_NET, "Delay Rule %s->%s: next delay: %lu\n",
                               libcfs_nid2str(attr->fa_src),
                               libcfs_nid2str(attr->fa_dst), rule->dl_delay_at);
@@ -835,9 +848,9 @@ lnet_delay_rule_add(struct lnet_fault_attr *attr)
                rule->dl_time_base = ktime_get_seconds() +
                                     attr->u.delay.la_interval;
                rule->dl_delay_time = ktime_get_seconds() +
-                                     cfs_rand() % attr->u.delay.la_interval;
+                                     prandom_u32_max(attr->u.delay.la_interval);
        } else {
-               rule->dl_delay_at = cfs_rand() % attr->u.delay.la_rate;
+               rule->dl_delay_at = prandom_u32_max(attr->u.delay.la_rate);
        }
 
        rule->dl_msg_send = -1;
@@ -984,10 +997,10 @@ lnet_delay_rule_reset(void)
 
                memset(&rule->dl_stat, 0, sizeof(rule->dl_stat));
                if (attr->u.delay.la_rate != 0) {
-                       rule->dl_delay_at = cfs_rand() % attr->u.delay.la_rate;
+                       rule->dl_delay_at = prandom_u32_max(attr->u.delay.la_rate);
                } else {
                        rule->dl_delay_time = ktime_get_seconds() +
-                                             cfs_rand() % attr->u.delay.la_interval;
+                                             prandom_u32_max(attr->u.delay.la_interval);
                        rule->dl_time_base = ktime_get_seconds() +
                                             attr->u.delay.la_interval;
                }