LU-9120 lnet: health error simulation

[fs/lustre-release.git] / lustre / utils / portals.c

diff --git a/lustre/utils/portals.c b/lustre/utils/portals.c
index 4ebf306..9349b6b 100644 (file)
--- a/lustre/utils/portals.c
+++ b/lustre/utils/portals.c
@@ -35,13 +35,13 @@
 #include <time.h>
 #include <linux/types.h>
 
-#include <libcfs/util/string.h>
 #include <libcfs/util/ioctl.h>
 #include <linux/lnet/libcfs_debug.h>
 #include <linux/lnet/lnet-dlc.h>
 #include <linux/lnet/lnetctl.h>
 #include <linux/lnet/nidstr.h>
 #include <linux/lnet/socklnd.h>
+#include <lustre/lustreapi.h>
 
 unsigned int libcfs_debug;
 unsigned int libcfs_printk = D_CANTMASK;
@@ -177,7 +177,7 @@ ptl_ipaddr_2_str(__u32 ipaddr, char *str, size_t strsize, int lookup)
                 net_ip = htonl (ipaddr);
                 he = gethostbyaddr (&net_ip, sizeof (net_ip), AF_INET);
                 if (he != NULL) {
-                       strlcpy(str, he->h_name, strsize);
+                       snprintf(str, strsize, "%s", he->h_name);
                         return (str);
                 }
         }
@@ -299,8 +299,6 @@ int jt_ptl_network(int argc, char **argv)
        __u32 net = LNET_NIDNET(LNET_NID_ANY);
        int rc;
 
-       fprintf(stderr, "This command has been deprecated. Plesae use 'lnetctl net add'\n");
-
        if (argc != 2) {
                fprintf(stderr, "usage: %s <net>|up|down\n", argv[0]);
                return -1;
@@ -359,8 +357,6 @@ jt_ptl_list_nids(int argc, char **argv)
         int                      count;
         int                      rc;
 
-       fprintf(stderr, "This command has been deprecated. Plesae use 'lnetctl net show'\n");
-
         all = (argc == 2) && (strcmp(argv[1], "all") == 0);
         /* Hack to pass back value */
         return_nid = (argc == 2) && (argv[1][0] == 1);
@@ -922,8 +918,6 @@ int jt_ptl_ping(int argc, char **argv)
         char                    *sep;
         int                      i;
 
-       fprintf(stderr, "This command has been deprecated. Plesae use 'lnetctl ping'\n");
-
         if (argc < 2) {
                 fprintf(stderr, "usage: %s id [timeout (secs)]\n", argv[0]);
                 return 0;
@@ -1229,8 +1223,6 @@ jt_ptl_print_routes (int argc, char **argv)
        int                       alive;
        unsigned int              pri;
 
-       fprintf(stderr, "This command has been deprecated. Plesae use 'lnetctl route show'\n");
-
        for (index = 0; ; index++) {
                LIBCFS_IOC_INIT_V2(data, cfg_hdr);
                data.cfg_count = index;
@@ -1323,6 +1315,57 @@ fault_attr_ptl_parse(char *ptl_str, __u64 *mask_p)
 }
 
 static int
+fault_attr_health_error_parse(char *error, __u32 *mask)
+{
+       if (!strcasecmp(error, "local_interrupt")) {
+               *mask |= HSTATUS_LOCAL_INTERRUPT_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "local_dropped")) {
+               *mask |= HSTATUS_LOCAL_DROPPED_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "local_aborted")) {
+               *mask |= HSTATUS_LOCAL_ABORTED_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "local_no_route")) {
+               *mask |= HSTATUS_LOCAL_NO_ROUTE_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "local_error")) {
+               *mask |= HSTATUS_LOCAL_ERROR_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "local_timeout")) {
+               *mask |= HSTATUS_LOCAL_TIMEOUT_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "remote_error")) {
+               *mask |= HSTATUS_REMOTE_ERROR_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "remote_dropped")) {
+               *mask |= HSTATUS_REMOTE_DROPPED_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "remote_timeout")) {
+               *mask |= HSTATUS_REMOTE_TIMEOUT_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "network_timeout")) {
+               *mask |= HSTATUS_NETWORK_TIMEOUT_BIT;
+               return 0;
+       }
+       if (!strcasecmp(error, "random")) {
+               *mask = HSTATUS_RANDOM;
+               return 0;
+       }
+
+       return -1;
+}
+
+static int
 fault_simul_rule_add(__u32 opc, char *name, int argc, char **argv)
 {
        struct libcfs_ioctl_data  data = { { 0 } };
@@ -1335,9 +1378,11 @@ fault_simul_rule_add(__u32 opc, char *name, int argc, char **argv)
        { .name = "dest",     .has_arg = required_argument, .val = 'd' },
        { .name = "rate",     .has_arg = required_argument, .val = 'r' },
        { .name = "interval", .has_arg = required_argument, .val = 'i' },
+       { .name = "random",   .has_arg = no_argument,       .val = 'n' },
        { .name = "latency",  .has_arg = required_argument, .val = 'l' },
        { .name = "portal",   .has_arg = required_argument, .val = 'p' },
        { .name = "message",  .has_arg = required_argument, .val = 'm' },
+       { .name = "health_error",  .has_arg = required_argument, .val = 'e' },
        { .name = NULL } };
 
        if (argc == 1) {
@@ -1346,7 +1391,7 @@ fault_simul_rule_add(__u32 opc, char *name, int argc, char **argv)
                return -1;
        }
 
-       optstr = opc == LNET_CTL_DROP_ADD ? "s:d:r:i:p:m:" : "s:d:r:l:p:m:";
+       optstr = opc == LNET_CTL_DROP_ADD ? "s:d:r:i:p:m:e:n" : "s:d:r:l:p:m:";
        memset(&attr, 0, sizeof(attr));
        while (1) {
                char c = getopt_long(argc, argv, optstr, opts, NULL);
@@ -1374,6 +1419,20 @@ fault_simul_rule_add(__u32 opc, char *name, int argc, char **argv)
                                attr.u.delay.la_rate = strtoul(optarg, NULL, 0);
                        break;
 
+               case 'e':
+                       if (opc == LNET_CTL_DROP_ADD) {
+                               rc = fault_attr_health_error_parse(optarg,
+                                       &attr.u.drop.da_health_error_mask);
+                               if (rc)
+                                       goto getopt_failed;
+                       }
+                       break;
+
+               case 'n':
+                       if (opc == LNET_CTL_DROP_ADD)
+                               attr.u.drop.da_random = true;
+                       break;
+
                case 'i': /* time interval (# seconds) for message drop */
                        if (opc == LNET_CTL_DROP_ADD)
                                attr.u.drop.da_interval = strtoul(optarg,
@@ -1416,6 +1475,12 @@ fault_simul_rule_add(__u32 opc, char *name, int argc, char **argv)
                                "but not both at the same time.\n");
                        return -1;
                }
+
+               if (attr.u.drop.da_random &&
+                   attr.u.drop.da_interval == 0) {
+                       fprintf(stderr, "please provide an interval to randomize\n");
+                       return -1;
+               }
        } else if (opc == LNET_CTL_DELAY_ADD) {
                if (!((attr.u.delay.la_rate == 0) ^
                      (attr.u.delay.la_interval == 0))) {