#define D_READA 0x00400000 /* read-ahead */
#define D_MMAP 0x00800000
#define D_CONFIG 0x01000000
+#define D_CONSOLE 0x02000000
/* If you change these values, please keep these files up to date...
* portals/utils/debug.c
* utils/lconf
#define CDEBUG(mask, format, a...) \
do { \
CHECK_STACK(CDEBUG_STACK); \
- if (((mask) & (D_ERROR | D_EMERG | D_WARNING)) || \
+ if (((mask) & (D_ERROR | D_EMERG | D_WARNING | D_CONSOLE)) || \
(portal_debug & (mask) && \
portal_subsystem_debug & DEBUG_SUBSYSTEM)) \
portals_debug_msg(DEBUG_SUBSYSTEM, mask, \
#define CERROR(format, a...) CDEBUG_LIMIT(D_ERROR, format, ## a)
#define CEMERG(format, a...) CDEBUG(D_EMERG, format, ## a)
+#define LCONSOLE(mask, format, a...) CDEBUG(D_CONSOLE | (mask), format, ## a)
+#define LCONSOLE_INFO(format, a...) CDEBUG_LIMIT(D_CONSOLE, format, ## a)
+#define LCONSOLE_WARN(format, a...) CDEBUG_LIMIT(D_CONSOLE | D_WARNING, format, ## a)
+#define LCONSOLE_ERROR(format, a...) CDEBUG_LIMIT(D_CONSOLE | D_ERROR, format, ## a)
+#define LCONSOLE_EMERG(format, a...) CDEBUG(D_CONSOLE | D_EMERG, format, ## a)
+
#define GOTO(label, rc) \
do { \
long GOTO__ret = (long)(rc); \
char *format, ...)
__attribute__ ((format (printf, 7, 8)));
-
static inline void cfs_slow_warning(cfs_time_t now, int seconds, char *msg)
{
if (cfs_time_after(cfs_time_current(),
/* Actual error */
LASSERT (rc < 0);
- if (!conn->ksnc_closing)
+ if (!conn->ksnc_closing) {
+ switch (rc) {
+ case -ECONNRESET:
+ LCONSOLE_WARN("Host %u.%u.%u.%u reset our connection "
+ "while we were sending data; it may have "
+ "rebooted.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
+ break;
+ default:
+ LCONSOLE_WARN("There was an unexpected network error "
+ "while writing to %u.%u.%u.%u: %d.\n",
+ HIPQUAD(conn->ksnc_ipaddr), rc);
+ break;
+ }
CERROR("[%p] Error %d on write to "LPX64
" ip %d.%d.%d.%d:%d\n", conn, rc,
conn->ksnc_peer->ksnp_nid,
HIPQUAD(conn->ksnc_ipaddr),
conn->ksnc_port);
+ }
ksocknal_close_conn_and_siblings (conn, rc);
ksocknal_tx_launched (tx);
int type;
ptl_hdr_t hdr;
ptl_magicversion_t *hmv;
+ char ipbuf[PTL_NALFMT_SIZE];
hmv = (ptl_magicversion_t *)&hdr.dest_nid;
LASSERT (sizeof (*hmv) == sizeof (hdr.dest_nid));
if (*nid == PTL_NID_ANY) { /* don't know peer's nid yet */
*nid = le64_to_cpu(hdr.src_nid);
} else if (*nid != le64_to_cpu (hdr.src_nid)) {
+ LCONSOLE_ERROR("Connected successfully to nid "LPX64" on host "
+ "%u.%u.%u.%u, but they claimed they were nid "
+ LPX64" (%s); please check your Lustre "
+ "configuration.\n",
+ *nid, HIPQUAD(conn->ksnc_ipaddr),
+ le64_to_cpu(hdr.src_nid),
+ portals_nid2str(SOCKNAL,
+ le64_to_cpu(hdr.src_nid),
+ ipbuf));
+
CERROR ("Connected to nid "LPX64"@%u.%u.%u.%u "
"but expecting "LPX64"\n",
le64_to_cpu (hdr.src_nid),
unsigned long flags;
int rc;
int type;
+ char *err_msg = NULL;
for (;;) {
for (type = 0; type < SOCKNAL_CONN_NTYPES; type++)
}
}
+ switch (rc) {
+ /* "normal" errors */
+ case -ECONNREFUSED:
+ LCONSOLE_ERROR("Connection was refused by host %u.%u.%u.%u on "
+ "port %d; check that Lustre is running on that "
+ "node.\n",
+ HIPQUAD(route->ksnr_ipaddr),
+ route->ksnr_port);
+ break;
+ case -EHOSTUNREACH:
+ case -ENETUNREACH:
+ LCONSOLE_ERROR("Host %u.%u.%u.%u was unreachable; the network "
+ "or that node may be down, or Lustre may be "
+ "misconfigured.\n",
+ HIPQUAD(route->ksnr_ipaddr));
+ break;
+ case -ETIMEDOUT:
+ LCONSOLE_ERROR("Connecting to host %u.%u.%u.%u on port %d took "
+ "too long; that node may be hung or "
+ "experiencing high load.\n",
+ HIPQUAD(route->ksnr_ipaddr),
+ route->ksnr_port);
+ break;
+ /* errors that should be rare */
+ case -EPROTO:
+ err_msg = "Portals could not negotiate a connection";
+ break;
+ case -EAGAIN:
+ case -EADDRINUSE:
+ /* -EAGAIN is out of ports, but we specify the ports
+ * manually. we really should never get this */
+ err_msg = "no privileged ports were available";
+ break;
+ default:
+ err_msg = "unknown error";
+ break;
+ }
+
+ if (err_msg) {
+ LCONSOLE_ERROR("There was an unexpected error connecting to host "
+ "%u.%u.%u.%u on port %d: %s (error code %d).\n",
+ HIPQUAD(route->ksnr_ipaddr),
+ route->ksnr_port,
+ err_msg, -rc);
+ }
+
/* Connection attempt failed */
write_lock_irqsave (&ksocknal_data.ksnd_global_lock, flags);
LASSERT (!conn->ksnc_closing);
if (SOCK_ERROR(conn->ksnc_sock) != 0) {
- /* Something (e.g. failed keepalive) set the socket error */
atomic_inc (&conn->ksnc_refcount);
+
+ switch (SOCK_ERROR(conn->ksnc_sock)) {
+ case ECONNRESET:
+ LCONSOLE_WARN("A connection with %u.%u.%u.%u "
+ "was reset; they may have "
+ "rebooted.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
+ break;
+ case ETIMEDOUT:
+ LCONSOLE_WARN("A connection with %u.%u.%u.%u "
+ "timed out; the network or that "
+ "node may be down.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
+ break;
+ default:
+ LCONSOLE_WARN("An unexpected network error "
+ "occurred with %u.%u.%u.%u: %d.\n",
+ HIPQUAD(conn->ksnc_ipaddr),
+ SOCK_ERROR(conn->ksnc_sock));
+ break;
+ }
+
+ /* Something (e.g. failed keepalive) set the socket error */
CERROR ("Socket error %d: "LPX64" %p %d.%d.%d.%d\n",
SOCK_ERROR(conn->ksnc_sock), peer->ksnp_nid,
conn, HIPQUAD(conn->ksnc_ipaddr));
+
return (conn);
}
conn->ksnc_rx_deadline)) {
/* Timed out incomplete incoming message */
atomic_inc (&conn->ksnc_refcount);
+ LCONSOLE_ERROR("A timeout occurred receiving data from "
+ "%u.%u.%u.%u; the network or that node "
+ "may be down.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
CERROR ("Timed out RX from "LPX64" %p %d.%d.%d.%d\n",
peer->ksnp_nid,conn,HIPQUAD(conn->ksnc_ipaddr));
return (conn);
/* Timed out messages queued for sending or
* buffered in the socket's send buffer */
atomic_inc (&conn->ksnc_refcount);
+ LCONSOLE_ERROR("A timeout occurred sending data to "
+ "%u.%u.%u.%u; the network or that node "
+ "may be down.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
CERROR ("Timed out TX to "LPX64" %s%d %p %d.%d.%d.%d\n",
peer->ksnp_nid,
list_empty (&conn->ksnc_tx_queue) ? "" : "Q ",
git://git.whamcloud.com / fs / lustre-release.git / commitdiff