import pacemaker-2.0.4-6.el8

# Conflicts:
#	.gitignore
#	.pacemaker.metadata
#	SPECS/pacemaker.spec
This commit is contained in:
CentOS Sources 2020-11-03 07:10:19 -05:00 committed by Cloud User
commit c6b88556ad
18 changed files with 28321 additions and 0 deletions

2
.gitignore vendored Normal file
View File

@ -0,0 +1,2 @@
SOURCES/nagios-agents-metadata-105ab8a.tar.gz
SOURCES/pacemaker-2deceaa.tar.gz

2
.pacemaker.metadata Normal file
View File

@ -0,0 +1,2 @@
ea6c0a27fd0ae8ce02f84a11f08a0d79377041c3 SOURCES/nagios-agents-metadata-105ab8a.tar.gz
78c94fdcf59cfb064d4433e1b8f71fd856eeec5f SOURCES/pacemaker-2deceaa.tar.gz

4947
SOURCES/001-rules.patch Normal file

File diff suppressed because it is too large Load Diff

8664
SOURCES/002-demote.patch Normal file

File diff suppressed because it is too large Load Diff

30
SOURCES/003-trace.patch Normal file
View File

@ -0,0 +1,30 @@
From 47c3e06b098c7e148c54675588d03b4d2bea40b5 Mon Sep 17 00:00:00 2001
From: Chris Lumens <clumens@redhat.com>
Date: Mon, 22 Jun 2020 16:20:01 -0400
Subject: [PATCH] Fix: libpacemaker: Don't allow a potential NULL in a format
string.
This is only tripping up F32 s390x builds, but I don't suppose there's
any reason it couldn't come up elsewhere later.
---
lib/pacemaker/pcmk_sched_constraints.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/lib/pacemaker/pcmk_sched_constraints.c b/lib/pacemaker/pcmk_sched_constraints.c
index 9c3a88d..d8c3e69 100644
--- a/lib/pacemaker/pcmk_sched_constraints.c
+++ b/lib/pacemaker/pcmk_sched_constraints.c
@@ -1595,8 +1595,8 @@ custom_action_order(pe_resource_t * lh_rsc, char *lh_action_task, pe_action_t *
order = calloc(1, sizeof(pe__ordering_t));
crm_trace("Creating[%d] %s %s %s - %s %s %s", data_set->order_id,
- lh_rsc?lh_rsc->id:"NA", lh_action_task, lh_action?lh_action->uuid:"NA",
- rh_rsc?rh_rsc->id:"NA", rh_action_task, rh_action?rh_action->uuid:"NA");
+ lh_rsc?lh_rsc->id:"NA", lh_action_task?lh_action_task:"NA", lh_action?lh_action->uuid:"NA",
+ rh_rsc?rh_rsc->id:"NA", rh_action_task?rh_action_task:"NA", rh_action?rh_action->uuid:"NA");
/* CRM_ASSERT(data_set->order_id != 291); */
--
1.8.3.1

27
SOURCES/004-test.patch Normal file
View File

@ -0,0 +1,27 @@
From 7ed7675615ada7d0be5654e0dcb26de60cf5b5e9 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Mon, 22 Jun 2020 20:03:56 -0500
Subject: [PATCH] Test: scheduler: explicitly disable concurrent-fencing in
on_fail_demote4
... so the expected output is the same regardless of what default the build was
compiled with
---
cts/scheduler/on_fail_demote4.xml | 1 +
1 file changed, 1 insertion(+)
diff --git a/cts/scheduler/on_fail_demote4.xml b/cts/scheduler/on_fail_demote4.xml
index eb4c4cc..1082266 100644
--- a/cts/scheduler/on_fail_demote4.xml
+++ b/cts/scheduler/on_fail_demote4.xml
@@ -8,6 +8,7 @@
<nvpair id="cts-shutdown-escalation" name="shutdown-escalation" value="5min"/>
<nvpair id="cts-batch-limit" name="batch-limit" value="10"/>
<nvpair id="cts-dc-deadtime" name="dc-deadtime" value="5s"/>
+ <nvpair id="cts-concurrent-fencing" name="concurrent-fencing" value="false"/>
<nvpair id="cib-bootstrap-options-have-watchdog" name="have-watchdog" value="false"/>
<nvpair id="cib-bootstrap-options-dc-version" name="dc-version" value="2.0.4-578.6e1b582.git.el7_8-6e1b582"/>
<nvpair id="cib-bootstrap-options-cluster-infrastructure" name="cluster-infrastructure" value="corosync"/>
--
1.8.3.1

View File

@ -0,0 +1,32 @@
From 85040eb19b9405464b01a7e67eb6769d2a03c611 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Fri, 19 Jun 2020 17:49:22 -0500
Subject: [PATCH] Doc: sysconfig: remove outdated reference to wildcards in
PCMK_trace_files
Wildcards stopped working when the log filtering implementation changed in
1.1.8 to support PCMK_trace_tags. It's not worth the effort to fix at this
point, so just update the comment in the sysconfig file.
---
daemons/pacemakerd/pacemaker.sysconfig | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/daemons/pacemakerd/pacemaker.sysconfig b/daemons/pacemakerd/pacemaker.sysconfig
index c7745d8..e4a5c4d 100644
--- a/daemons/pacemakerd/pacemaker.sysconfig
+++ b/daemons/pacemakerd/pacemaker.sysconfig
@@ -34,9 +34,8 @@
# Log all messages from a comma-separated list of functions.
# PCMK_trace_functions=function1,function2,function3
-# Log all messages from a comma-separated list of files (no path).
-# Wildcards are supported, e.g. PCMK_trace_files=prefix*.c
-# PCMK_trace_files=file.c,other.h
+# Log all messages from a comma-separated list of file names (without path).
+# PCMK_trace_files=file1.c,file2.c
# Log all messages matching comma-separated list of formats.
# PCMK_trace_formats="Sent delete %d"
--
1.8.3.1

File diff suppressed because it is too large Load Diff

4146
SOURCES/007-ipc_model.patch Normal file

File diff suppressed because it is too large Load Diff

685
SOURCES/008-crm_node.patch Normal file
View File

@ -0,0 +1,685 @@
From e01a1178fd8f5b99895683b3af9998e9485d12a4 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Fri, 24 Apr 2020 16:42:02 -0500
Subject: [PATCH 1/3] Feature: controller: add new IPC API command for getting
node list
This is based on and will replace the corresponding functionality in
pacemakerd.
---
daemons/controld/controld_messages.c | 44 ++++++++++++++++++++++++++++--
include/crm/common/ipc_controld.h | 13 +++++++++
include/crm_internal.h | 1 +
lib/common/ipc_controld.c | 53 ++++++++++++++++++++++++++++++++++++
4 files changed, 109 insertions(+), 2 deletions(-)
diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c
index 0be04d0..423f006 100644
--- a/daemons/controld/controld_messages.c
+++ b/daemons/controld/controld_messages.c
@@ -375,10 +375,11 @@ relay_message(xmlNode * msg, gboolean originated_locally)
is_local = 0;
} else if (is_for_crm) {
- if (safe_str_eq(task, CRM_OP_NODE_INFO)) {
+ if (safe_str_eq(task, CRM_OP_NODE_INFO)
+ || safe_str_eq(task, PCMK__CONTROLD_CMD_NODES)) {
/* Node info requests do not specify a host, which is normally
* treated as "all hosts", because the whole point is that the
- * client doesn't know the local node name. Always handle these
+ * client may not know the local node name. Always handle these
* requests locally.
*/
is_local = 1;
@@ -784,6 +785,42 @@ handle_ping(xmlNode *msg)
}
/*!
+ * \brief Handle a PCMK__CONTROLD_CMD_NODES message
+ *
+ * \return Next FSA input
+ */
+static enum crmd_fsa_input
+handle_node_list(xmlNode *request)
+{
+ GHashTableIter iter;
+ crm_node_t *node = NULL;
+ xmlNode *reply = NULL;
+ xmlNode *reply_data = NULL;
+
+ // Create message data for reply
+ reply_data = create_xml_node(NULL, XML_CIB_TAG_NODES);
+ g_hash_table_iter_init(&iter, crm_peer_cache);
+ while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) {
+ xmlNode *xml = create_xml_node(reply_data, XML_CIB_TAG_NODE);
+
+ crm_xml_add_ll(xml, XML_ATTR_ID, (long long) node->id); // uint32_t
+ crm_xml_add(xml, XML_ATTR_UNAME, node->uname);
+ crm_xml_add(xml, XML_NODE_IN_CLUSTER, node->state);
+ }
+
+ // Create and send reply
+ reply = create_reply(request, reply_data);
+ free_xml(reply_data);
+ if (reply) {
+ (void) relay_message(reply, TRUE);
+ free_xml(reply);
+ }
+
+ // Nothing further to do
+ return I_NULL;
+}
+
+/*!
* \brief Handle a CRM_OP_NODE_INFO request
*
* \param[in] msg Message XML
@@ -1080,6 +1117,9 @@ handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause)
remote_ra_process_maintenance_nodes(xml);
+ } else if (strcmp(op, PCMK__CONTROLD_CMD_NODES) == 0) {
+ return handle_node_list(stored_msg);
+
/*========== (NOT_DC)-Only Actions ==========*/
} else if (!AM_I_DC) {
diff --git a/include/crm/common/ipc_controld.h b/include/crm/common/ipc_controld.h
index 0ebabfc..b817357 100644
--- a/include/crm/common/ipc_controld.h
+++ b/include/crm/common/ipc_controld.h
@@ -22,6 +22,7 @@ extern "C" {
*/
#include <stdbool.h> // bool
+#include <glib.h> // GList
#include <libxml/tree.h> // xmlNode
#include <crm/common/ipc.h> // pcmk_ipc_api_t
@@ -32,8 +33,16 @@ enum pcmk_controld_api_reply {
pcmk_controld_reply_info,
pcmk_controld_reply_resource,
pcmk_controld_reply_ping,
+ pcmk_controld_reply_nodes,
};
+// Node information passed with pcmk_controld_reply_nodes
+typedef struct {
+ uint32_t id;
+ const char *uname;
+ const char *state;
+} pcmk_controld_api_node_t;
+
/*!
* Controller reply passed to event callback
*
@@ -72,6 +81,9 @@ typedef struct {
const char *fsa_state;
const char *result;
} ping;
+
+ // pcmk_controld_reply_nodes
+ GList *nodes; // list of pcmk_controld_api_node_t *
} data;
} pcmk_controld_api_reply_t;
@@ -88,6 +100,7 @@ int pcmk_controld_api_refresh(pcmk_ipc_api_t *api, const char *target_node,
const char *provider, const char *type,
bool cib_only);
int pcmk_controld_api_ping(pcmk_ipc_api_t *api, const char *node_name);
+int pcmk_controld_api_list_nodes(pcmk_ipc_api_t *api);
int pcmk_controld_api_shutdown(pcmk_ipc_api_t *api, const char *node_name);
int pcmk_controld_api_start_election(pcmk_ipc_api_t *api);
unsigned int pcmk_controld_api_replies_expected(pcmk_ipc_api_t *api);
diff --git a/include/crm_internal.h b/include/crm_internal.h
index fd56fc6..cf8999f 100644
--- a/include/crm_internal.h
+++ b/include/crm_internal.h
@@ -122,6 +122,7 @@ pid_t pcmk_locate_sbd(void);
#define PCMK__ATTRD_CMD_SYNC_RESPONSE "sync-response"
#define PCMK__ATTRD_CMD_CLEAR_FAILURE "clear-failure"
+#define PCMK__CONTROLD_CMD_NODES "list-nodes"
/*
* Environment variables used by Pacemaker
diff --git a/lib/common/ipc_controld.c b/lib/common/ipc_controld.c
index 22bb733..a733dd3 100644
--- a/lib/common/ipc_controld.c
+++ b/lib/common/ipc_controld.c
@@ -120,6 +120,28 @@ set_ping_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data)
data->data.ping.result = crm_element_value(msg_data, XML_PING_ATTR_STATUS);
}
+static void
+set_nodes_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data)
+{
+ pcmk_controld_api_node_t *node_info;
+
+ data->reply_type = pcmk_controld_reply_nodes;
+ for (xmlNode *node = first_named_child(msg_data, XML_CIB_TAG_NODE);
+ node != NULL; node = crm_next_same_xml(node)) {
+
+ long long id_ll = 0;
+
+ node_info = calloc(1, sizeof(pcmk_controld_api_node_t));
+ crm_element_value_ll(node, XML_ATTR_ID, &id_ll);
+ if (id_ll > 0) {
+ node_info->id = id_ll;
+ }
+ node_info->uname = crm_element_value(node, XML_ATTR_UNAME);
+ node_info->state = crm_element_value(node, XML_NODE_IN_CLUSTER);
+ data->data.nodes = g_list_prepend(data->data.nodes, node_info);
+ }
+}
+
static bool
reply_expected(pcmk_ipc_api_t *api, xmlNode *request)
{
@@ -201,6 +223,9 @@ dispatch(pcmk_ipc_api_t *api, xmlNode *reply)
} else if (!strcmp(value, CRM_OP_PING)) {
set_ping_data(&reply_data, msg_data);
+ } else if (!strcmp(value, PCMK__CONTROLD_CMD_NODES)) {
+ set_nodes_data(&reply_data, msg_data);
+
} else {
crm_debug("Unrecognizable controller message: unknown command '%s'",
value);
@@ -210,6 +235,11 @@ dispatch(pcmk_ipc_api_t *api, xmlNode *reply)
}
pcmk__call_ipc_callback(api, pcmk_ipc_event_reply, status, &reply_data);
+
+ // Free any reply data that was allocated
+ if (safe_str_eq(value, PCMK__CONTROLD_CMD_NODES)) {
+ g_list_free_full(reply_data.data.nodes, free);
+ }
}
pcmk__ipc_methods_t *
@@ -376,6 +406,29 @@ pcmk_controld_api_ping(pcmk_ipc_api_t *api, const char *node_name)
}
/*!
+ * \brief Ask the controller for cluster information
+ *
+ * \param[in] api Controller connection
+ *
+ * \return Standard Pacemaker return code
+ * \note Event callback will get a reply of type pcmk_controld_reply_nodes.
+ */
+int
+pcmk_controld_api_list_nodes(pcmk_ipc_api_t *api)
+{
+ xmlNode *request;
+ int rc = EINVAL;
+
+ request = create_controller_request(api, PCMK__CONTROLD_CMD_NODES, NULL,
+ NULL);
+ if (request != NULL) {
+ rc = send_controller_request(api, request, true);
+ free_xml(request);
+ }
+ return rc;
+}
+
+/*!
* \internal
* \brief Ask the controller to shut down
*
--
1.8.3.1
From 74e2d8d18bf534c1ec6f0e0f44a90772d393a553 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Thu, 2 Jul 2020 11:51:56 -0500
Subject: [PATCH 2/3] Refactor: functionize numeric comparisons of strings
This moves the guts of sort_node_uname() from libpe_status to a new function,
pcmk_numeric_strcasecmp(), in libcrmcommon, so it can be used with strings and
not just pe_node_t objects.
---
include/crm/common/util.h | 1 +
lib/common/strings.c | 65 +++++++++++++++++++++++++++++++++++++++++++++++
lib/pengine/utils.c | 50 ++----------------------------------
3 files changed, 68 insertions(+), 48 deletions(-)
diff --git a/include/crm/common/util.h b/include/crm/common/util.h
index 67d74d2..bb97b0a 100644
--- a/include/crm/common/util.h
+++ b/include/crm/common/util.h
@@ -59,6 +59,7 @@ gboolean crm_strcase_equal(gconstpointer a, gconstpointer b);
char *crm_strdup_printf(char const *format, ...) __attribute__ ((__format__ (__printf__, 1, 2)));
int pcmk__parse_ll_range(const char *srcstring, long long *start, long long *end);
gboolean pcmk__str_in_list(GList *lst, const gchar *s);
+int pcmk_numeric_strcasecmp(const char *s1, const char *s2);
# define safe_str_eq(a, b) crm_str_eq(a, b, FALSE)
# define crm_str_hash g_str_hash_traditional
diff --git a/lib/common/strings.c b/lib/common/strings.c
index 4562738..bd68ccf 100644
--- a/lib/common/strings.c
+++ b/lib/common/strings.c
@@ -16,6 +16,7 @@
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
+#include <ctype.h>
#include <limits.h>
#include <bzlib.h>
#include <sys/types.h>
@@ -715,3 +716,67 @@ pcmk__str_none_of(const char *s, ...)
return g_list_find_custom(lst, s, (GCompareFunc) strcmp) != NULL;
}
+
+/*
+ * \brief Sort strings, with numeric portions sorted numerically
+ *
+ * Sort two strings case-insensitively like strcasecmp(), but with any numeric
+ * portions of the string sorted numerically. This is particularly useful for
+ * node names (for example, "node10" will sort higher than "node9" but lower
+ * than "remotenode9").
+ *
+ * \param[in] s1 First string to compare (must not be NULL)
+ * \param[in] s2 Second string to compare (must not be NULL)
+ *
+ * \retval -1 \p s1 comes before \p s2
+ * \retval 0 \p s1 and \p s2 are equal
+ * \retval 1 \p s1 comes after \p s2
+ */
+int
+pcmk_numeric_strcasecmp(const char *s1, const char *s2)
+{
+ while (*s1 && *s2) {
+ if (isdigit(*s1) && isdigit(*s2)) {
+ // If node names contain a number, sort numerically
+
+ char *end1 = NULL;
+ char *end2 = NULL;
+ long num1 = strtol(s1, &end1, 10);
+ long num2 = strtol(s2, &end2, 10);
+
+ // allow ordering e.g. 007 > 7
+ size_t len1 = end1 - s1;
+ size_t len2 = end2 - s2;
+
+ if (num1 < num2) {
+ return -1;
+ } else if (num1 > num2) {
+ return 1;
+ } else if (len1 < len2) {
+ return -1;
+ } else if (len1 > len2) {
+ return 1;
+ }
+ s1 = end1;
+ s2 = end2;
+ } else {
+ // Compare non-digits case-insensitively
+ int lower1 = tolower(*s1);
+ int lower2 = tolower(*s2);
+
+ if (lower1 < lower2) {
+ return -1;
+ } else if (lower1 > lower2) {
+ return 1;
+ }
+ ++s1;
+ ++s2;
+ }
+ }
+ if (!*s1 && *s2) {
+ return -1;
+ } else if (*s1 && !*s2) {
+ return 1;
+ }
+ return 0;
+}
diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c
index ce3c260..584def4 100644
--- a/lib/pengine/utils.c
+++ b/lib/pengine/utils.c
@@ -13,7 +13,6 @@
#include <crm/common/xml.h>
#include <crm/common/util.h>
-#include <ctype.h>
#include <glib.h>
#include <stdbool.h>
@@ -214,53 +213,8 @@ pe__node_list2table(GList *list)
gint
sort_node_uname(gconstpointer a, gconstpointer b)
{
- const char *name_a = ((const pe_node_t *) a)->details->uname;
- const char *name_b = ((const pe_node_t *) b)->details->uname;
-
- while (*name_a && *name_b) {
- if (isdigit(*name_a) && isdigit(*name_b)) {
- // If node names contain a number, sort numerically
-
- char *end_a = NULL;
- char *end_b = NULL;
- long num_a = strtol(name_a, &end_a, 10);
- long num_b = strtol(name_b, &end_b, 10);
-
- // allow ordering e.g. 007 > 7
- size_t len_a = end_a - name_a;
- size_t len_b = end_b - name_b;
-
- if (num_a < num_b) {
- return -1;
- } else if (num_a > num_b) {
- return 1;
- } else if (len_a < len_b) {
- return -1;
- } else if (len_a > len_b) {
- return 1;
- }
- name_a = end_a;
- name_b = end_b;
- } else {
- // Compare non-digits case-insensitively
- int lower_a = tolower(*name_a);
- int lower_b = tolower(*name_b);
-
- if (lower_a < lower_b) {
- return -1;
- } else if (lower_a > lower_b) {
- return 1;
- }
- ++name_a;
- ++name_b;
- }
- }
- if (!*name_a && *name_b) {
- return -1;
- } else if (*name_a && !*name_b) {
- return 1;
- }
- return 0;
+ return pcmk_numeric_strcasecmp(((const pe_node_t *) a)->details->uname,
+ ((const pe_node_t *) b)->details->uname);
}
/*!
--
1.8.3.1
From 8461509158e06365122dc741c527c83c94e966ce Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Fri, 24 Apr 2020 19:35:19 -0500
Subject: [PATCH 3/3] Fix: tools: crm_node -l and -p now work from Pacemaker
Remote nodes
crm_node now asks the controller for the cluster node list, instead of
pacemakerd. This allows it to work from Pacemaker Remote nodes, since
controller IPC is proxied but pacemakerd IPC is not.
---
tools/crm_node.c | 176 +++++++++++++++++++++----------------------------------
1 file changed, 67 insertions(+), 109 deletions(-)
diff --git a/tools/crm_node.c b/tools/crm_node.c
index 57c2ee5..146454d 100644
--- a/tools/crm_node.c
+++ b/tools/crm_node.c
@@ -130,6 +130,16 @@ remove_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError *
return TRUE;
}
+static gint
+sort_node(gconstpointer a, gconstpointer b)
+{
+ const pcmk_controld_api_node_t *node_a = a;
+ const pcmk_controld_api_node_t *node_b = b;
+
+ return pcmk_numeric_strcasecmp((node_a->uname? node_a->uname : ""),
+ (node_b->uname? node_b->uname : ""));
+}
+
static void
controller_event_cb(pcmk_ipc_api_t *controld_api,
enum pcmk_ipc_event event_type, crm_exit_t status,
@@ -157,15 +167,16 @@ controller_event_cb(pcmk_ipc_api_t *controld_api,
crm_exit_str(status));
goto done;
}
- if (reply->reply_type != pcmk_controld_reply_info) {
- fprintf(stderr, "error: Unknown reply type %d from controller\n",
- reply->reply_type);
- goto done;
- }
// Parse desired info from reply and display to user
switch (options.command) {
case 'i':
+ if (reply->reply_type != pcmk_controld_reply_info) {
+ fprintf(stderr,
+ "error: Unknown reply type %d from controller\n",
+ reply->reply_type);
+ goto done;
+ }
if (reply->data.node_info.id == 0) {
fprintf(stderr,
"error: Controller reply did not contain node ID\n");
@@ -177,6 +188,12 @@ controller_event_cb(pcmk_ipc_api_t *controld_api,
case 'n':
case 'N':
+ if (reply->reply_type != pcmk_controld_reply_info) {
+ fprintf(stderr,
+ "error: Unknown reply type %d from controller\n",
+ reply->reply_type);
+ goto done;
+ }
if (reply->data.node_info.uname == NULL) {
fprintf(stderr, "Node is not known to cluster\n");
exit_code = CRM_EX_NOHOST;
@@ -186,6 +203,12 @@ controller_event_cb(pcmk_ipc_api_t *controld_api,
break;
case 'q':
+ if (reply->reply_type != pcmk_controld_reply_info) {
+ fprintf(stderr,
+ "error: Unknown reply type %d from controller\n",
+ reply->reply_type);
+ goto done;
+ }
printf("%d\n", reply->data.node_info.have_quorum);
if (!(reply->data.node_info.have_quorum)) {
exit_code = CRM_EX_QUORUM;
@@ -193,6 +216,36 @@ controller_event_cb(pcmk_ipc_api_t *controld_api,
}
break;
+ case 'l':
+ case 'p':
+ if (reply->reply_type != pcmk_controld_reply_nodes) {
+ fprintf(stderr,
+ "error: Unknown reply type %d from controller\n",
+ reply->reply_type);
+ goto done;
+ }
+ reply->data.nodes = g_list_sort(reply->data.nodes, sort_node);
+ for (GList *node_iter = reply->data.nodes;
+ node_iter != NULL; node_iter = node_iter->next) {
+
+ pcmk_controld_api_node_t *node = node_iter->data;
+ const char *uname = (node->uname? node->uname : "");
+ const char *state = (node->state? node->state : "");
+
+ if (options.command == 'l') {
+ printf("%lu %s %s\n",
+ (unsigned long) node->id, uname, state);
+
+ // i.e. CRM_NODE_MEMBER, but we don't want to include cluster.h
+ } else if (!strcmp(state, "member")) {
+ printf("%s ", uname);
+ }
+ }
+ if (options.command == 'p') {
+ printf("\n");
+ }
+ break;
+
default:
fprintf(stderr, "internal error: Controller reply not expected\n");
exit_code = CRM_EX_SOFTWARE;
@@ -207,7 +260,7 @@ done:
}
static void
-run_controller_mainloop(uint32_t nodeid)
+run_controller_mainloop(uint32_t nodeid, bool list_nodes)
{
pcmk_ipc_api_t *controld_api = NULL;
int rc;
@@ -233,7 +286,11 @@ run_controller_mainloop(uint32_t nodeid)
return;
}
- rc = pcmk_controld_api_node_info(controld_api, nodeid);
+ if (list_nodes) {
+ rc = pcmk_controld_api_list_nodes(controld_api);
+ } else {
+ rc = pcmk_controld_api_node_info(controld_api, nodeid);
+ }
if (rc != pcmk_rc_ok) {
fprintf(stderr, "error: Could not ping controller: %s\n",
pcmk_rc_str(rc));
@@ -263,7 +320,7 @@ print_node_name(void)
} else {
// Otherwise ask the controller
- run_controller_mainloop(0);
+ run_controller_mainloop(0, false);
}
}
@@ -444,105 +501,6 @@ remove_node(const char *target_uname)
exit_code = CRM_EX_OK;
}
-static gint
-compare_node_xml(gconstpointer a, gconstpointer b)
-{
- const char *a_name = crm_element_value((xmlNode*) a, "uname");
- const char *b_name = crm_element_value((xmlNode*) b, "uname");
-
- return strcmp((a_name? a_name : ""), (b_name? b_name : ""));
-}
-
-static int
-node_mcp_dispatch(const char *buffer, ssize_t length, gpointer userdata)
-{
- GList *nodes = NULL;
- xmlNode *node = NULL;
- xmlNode *msg = string2xml(buffer);
- const char *uname;
- const char *state;
-
- if (msg == NULL) {
- fprintf(stderr, "error: Could not understand pacemakerd response\n");
- exit_code = CRM_EX_PROTOCOL;
- g_main_loop_quit(mainloop);
- return 0;
- }
-
- crm_log_xml_trace(msg, "message");
-
- for (node = __xml_first_child(msg); node != NULL; node = __xml_next(node)) {
- nodes = g_list_insert_sorted(nodes, node, compare_node_xml);
- }
-
- for (GList *iter = nodes; iter; iter = iter->next) {
- node = (xmlNode*) iter->data;
- uname = crm_element_value(node, "uname");
- state = crm_element_value(node, "state");
-
- if (options.command == 'l') {
- int id = 0;
-
- crm_element_value_int(node, "id", &id);
- printf("%d %s %s\n", id, (uname? uname : ""), (state? state : ""));
-
- // This is CRM_NODE_MEMBER but we don't want to include cluster header
- } else if ((options.command == 'p') && safe_str_eq(state, "member")) {
- printf("%s ", (uname? uname : ""));
- }
- }
- if (options.command == 'p') {
- fprintf(stdout, "\n");
- }
-
- free_xml(msg);
- exit_code = CRM_EX_OK;
- g_main_loop_quit(mainloop);
- return 0;
-}
-
-static void
-lost_pacemakerd(gpointer user_data)
-{
- fprintf(stderr, "error: Lost connection to cluster\n");
- exit_code = CRM_EX_DISCONNECT;
- g_main_loop_quit(mainloop);
-}
-
-static void
-run_pacemakerd_mainloop(void)
-{
- crm_ipc_t *ipc = NULL;
- xmlNode *poke = NULL;
- mainloop_io_t *source = NULL;
-
- struct ipc_client_callbacks ipc_callbacks = {
- .dispatch = node_mcp_dispatch,
- .destroy = lost_pacemakerd
- };
-
- source = mainloop_add_ipc_client(CRM_SYSTEM_MCP, G_PRIORITY_DEFAULT, 0,
- NULL, &ipc_callbacks);
- ipc = mainloop_get_ipc_client(source);
- if (ipc == NULL) {
- fprintf(stderr,
- "error: Could not connect to cluster (is it running?)\n");
- exit_code = CRM_EX_DISCONNECT;
- return;
- }
-
- // Sending anything will get us a list of nodes
- poke = create_xml_node(NULL, "poke");
- crm_ipc_send(ipc, poke, 0, 0, NULL);
- free_xml(poke);
-
- // Handle reply via node_mcp_dispatch()
- mainloop = g_main_loop_new(NULL, FALSE);
- g_main_loop_run(mainloop);
- g_main_loop_unref(mainloop);
- mainloop = NULL;
-}
-
static GOptionContext *
build_arg_context(pcmk__common_args_t *args, GOptionGroup *group) {
GOptionContext *context = NULL;
@@ -627,11 +585,11 @@ main(int argc, char **argv)
case 'i':
case 'q':
case 'N':
- run_controller_mainloop(options.nodeid);
+ run_controller_mainloop(options.nodeid, false);
break;
case 'l':
case 'p':
- run_pacemakerd_mainloop();
+ run_controller_mainloop(0, true);
break;
default:
break;
--
1.8.3.1

View File

@ -0,0 +1,34 @@
From b542a8f667002519fbc07693a796553746c43c12 Mon Sep 17 00:00:00 2001
From: Reid Wahl <nrwahl@protonmail.com>
Date: Sat, 11 Jul 2020 18:31:55 -0700
Subject: [PATCH] Log: controld: Show action timer plus cluster-delay in
action_timer cb
`action_timer_callback()` prints a misleading error message. If it
times out waiting for an action result, the error message prints the
timeout value followed by "(action timeout plus cluster-delay)".
However, only the `action->timeout` value is displayed. `cluster-delay`
is not added in.
Resolves: RHBZ#1856035
---
daemons/controld/controld_te_callbacks.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/daemons/controld/controld_te_callbacks.c b/daemons/controld/controld_te_callbacks.c
index 8506f26..6ddaffe 100644
--- a/daemons/controld/controld_te_callbacks.c
+++ b/daemons/controld/controld_te_callbacks.c
@@ -697,7 +697,8 @@ action_timer_callback(gpointer data)
crm_err("Node %s did not send %s result (via %s) within %dms "
"(action timeout plus cluster-delay)",
(on_node? on_node : ""), (task? task : "unknown action"),
- (via_node? via_node : "controller"), timer->timeout);
+ (via_node? via_node : "controller"),
+ timer->timeout + transition_graph->network_delay);
print_action(LOG_ERR, "Aborting transition, action lost: ", timer->action);
timer->action->failed = TRUE;
--
1.8.3.1

476
SOURCES/010-crm_mon.patch Normal file
View File

@ -0,0 +1,476 @@
From 7056ae08bfa5cafeec9c454cb40aefa7553af6df Mon Sep 17 00:00:00 2001
From: Chris Lumens <clumens@redhat.com>
Date: Thu, 16 Jul 2020 12:53:24 -0400
Subject: [PATCH 1/4] Fix: libcrmcommon: Set out->priv to NULL in free_priv.
init won't do anything if priv is not NULL, so when the private data is
freed, also set it to NULL. This prevents segfaults when reset is
called.
---
lib/common/output_html.c | 1 +
lib/common/output_log.c | 1 +
lib/common/output_text.c | 1 +
lib/common/output_xml.c | 1 +
tools/crm_mon_curses.c | 1 +
5 files changed, 5 insertions(+)
diff --git a/lib/common/output_html.c b/lib/common/output_html.c
index c8f0088..fc06641 100644
--- a/lib/common/output_html.c
+++ b/lib/common/output_html.c
@@ -72,6 +72,7 @@ html_free_priv(pcmk__output_t *out) {
g_queue_free(priv->parent_q);
g_slist_free(priv->errors);
free(priv);
+ out->priv = NULL;
}
static bool
diff --git a/lib/common/output_log.c b/lib/common/output_log.c
index 5b45ce4..0208046 100644
--- a/lib/common/output_log.c
+++ b/lib/common/output_log.c
@@ -44,6 +44,7 @@ log_free_priv(pcmk__output_t *out) {
g_queue_free(priv->prefixes);
free(priv);
+ out->priv = NULL;
}
static bool
diff --git a/lib/common/output_text.c b/lib/common/output_text.c
index 54c409a..8f15849 100644
--- a/lib/common/output_text.c
+++ b/lib/common/output_text.c
@@ -43,6 +43,7 @@ text_free_priv(pcmk__output_t *out) {
g_queue_free(priv->parent_q);
free(priv);
+ out->priv = NULL;
}
static bool
diff --git a/lib/common/output_xml.c b/lib/common/output_xml.c
index 8565bfe..858da3f 100644
--- a/lib/common/output_xml.c
+++ b/lib/common/output_xml.c
@@ -54,6 +54,7 @@ xml_free_priv(pcmk__output_t *out) {
g_queue_free(priv->parent_q);
g_slist_free(priv->errors);
free(priv);
+ out->priv = NULL;
}
static bool
diff --git a/tools/crm_mon_curses.c b/tools/crm_mon_curses.c
index d93b847..e9cc023 100644
--- a/tools/crm_mon_curses.c
+++ b/tools/crm_mon_curses.c
@@ -46,6 +46,7 @@ curses_free_priv(pcmk__output_t *out) {
g_queue_free(priv->parent_q);
free(priv);
+ out->priv = NULL;
}
static bool
--
1.8.3.1
From 3779152993ca0e88dc407c918882568217f1b630 Mon Sep 17 00:00:00 2001
From: Chris Lumens <clumens@redhat.com>
Date: Thu, 16 Jul 2020 13:50:24 -0400
Subject: [PATCH 2/4] Fix: libcrmcommon: Make reset and finish work more
similarly.
When finish is called for HTML and XML output formats, various extra
nodes and headers are added, errors are added, etc. None of this stuff
happens on reset. For the HTML format, this also means things like the
CGI headers and title don't get added when reset is called. Make these
two functions much more similar.
Regression in 2.0.3.
See: rhbz#1857728
---
lib/common/output_html.c | 26 ++++++++++++++++----------
lib/common/output_xml.c | 30 ++++++++++++++++--------------
2 files changed, 32 insertions(+), 24 deletions(-)
diff --git a/lib/common/output_html.c b/lib/common/output_html.c
index fc06641..6127df2 100644
--- a/lib/common/output_html.c
+++ b/lib/common/output_html.c
@@ -113,18 +113,11 @@ add_error_node(gpointer data, gpointer user_data) {
}
static void
-html_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_dest) {
+finish_reset_common(pcmk__output_t *out, crm_exit_t exit_status, bool print) {
private_data_t *priv = out->priv;
htmlNodePtr head_node = NULL;
htmlNodePtr charset_node = NULL;
- /* If root is NULL, html_init failed and we are being called from pcmk__output_free
- * in the pcmk__output_new path.
- */
- if (priv == NULL || priv->root == NULL) {
- return;
- }
-
if (cgi_output && print) {
fprintf(out->dest, "Content-Type: text/html\n\n");
}
@@ -174,6 +167,20 @@ html_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy
if (print) {
htmlDocDump(out->dest, priv->root->doc);
}
+}
+
+static void
+html_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_dest) {
+ private_data_t *priv = out->priv;
+
+ /* If root is NULL, html_init failed and we are being called from pcmk__output_free
+ * in the pcmk__output_new path.
+ */
+ if (priv == NULL || priv->root == NULL) {
+ return;
+ }
+
+ finish_reset_common(out, exit_status, print);
if (copy_dest != NULL) {
*copy_dest = copy_xml(priv->root);
@@ -185,8 +192,7 @@ html_reset(pcmk__output_t *out) {
CRM_ASSERT(out != NULL);
if (out->priv != NULL) {
- private_data_t *priv = out->priv;
- htmlDocDump(out->dest, priv->root->doc);
+ finish_reset_common(out, CRM_EX_OK, true);
}
html_free_priv(out);
diff --git a/lib/common/output_xml.c b/lib/common/output_xml.c
index 858da3f..b64a71d 100644
--- a/lib/common/output_xml.c
+++ b/lib/common/output_xml.c
@@ -106,17 +106,10 @@ add_error_node(gpointer data, gpointer user_data) {
}
static void
-xml_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_dest) {
+finish_reset_common(pcmk__output_t *out, crm_exit_t exit_status, bool print) {
xmlNodePtr node;
private_data_t *priv = out->priv;
- /* If root is NULL, xml_init failed and we are being called from pcmk__output_free
- * in the pcmk__output_new path.
- */
- if (priv == NULL || priv->root == NULL) {
- return;
- }
-
if (legacy_xml) {
GSList *node = priv->errors;
@@ -148,6 +141,20 @@ xml_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_
fprintf(out->dest, "%s", buf);
free(buf);
}
+}
+
+static void
+xml_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_dest) {
+ private_data_t *priv = out->priv;
+
+ /* If root is NULL, xml_init failed and we are being called from pcmk__output_free
+ * in the pcmk__output_new path.
+ */
+ if (priv == NULL || priv->root == NULL) {
+ return;
+ }
+
+ finish_reset_common(out, exit_status, print);
if (copy_dest != NULL) {
*copy_dest = copy_xml(priv->root);
@@ -156,15 +163,10 @@ xml_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_
static void
xml_reset(pcmk__output_t *out) {
- char *buf = NULL;
-
CRM_ASSERT(out != NULL);
if (out->priv != NULL) {
- private_data_t *priv = out->priv;
- buf = dump_xml_formatted_with_text(priv->root);
- fprintf(out->dest, "%s", buf);
- free(buf);
+ finish_reset_common(out, CRM_EX_OK, true);
}
xml_free_priv(out);
--
1.8.3.1
From 0f8e4ca5d9a429c934f1e91a1bdf572efd07e0db Mon Sep 17 00:00:00 2001
From: Chris Lumens <clumens@redhat.com>
Date: Thu, 16 Jul 2020 16:09:08 -0400
Subject: [PATCH 3/4] Fix: tools, libcrmcommon: Reopen the output dest on
reset.
This is needed when running crm_mon as a daemon. When we do a reset,
we need to clear out any existing output destination and start writing
again from the beginning. This really only matters when the destination
is a file.
The extra freopen at the end of crm_mon is to handle when crm_mon is
killed. We need to reset the output file to its beginning before
calling finish.
---
lib/common/output_html.c | 3 +++
lib/common/output_log.c | 3 +++
lib/common/output_text.c | 3 +++
lib/common/output_xml.c | 3 +++
tools/crm_mon.c | 9 +++++++++
5 files changed, 21 insertions(+)
diff --git a/lib/common/output_html.c b/lib/common/output_html.c
index 6127df2..6e21031 100644
--- a/lib/common/output_html.c
+++ b/lib/common/output_html.c
@@ -191,6 +191,9 @@ static void
html_reset(pcmk__output_t *out) {
CRM_ASSERT(out != NULL);
+ out->dest = freopen(NULL, "w", out->dest);
+ CRM_ASSERT(out->dest != NULL);
+
if (out->priv != NULL) {
finish_reset_common(out, CRM_EX_OK, true);
}
diff --git a/lib/common/output_log.c b/lib/common/output_log.c
index 0208046..8422ac2 100644
--- a/lib/common/output_log.c
+++ b/lib/common/output_log.c
@@ -72,6 +72,9 @@ static void
log_reset(pcmk__output_t *out) {
CRM_ASSERT(out != NULL);
+ out->dest = freopen(NULL, "w", out->dest);
+ CRM_ASSERT(out->dest != NULL);
+
log_free_priv(out);
log_init(out);
}
diff --git a/lib/common/output_text.c b/lib/common/output_text.c
index 8f15849..2f7e5b0 100644
--- a/lib/common/output_text.c
+++ b/lib/common/output_text.c
@@ -75,6 +75,9 @@ static void
text_reset(pcmk__output_t *out) {
CRM_ASSERT(out != NULL);
+ out->dest = freopen(NULL, "w", out->dest);
+ CRM_ASSERT(out->dest != NULL);
+
text_free_priv(out);
text_init(out);
}
diff --git a/lib/common/output_xml.c b/lib/common/output_xml.c
index b64a71d..9f8e01b 100644
--- a/lib/common/output_xml.c
+++ b/lib/common/output_xml.c
@@ -165,6 +165,9 @@ static void
xml_reset(pcmk__output_t *out) {
CRM_ASSERT(out != NULL);
+ out->dest = freopen(NULL, "w", out->dest);
+ CRM_ASSERT(out->dest != NULL);
+
if (out->priv != NULL) {
finish_reset_common(out, CRM_EX_OK, true);
}
diff --git a/tools/crm_mon.c b/tools/crm_mon.c
index b2e143b..10624c1 100644
--- a/tools/crm_mon.c
+++ b/tools/crm_mon.c
@@ -2014,6 +2014,10 @@ mon_refresh_display(gpointer user_data)
break;
}
+ if (options.daemonize) {
+ out->reset(out);
+ }
+
stonith_history_free(stonith_history);
stonith_history = NULL;
pe_reset_working_set(mon_data_set);
@@ -2179,6 +2183,11 @@ clean_up(crm_exit_t exit_code)
* crm_mon to be able to do so.
*/
if (out != NULL) {
+ if (options.daemonize) {
+ out->dest = freopen(NULL, "w", out->dest);
+ CRM_ASSERT(out->dest != NULL);
+ }
+
switch (output_format) {
case mon_output_cgi:
case mon_output_html:
--
1.8.3.1
From b655c039414d2c7af77c3532222b04684ef1f3d0 Mon Sep 17 00:00:00 2001
From: Chris Lumens <clumens@redhat.com>
Date: Fri, 17 Jul 2020 10:58:32 -0400
Subject: [PATCH 4/4] Fix: tools: Add the http-equiv header to crm_mon at the
right time.
This header is only getting added on termination, which is not a lot of
help if you're running crm_mon in daemonize mode. Putting this header
in at the right time requires a couple changes:
* pcmk__html_add_header doesn't need a parent argument. It was not
being used in the first place.
* The extra_headers list in output_html.c should not be freed in the
reset function. This means it would get freed after every time the
daemonized output is refreshed, which means the header would have to be
added every time too. extra_headers will now only be freed when the
program exits. This is a behavior change, but I can't see why it's
going to be a problem.
* To support that, we need to copy each item in the extra_headers list
when it gets added to the output XML document. This prevents segfaults
when we later free that document.
* handle_html_output no longer needs to exist. That function only
existed to add the http-equiv header at the end, which is wrong.
---
include/crm/common/output.h | 5 ++---
lib/common/output_html.c | 7 ++++---
tools/crm_mon.c | 26 +++++++-------------------
3 files changed, 13 insertions(+), 25 deletions(-)
diff --git a/include/crm/common/output.h b/include/crm/common/output.h
index e7c9417..186bcfe 100644
--- a/include/crm/common/output.h
+++ b/include/crm/common/output.h
@@ -703,15 +703,14 @@ pcmk__output_create_html_node(pcmk__output_t *out, const char *element_name, con
* the following code would generate the tag "<meta http-equiv='refresh' content='19'>":
*
* \code
- * pcmk__html_add_header(parent, "meta", "http-equiv", "refresh", "content", "19", NULL);
+ * pcmk__html_add_header("meta", "http-equiv", "refresh", "content", "19", NULL);
* \endcode
*
- * \param[in,out] parent The node that will be the parent of the new node.
* \param[in] name The HTML tag for the new node.
* \param[in] ... A NULL-terminated key/value list of attributes.
*/
void
-pcmk__html_add_header(xmlNodePtr parent, const char *name, ...)
+pcmk__html_add_header(const char *name, ...)
G_GNUC_NULL_TERMINATED;
#ifdef __cplusplus
diff --git a/lib/common/output_html.c b/lib/common/output_html.c
index 6e21031..259e412 100644
--- a/lib/common/output_html.c
+++ b/lib/common/output_html.c
@@ -139,7 +139,7 @@ finish_reset_common(pcmk__output_t *out, crm_exit_t exit_status, bool print) {
/* Add any extra header nodes the caller might have created. */
for (int i = 0; i < g_slist_length(extra_headers); i++) {
- xmlAddChild(head_node, g_slist_nth_data(extra_headers, i));
+ xmlAddChild(head_node, xmlCopyNode(g_slist_nth_data(extra_headers, i), 1));
}
/* Stylesheets are included two different ways. The first is via a built-in
@@ -185,6 +185,8 @@ html_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy
if (copy_dest != NULL) {
*copy_dest = copy_xml(priv->root);
}
+
+ g_slist_free_full(extra_headers, (GDestroyNotify) xmlFreeNode);
}
static void
@@ -199,7 +201,6 @@ html_reset(pcmk__output_t *out) {
}
html_free_priv(out);
- g_slist_free_full(extra_headers, (GDestroyNotify) xmlFreeNode);
html_init(out);
}
@@ -412,7 +413,7 @@ pcmk__output_create_html_node(pcmk__output_t *out, const char *element_name, con
}
void
-pcmk__html_add_header(xmlNodePtr parent, const char *name, ...) {
+pcmk__html_add_header(const char *name, ...) {
htmlNodePtr header_node;
va_list ap;
diff --git a/tools/crm_mon.c b/tools/crm_mon.c
index 10624c1..7fd2b9c 100644
--- a/tools/crm_mon.c
+++ b/tools/crm_mon.c
@@ -1346,6 +1346,12 @@ main(int argc, char **argv)
options.mon_ops |= mon_op_print_timing | mon_op_inactive_resources;
}
+ if ((output_format == mon_output_html || output_format == mon_output_cgi) &&
+ out->dest != stdout) {
+ pcmk__html_add_header("meta", "http-equiv", "refresh", "content",
+ crm_itoa(options.reconnect_msec/1000), NULL);
+ }
+
crm_info("Starting %s", crm_system_name);
if (cib) {
@@ -2106,15 +2112,6 @@ clean_up_connections(void)
}
}
-static void
-handle_html_output(crm_exit_t exit_code) {
- xmlNodePtr html = NULL;
-
- pcmk__html_add_header(html, "meta", "http-equiv", "refresh", "content",
- crm_itoa(options.reconnect_msec/1000), NULL);
- out->finish(out, exit_code, true, (void **) &html);
-}
-
/*
* De-init ncurses, disconnect from the CIB manager, disconnect fencing,
* deallocate memory and show usage-message if requested.
@@ -2188,16 +2185,7 @@ clean_up(crm_exit_t exit_code)
CRM_ASSERT(out->dest != NULL);
}
- switch (output_format) {
- case mon_output_cgi:
- case mon_output_html:
- handle_html_output(exit_code);
- break;
-
- default:
- out->finish(out, exit_code, true, NULL);
- break;
- }
+ out->finish(out, exit_code, true, NULL);
pcmk__output_free(out);
pcmk__unregister_formats();
--
1.8.3.1

270
SOURCES/011-cts.patch Normal file
View File

@ -0,0 +1,270 @@
From 4e190ebc5460563bae2586b28afb0415f2eb3d1a Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Wed, 1 Jul 2020 20:38:16 -0500
Subject: [PATCH 1/4] Test: CTS: libqb shared memory creates directories now
... so use "rm -rf" instead of "rm -f"
---
cts/CTS.py.in | 2 +-
cts/CTSaudits.py | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/cts/CTS.py.in b/cts/CTS.py.in
index c418318..091bb1f 100644
--- a/cts/CTS.py.in
+++ b/cts/CTS.py.in
@@ -546,7 +546,7 @@ class ClusterManager(UserDict):
if self.rsh(node, self.templates["StopCmd"]) == 0:
# Make sure we can continue even if corosync leaks
# fdata-* is the old name
- #self.rsh(node, "rm -f /dev/shm/qb-* /dev/shm/fdata-*")
+ #self.rsh(node, "rm -rf /dev/shm/qb-* /dev/shm/fdata-*")
self.ShouldBeStatus[node] = "down"
self.cluster_stable(self.Env["DeadTime"])
return 1
diff --git a/cts/CTSaudits.py b/cts/CTSaudits.py
index b7e0827..cc82171 100755
--- a/cts/CTSaudits.py
+++ b/cts/CTSaudits.py
@@ -233,7 +233,7 @@ class FileAudit(ClusterAudit):
for line in lsout:
self.CM.debug("ps[%s]: %s" % (node, line))
- self.CM.rsh(node, "rm -f /dev/shm/qb-*")
+ self.CM.rsh(node, "rm -rf /dev/shm/qb-*")
else:
self.CM.debug("Skipping %s" % node)
--
1.8.3.1
From 4316507d50d51c7864d8d34aac1da31a232b9f42 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Thu, 2 Jul 2020 16:09:20 -0500
Subject: [PATCH 2/4] Test: CTS: ignore error logged by recent pcs versions
... because it is expected when a node is fenced, and we should already see
pacemaker errors if a node is unexpectedly fenced
---
cts/patterns.py | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/cts/patterns.py b/cts/patterns.py
index 96d6471..7eed90c 100644
--- a/cts/patterns.py
+++ b/cts/patterns.py
@@ -21,6 +21,10 @@ class BasePatterns(object):
# Logging bug in some versions of libvirtd
r"libvirtd.*: internal error: Failed to parse PCI config address",
+
+ # pcs can log this when node is fenced, but fencing is OK in some
+ # tests (and we will catch it in pacemaker logs when not OK)
+ r"pcs.daemon:No response from: .* request: get_configs, error:",
]
self.BadNews = []
self.components = {}
--
1.8.3.1
From 598ae0f65bad6ed16978d1ab6e24e8e358e0a1a4 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Thu, 2 Jul 2020 20:40:00 -0500
Subject: [PATCH 3/4] Low: libcrmcommon: avoid assertion on controller protocol
errors
Previously, after a protocol error, we would set reply to NULL and then try to
call crm_element_value() on it, which would log an assertion.
---
lib/common/ipc_controld.c | 46 ++++++++++++++++++++++------------------------
1 file changed, 22 insertions(+), 24 deletions(-)
diff --git a/lib/common/ipc_controld.c b/lib/common/ipc_controld.c
index 5917cc5..22cb9e0 100644
--- a/lib/common/ipc_controld.c
+++ b/lib/common/ipc_controld.c
@@ -187,53 +187,51 @@ dispatch(pcmk_ipc_api_t *api, xmlNode *reply)
crm_debug("Unrecognizable controller message: invalid message type '%s'",
crm_str(value));
status = CRM_EX_PROTOCOL;
- reply = NULL;
+ goto done;
}
if (crm_element_value(reply, XML_ATTR_REFERENCE) == NULL) {
crm_debug("Unrecognizable controller message: no reference");
status = CRM_EX_PROTOCOL;
- reply = NULL;
+ goto done;
}
value = crm_element_value(reply, F_CRM_TASK);
if (value == NULL) {
crm_debug("Unrecognizable controller message: no command name");
status = CRM_EX_PROTOCOL;
- reply = NULL;
+ goto done;
}
// Parse useful info from reply
- if (reply != NULL) {
- reply_data.feature_set = crm_element_value(reply, XML_ATTR_VERSION);
- reply_data.host_from = crm_element_value(reply, F_CRM_HOST_FROM);
- msg_data = get_message_xml(reply, F_CRM_DATA);
+ reply_data.feature_set = crm_element_value(reply, XML_ATTR_VERSION);
+ reply_data.host_from = crm_element_value(reply, F_CRM_HOST_FROM);
+ msg_data = get_message_xml(reply, F_CRM_DATA);
- if (!strcmp(value, CRM_OP_REPROBE)) {
- reply_data.reply_type = pcmk_controld_reply_reprobe;
+ if (!strcmp(value, CRM_OP_REPROBE)) {
+ reply_data.reply_type = pcmk_controld_reply_reprobe;
- } else if (!strcmp(value, CRM_OP_NODE_INFO)) {
- set_node_info_data(&reply_data, msg_data);
+ } else if (!strcmp(value, CRM_OP_NODE_INFO)) {
+ set_node_info_data(&reply_data, msg_data);
- } else if (!strcmp(value, CRM_OP_INVOKE_LRM)) {
- reply_data.reply_type = pcmk_controld_reply_resource;
- reply_data.data.resource.node_state = msg_data;
+ } else if (!strcmp(value, CRM_OP_INVOKE_LRM)) {
+ reply_data.reply_type = pcmk_controld_reply_resource;
+ reply_data.data.resource.node_state = msg_data;
- } else if (!strcmp(value, CRM_OP_PING)) {
- set_ping_data(&reply_data, msg_data);
+ } else if (!strcmp(value, CRM_OP_PING)) {
+ set_ping_data(&reply_data, msg_data);
- } else if (!strcmp(value, PCMK__CONTROLD_CMD_NODES)) {
- set_nodes_data(&reply_data, msg_data);
+ } else if (!strcmp(value, PCMK__CONTROLD_CMD_NODES)) {
+ set_nodes_data(&reply_data, msg_data);
- } else {
- crm_debug("Unrecognizable controller message: unknown command '%s'",
- value);
- status = CRM_EX_PROTOCOL;
- reply = NULL;
- }
+ } else {
+ crm_debug("Unrecognizable controller message: unknown command '%s'",
+ value);
+ status = CRM_EX_PROTOCOL;
}
+done:
pcmk__call_ipc_callback(api, pcmk_ipc_event_reply, status, &reply_data);
// Free any reply data that was allocated
--
1.8.3.1
From 5ae4101b60f8c0cd96eb2097a65a59aaa1750d73 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Fri, 17 Jul 2020 17:20:23 -0500
Subject: [PATCH 4/4] Log: fencer: don't log assertion if unable to create full
request reply
Previously, we would log an assertion and a warning if asked to create a reply
to a NULL request. However there is a possible sequence for this to happen:
- Some nodes are up and some down at cluster start-up
- One node is elected DC and schedules fencing of the down nodes
- Fencing is initiated for one of the down nodes
- One of the other down nodes comes up and is elected DC
- The fencing result comes back and all peers (including new DC) are notified
- New DC tries to create a notification for its client (the controller)
but doesn't know anything about the initial request
For now, just log a warning and drop the assertion. Longer term, maybe we
should synchronize in-flight request information when a fencer joins the
process group.
---
daemons/fenced/fenced_commands.c | 55 +++++++++++++++++++++++-----------------
1 file changed, 32 insertions(+), 23 deletions(-)
diff --git a/daemons/fenced/fenced_commands.c b/daemons/fenced/fenced_commands.c
index 05c5437..9c27d61 100644
--- a/daemons/fenced/fenced_commands.c
+++ b/daemons/fenced/fenced_commands.c
@@ -2336,22 +2336,8 @@ stonith_fence(xmlNode * msg)
xmlNode *
stonith_construct_reply(xmlNode * request, const char *output, xmlNode * data, int rc)
{
- int lpc = 0;
xmlNode *reply = NULL;
- const char *name = NULL;
- const char *value = NULL;
-
- const char *names[] = {
- F_STONITH_OPERATION,
- F_STONITH_CALLID,
- F_STONITH_CLIENTID,
- F_STONITH_CLIENTNAME,
- F_STONITH_REMOTE_OP_ID,
- F_STONITH_CALLOPTS
- };
-
- crm_trace("Creating a basic reply");
reply = create_xml_node(NULL, T_STONITH_REPLY);
crm_xml_add(reply, "st_origin", __FUNCTION__);
@@ -2359,16 +2345,39 @@ stonith_construct_reply(xmlNode * request, const char *output, xmlNode * data, i
crm_xml_add(reply, "st_output", output);
crm_xml_add_int(reply, F_STONITH_RC, rc);
- CRM_CHECK(request != NULL, crm_warn("Can't create a sane reply"); return reply);
- for (lpc = 0; lpc < DIMOF(names); lpc++) {
- name = names[lpc];
- value = crm_element_value(request, name);
- crm_xml_add(reply, name, value);
- }
+ if (request == NULL) {
+ /* Most likely, this is the result of a stonith operation that was
+ * initiated before we came up. Unfortunately that means we lack enough
+ * information to provide clients with a full result.
+ *
+ * @TODO Maybe synchronize this information at start-up?
+ */
+ crm_warn("Missing request information for client notifications for "
+ "operation with result %d (initiated before we came up?)", rc);
- if (data != NULL) {
- crm_trace("Attaching reply output");
- add_message_xml(reply, F_STONITH_CALLDATA, data);
+ } else {
+ const char *name = NULL;
+ const char *value = NULL;
+
+ const char *names[] = {
+ F_STONITH_OPERATION,
+ F_STONITH_CALLID,
+ F_STONITH_CLIENTID,
+ F_STONITH_CLIENTNAME,
+ F_STONITH_REMOTE_OP_ID,
+ F_STONITH_CALLOPTS
+ };
+
+ crm_trace("Creating a result reply with%s reply output (rc=%d)",
+ (data? "" : "out"), rc);
+ for (int lpc = 0; lpc < DIMOF(names); lpc++) {
+ name = names[lpc];
+ value = crm_element_value(request, name);
+ crm_xml_add(reply, name, value);
+ }
+ if (data != NULL) {
+ add_message_xml(reply, F_STONITH_CALLDATA, data);
+ }
}
return reply;
}
--
1.8.3.1

85
SOURCES/012-ipc_fix.patch Normal file
View File

@ -0,0 +1,85 @@
From f7389ac6f67804f20393951462a59a0b505dfe03 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Tue, 21 Jul 2020 16:41:18 -0500
Subject: [PATCH] Fix: executor: only send executor notifications to executor
clients
This bug has existed since Pacemaker Remote was first implemented, but was
hidden until crm_node -l/-p was recently modified to go through controller IPC,
because other command-line IPC API clients either fire-and-forget IPC requests
or merely count replies, rather than parse the content of replies.
Previously, when the executor sent notifications of results, it broadcast the
notification to all IPC clients. Normally this behavior makes sense, but for
the executor in particular, it may be running as pacemaker-remoted, in which
case its IPC clients include not only clients that connected to the executor
IPC, but clients that connected via proxy to other IPC APIs on the cluster node
hosting the remote connection.
With crm_node -l/-p, this meant that it was possible for an executor API
notification to arrive while crm_node was waiting for a controller IPC reply.
It would not find the information it needed and would report a protocol
violation error.
The fix is to send executor notifications only to executor clients.
---
daemons/execd/execd_commands.c | 9 +++++++++
daemons/execd/remoted_proxy.c | 5 +++++
include/crm/common/ipc_internal.h | 5 +++--
3 files changed, 17 insertions(+), 2 deletions(-)
diff --git a/daemons/execd/execd_commands.c b/daemons/execd/execd_commands.c
index aaf2976..685fcc7 100644
--- a/daemons/execd/execd_commands.c
+++ b/daemons/execd/execd_commands.c
@@ -507,6 +507,15 @@ send_client_notify(gpointer key, gpointer value, gpointer user_data)
crm_trace("Skipping notification to client without name");
return;
}
+ if (is_set(client->flags, pcmk__client_to_proxy)) {
+ /* We only want to notify clients of the executor IPC API. If we are
+ * running as Pacemaker Remote, we may have clients proxied to other
+ * IPC services in the cluster, so skip those.
+ */
+ crm_trace("Skipping executor API notification to %s IPC client",
+ client->name);
+ return;
+ }
rc = lrmd_server_send_notify(client, update_msg);
if (rc == pcmk_rc_ok) {
diff --git a/daemons/execd/remoted_proxy.c b/daemons/execd/remoted_proxy.c
index dda7eed..5c58de4 100644
--- a/daemons/execd/remoted_proxy.c
+++ b/daemons/execd/remoted_proxy.c
@@ -88,6 +88,11 @@ ipc_proxy_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid, const char *ipc
client->userdata = strdup(ipc_proxy->id);
client->name = crm_strdup_printf("proxy-%s-%d-%.8s", ipc_channel, client->pid, client->id);
+ /* Allow remote executor to distinguish between proxied local clients and
+ * actual executor API clients
+ */
+ set_bit(client->flags, pcmk__client_to_proxy);
+
g_hash_table_insert(ipc_clients, client->id, client);
msg = create_xml_node(NULL, T_LRMD_IPC_PROXY);
diff --git a/include/crm/common/ipc_internal.h b/include/crm/common/ipc_internal.h
index 6a1fcf3..91b3435 100644
--- a/include/crm/common/ipc_internal.h
+++ b/include/crm/common/ipc_internal.h
@@ -121,8 +121,9 @@ struct pcmk__remote_s {
};
enum pcmk__client_flags {
- pcmk__client_proxied = 0x00001, /* ipc_proxy code only */
- pcmk__client_privileged = 0x00002, /* root or cluster user */
+ pcmk__client_proxied = (1 << 0), // Remote client behind proxy
+ pcmk__client_privileged = (1 << 1), // root or cluster user
+ pcmk__client_to_proxy = (1 << 2), // Local client to be proxied
};
struct pcmk__client_s {
--
1.8.3.1

1072
SOURCES/013-pacemakerd.patch Normal file

File diff suppressed because it is too large Load Diff

1187
SOURCES/014-sbd.patch Normal file

File diff suppressed because it is too large Load Diff

123
SOURCES/015-cibsecret.patch Normal file
View File

@ -0,0 +1,123 @@
From 2e7a40570d6b21534ec0215ac5ebc174796cf17c Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Thu, 20 Aug 2020 10:02:20 -0500
Subject: [PATCH 1/2] Refactor: tools: rename function in cibsecret to be more
clear
It led me to initially misdiagnose a problem.
---
tools/cibsecret.in | 26 +++++++++++++-------------
1 file changed, 13 insertions(+), 13 deletions(-)
diff --git a/tools/cibsecret.in b/tools/cibsecret.in
index 9b74ba3..dabbfc0 100644
--- a/tools/cibsecret.in
+++ b/tools/cibsecret.in
@@ -162,28 +162,28 @@ check_env() {
}
# This must be called (and return success) before calling $rsh or $rcp_to_from
-get_live_nodes() {
- # Get a list of all cluster nodes
+get_live_peers() {
+ # Get a list of all other cluster nodes
GLN_ALL_NODES="$(crm_node -l | awk '{print $2}' | grep -v "$(uname -n)")"
# Make a list of those that respond to pings
if [ "$(id -u)" = "0" ] && which fping >/dev/null 2>&1; then
- LIVE_NODES=$(fping -a $GLN_ALL_NODES 2>/dev/null)
+ LIVE_NODES=$(fping -a $GLP_ALL_PEERS 2>/dev/null)
else
LIVE_NODES=""
- for GLN_NODE in $GLN_ALL_NODES; do \
- ping -c 2 -q "$GLN_NODE" >/dev/null 2>&1 &&
- LIVE_NODES="$LIVE_NODES $GLN_NODE"
+ for GLP_NODE in $GLP_ALL_PEERS; do \
+ ping -c 2 -q "$GLP_NODE" >/dev/null 2>&1 &&
+ LIVE_NODES="$LIVE_NODES $GLP_NODE"
done
fi
# Warn the user about any that didn't respond to pings
- GLN_DOWN="$( (for GLN_NODE in $LIVE_NODES $GLN_ALL_NODES; do echo "$GLN_NODE"; done) | sort | uniq -u)"
- if [ "$(echo "$GLN_DOWN" | wc -w)" = "1" ]; then
- warn "node $GLN_DOWN is down"
+ GLP_DOWN="$( (for GLP_NODE in $LIVE_NODES $GLP_ALL_PEERS; do echo "$GLP_NODE"; done) | sort | uniq -u)"
+ if [ "$(echo "$GLP_DOWN" | wc -w)" = "1" ]; then
+ warn "node $GLP_DOWN is down"
warn "you'll need to update it using \"$PROG sync\" later"
- elif [ -n "$GLN_DOWN" ]; then
- warn "nodes $(echo "$GLN_DOWN" | tr '\n' ' ')are down"
+ elif [ -n "$GLP_DOWN" ]; then
+ warn "nodes $(echo "$GLP_DOWN" | tr '\n' ' ')are down"
warn "you'll need to update them using \"$PROG sync\" later"
fi
@@ -235,7 +235,7 @@ scp_fun() {
# TODO: this procedure should be replaced with csync2
# provided that csync2 has already been configured
sync_files() {
- get_live_nodes || return
+ get_live_peers || return
info "syncing $LRM_CIBSECRETS to $(echo "$LIVE_NODES" | tr '\n' ' ') ..."
$rsh rm -rf "$LRM_CIBSECRETS" &&
$rsh mkdir -p "$(dirname "$LRM_CIBSECRETS")" &&
@@ -244,7 +244,7 @@ sync_files() {
sync_one() {
SO_FILE="$1"
- get_live_nodes || return
+ get_live_peers || return
info "syncing $SO_FILE to $(echo "$LIVE_NODES" | tr '\n' ' ') ..."
$rsh mkdir -p "$(dirname "$SO_FILE")" &&
if [ -f "$SO_FILE" ]; then
--
1.8.3.1
From 9c1517e6a681f35d62b4714e854b258c17ab5e59 Mon Sep 17 00:00:00 2001
From: Ken Gaillot <kgaillot@redhat.com>
Date: Thu, 20 Aug 2020 10:03:23 -0500
Subject: [PATCH 2/2] Fix: tools: properly detect local node name
cibsecret had two serious problems when generating a list of other nodes to
sync secrets to:
* It used `uname -n` to remove the local node from the list. If the local node
name is different from its uname, this could cause local secrets to be
removed from the local node rather than synced to other nodes.
* It removed not just the local node name, but any node name that contained
the local node name as a substring (e.g. "node1" and "node10"). This could
cause secrets to not be synced to such nodes.
Now, use `crm_node -n` to determine the local node name, check crm_node for
errors to get better error messages, and remove only the node name that matches
the local node name in its entirety.
---
tools/cibsecret.in | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/tools/cibsecret.in b/tools/cibsecret.in
index dabbfc0..568833c 100644
--- a/tools/cibsecret.in
+++ b/tools/cibsecret.in
@@ -163,8 +163,14 @@ check_env() {
# This must be called (and return success) before calling $rsh or $rcp_to_from
get_live_peers() {
+ # Get local node name
+ GLP_LOCAL_NODE="$(crm_node -n)"
+ [ $? -eq 0 ] || fatal "couldn't get local node name"
+
# Get a list of all other cluster nodes
- GLN_ALL_NODES="$(crm_node -l | awk '{print $2}' | grep -v "$(uname -n)")"
+ GLP_ALL_PEERS="$(crm_node -l)"
+ [ $? -eq 0 ] || fatal "couldn't determine cluster nodes"
+ GLP_ALL_PEERS="$(echo "$GLP_ALL_PEERS" | awk '{print $2}' | grep -v "^${GLP_LOCAL_NODE}$")"
# Make a list of those that respond to pings
if [ "$(id -u)" = "0" ] && which fping >/dev/null 2>&1; then
--
1.8.3.1

1341
SPECS/pacemaker.spec Normal file

File diff suppressed because it is too large Load Diff