nfs-utils: support for lifting grace period early

index 9dbe5d908336..828a6991f8e6 100644
@@ -42,6 +42,8 @@
 #define NSM_TIMEOUT	2
 #define NSM_MAX_TIMEOUT	120	/* don't make this too big */
 
+#define NLM_END_GRACE_FILE	"/proc/fs/lockd/nlm_end_grace"
+
 struct nsm_host {
 	struct nsm_host *	next;
 	char *			name;

@@ -450,6 +452,28 @@ retry:
 	return sock;
 }
 
+/* Inform the kernel that it's OK to lift lockd's grace period */
+static void
+nsm_lift_grace_period(void)
+{
+	int fd;
+
+	fd = open(NLM_END_GRACE_FILE, O_WRONLY);
+	if (fd < 0) {
+		/* Don't warn if file isn't present */
+		if (errno != ENOENT)
+			xlog(L_WARNING, "Unable to open %s: %m",
+				NLM_END_GRACE_FILE);
+		return;
+	}
+
+	if (write(fd, "Y", 1) < 0)
+		xlog(L_WARNING, "Unable to write to %s: %m", NLM_END_GRACE_FILE);
+
+	close(fd);
+	return;
+}
+
 int
 main(int argc, char **argv)
 {

@@ -534,6 +558,7 @@ usage:		fprintf(stderr,
 	(void)nsm_retire_monitored_hosts();
 	if (nsm_load_notify_list(smn_get_host) == 0) {
 		xlog(D_GENERAL, "No hosts to notify; exiting");
+		nsm_lift_grace_period();
 		return 0;
 	}
 

index bac678980938..352f029d55c0 100644
@@ -21,17 +21,15 @@
  * Explanation:
  *
  * This file contains the code to manage the sqlite backend database for the
- * clstated upcall daemon.
+ * nfsdcltrack usermodehelper upcall program.
  *
  * The main database is called main.sqlite and contains the following tables:
  *
  * parameters: simple key/value pairs for storing database info
  *
- * clients: one column containing a BLOB with the as sent by the client
- * 	    and a timestamp (in epoch seconds) of when the record was
- * 	    established
- *
- * FIXME: should we also record the fsid being accessed?
+ * clients: an "id" column containing a BLOB with the long-form clientid as
+ * 	    sent by the client, a "time" column containing a timestamp (in
+ * 	    epoch seconds) of when the record was last updated.
  */
 
 #ifdef HAVE_CONFIG_H

index 352f029d55c0..52fcaa4d7ca3 100644
@@ -50,10 +50,10 @@
 
 #include "xlog.h"
 
-#define CLD_SQLITE_SCHEMA_VERSION 1
+#define CLTRACK_SQLITE_LATEST_SCHEMA_VERSION 1
 
 /* in milliseconds */
-#define CLD_SQLITE_BUSY_TIMEOUT 10000
+#define CLTRACK_SQLITE_BUSY_TIMEOUT 10000
 
 /* private data structures */
 

@@ -111,7 +111,7 @@ sqlite_prepare_dbh(const char *topdir)
 		return ret;
 	}
 
-	ret = sqlite3_busy_timeout(dbh, CLD_SQLITE_BUSY_TIMEOUT);
+	ret = sqlite3_busy_timeout(dbh, CLTRACK_SQLITE_BUSY_TIMEOUT);
 	if (ret != SQLITE_OK) {
 		xlog(L_ERROR, "Unable to set sqlite busy timeout: %d", ret);
 		sqlite3_close(dbh);

@@ -156,7 +156,7 @@ sqlite_maindb_init(const char *topdir)
 	/* insert version into table -- ignore error if it fails */
 	ret = snprintf(buf, sizeof(buf),
 		       "INSERT OR IGNORE INTO parameters values (\"version\", "
-		       "\"%d\");", CLD_SQLITE_SCHEMA_VERSION);
+		       "\"%d\");", CLTRACK_SQLITE_CURRENT_SCHEMA_VERSION);
 	if (ret < 0) {
 		goto out_err;
 	} else if ((size_t)ret >= sizeof(buf)) {

@@ -189,10 +189,10 @@ sqlite_maindb_init(const char *topdir)
 
 	/* process SELECT result */
 	ret = sqlite3_column_int(stmt, 0);
-	if (ret != CLD_SQLITE_SCHEMA_VERSION) {
+	if (ret != CLTRACK_SQLITE_CURRENT_SCHEMA_VERSION) {
 		xlog(L_ERROR, "Unsupported database schema version! "
 			"Expected %d, got %d.",
-			CLD_SQLITE_SCHEMA_VERSION, ret);
+			CLTRACK_SQLITE_CURRENT_SCHEMA_VERSION, ret);
 		ret = -EINVAL;
 		goto out_err;
 	}

index 4334340197b1..f2813610be73 100644
@@ -241,7 +241,7 @@ cltrack_init(const char __attribute__((unused)) *unused)
 	}
 
 	/* set up storage db */
-	ret = sqlite_maindb_init(storagedir);
+	ret = sqlite_prepare_dbh(storagedir);
 	if (ret) {
 		xlog(L_ERROR, "Failed to init database: %d", ret);
 		/*

276 lines not on this page · raw message

index ebf04c3cdbaa..3713bfb66e2f 100644
@@ -21,7 +21,6 @@
 #define _SQLITE_H_
 
 int sqlite_prepare_dbh(const char *topdir);
-int sqlite_maindb_init(const char *topdir);
 int sqlite_insert_client(const unsigned char *clname, const size_t namelen);
 int sqlite_remove_client(const unsigned char *clname, const size_t namelen);
 int sqlite_check_client(const unsigned char *clname, const size_t namelen);

index dde2f7ff8621..e260e81b1722 100644
@@ -29,7 +29,10 @@
  *
  * clients: an "id" column containing a BLOB with the long-form clientid as
  * 	    sent by the client, a "time" column containing a timestamp (in
- * 	    epoch seconds) of when the record was last updated.
+ * 	    epoch seconds) of when the record was last updated, and a
+ * 	    "reclaim_complete" column containing a timestamp (in epoch seconds)
+ * 	    of when the last "create" operation came in for v4.1+ clients.
+ * 	    v4.0 clients should always have this set to 0.
  */
 
 #ifdef HAVE_CONFIG_H

@@ -50,7 +53,7 @@
 
 #include "xlog.h"
 
-#define CLTRACK_SQLITE_LATEST_SCHEMA_VERSION 1
+#define CLTRACK_SQLITE_LATEST_SCHEMA_VERSION 2
 
 /* in milliseconds */
 #define CLTRACK_SQLITE_BUSY_TIMEOUT 10000

@@ -120,6 +123,81 @@ out:
 	return ret;
 }
 
+static int
+sqlite_maindb_update_v1_to_v2(void)
+{
+	int ret, ret2;
+	char *err;
+
+	/* begin transaction */
+	ret = sqlite3_exec(dbh, "BEGIN EXCLUSIVE TRANSACTION;", NULL, NULL,
+				&err);
+	if (ret != SQLITE_OK) {
+		xlog(L_ERROR, "Unable to begin transaction: %s", err);
+		goto rollback;
+	}
+
+	/*
+	 * Check schema version again. This time, under an exclusive
+	 * transaction to guard against racing DB setup attempts
+	 */
+	ret = sqlite_query_schema_version();
+	switch (ret) {
+	case 1:
+		/* Still at v1 -- do conversion */
+		break;
+	case CLTRACK_SQLITE_LATEST_SCHEMA_VERSION:
+		/* Someone else raced in and set it up */
+		ret = 0;
+		goto rollback;
+	default:
+		/* Something went wrong -- fail! */
+		ret = -EINVAL;
+		goto rollback;
+	}
+
+	/* create v2 clients table */
+	ret = sqlite3_exec(dbh, "ALTER TABLE clients ADD COLUMN "
+				"reclaim_complete INTEGER;",
+				NULL, NULL, &err);
+	if (ret != SQLITE_OK) {
+		xlog(L_ERROR, "Unable to update clients table: %s", err);
+		goto rollback;
+	}
+
+	ret = snprintf(buf, sizeof(buf), "UPDATE parameters SET value = %d "
+			"WHERE key = \"version\";",
+			CLTRACK_SQLITE_LATEST_SCHEMA_VERSION);
+	if (ret < 0) {
+		xlog(L_ERROR, "sprintf failed!");
+		goto rollback;
+	} else if ((size_t)ret >= sizeof(buf)) {
+		xlog(L_ERROR, "sprintf output too long! (%d chars)", ret);
+		ret = -EINVAL;
+		goto rollback;
+	}
+
+	ret = sqlite3_exec(dbh, (const char *)buf, NULL, NULL, &err);
+	if (ret != SQLITE_OK) {
+		xlog(L_ERROR, "Unable to update schema version: %s", err);
+		goto rollback;
+	}
+
+	ret = sqlite3_exec(dbh, "COMMIT TRANSACTION;", NULL, NULL, &err);
+	if (ret != SQLITE_OK) {
+		xlog(L_ERROR, "Unable to commit transaction: %s", err);
+		goto rollback;
+	}
+out:
+	sqlite3_free(err);
+	return ret;
+rollback:
+	ret2 = sqlite3_exec(dbh, "ROLLBACK TRANSACTION;", NULL, NULL, &err);
+	if (ret2 != SQLITE_OK)
+		xlog(L_ERROR, "Unable to rollback transaction: %s", err);
+	goto out;
+}
+
 /*
  * Start an exclusive transaction and recheck the DB schema version. If it's
  * still zero (indicating a new database) then set it up. If that all works,

@@ -127,9 +205,9 @@ out:
  * transaction. On any error, rollback the transaction.
  */
 int
-sqlite_maindb_init_v1(void)
+sqlite_maindb_init_v2(void)
 {
-	int ret;
+	int ret, ret2;
 	char *err = NULL;
 
 	/* Start a transaction */

@@ -169,7 +247,7 @@ sqlite_maindb_init_v1(void)
 
 	/* create the "clients" table */
 	ret = sqlite3_exec(dbh, "CREATE TABLE clients (id BLOB PRIMARY KEY, "
-				"time INTEGER);",
+				"time INTEGER, reclaim_complete INTEGER);",
 				NULL, NULL, &err);
 	if (ret != SQLITE_OK) {
 		xlog(L_ERROR, "Unable to create clients table: %s", err);

@@ -207,7 +285,9 @@ out:
 
 rollback:
 	/* Attempt to rollback the transaction */
-	sqlite3_exec(dbh, "ROLLBACK TRANSACTION;", NULL, NULL, &err);
+	ret2 = sqlite3_exec(dbh, "ROLLBACK TRANSACTION;", NULL, NULL, &err);
+	if (ret2 != SQLITE_OK)
+		xlog(L_ERROR, "Unable to rollback transaction: %s", err);
 	goto out;
 }
 

@@ -255,9 +335,15 @@ sqlite_prepare_dbh(const char *topdir)
 		/* DB is already set up. Do nothing */
 		ret = 0;
 		break;
+	case 1:
+		/* Old DB -- update to new schema */
+		ret = sqlite_maindb_update_v1_to_v2();
+		if (ret)
+			goto out_close;
+		break;
 	case 0:
 		/* Query failed -- try to set up new DB */
-		ret = sqlite_maindb_init_v1();
+		ret = sqlite_maindb_init_v2();
 		if (ret)
 			goto out_close;
 		break;

@@ -289,7 +375,7 @@ sqlite_insert_client(const unsigned char *clname, const size_t namelen)
 	sqlite3_stmt *stmt = NULL;
 
 	ret = sqlite3_prepare_v2(dbh, "INSERT OR REPLACE INTO clients VALUES "
-				      "(?, strftime('%s', 'now'));", -1,
+					"(?, strftime('%s', 'now'), 0);", -1,
 					&stmt, NULL);
 	if (ret != SQLITE_OK) {
 		xlog(L_ERROR, "%s: insert statement prepare failed: %s",

index f2813610be73..cbc98c7edc11 100644
@@ -37,6 +37,7 @@
 #include <libgen.h>
 #include <sys/inotify.h>
 #include <dirent.h>
+#include <limits.h>
 #ifdef HAVE_SYS_CAPABILITY_H
 #include <sys/prctl.h>
 #include <sys/capability.h>

@@ -254,6 +255,30 @@ cltrack_init(const char __attribute__((unused)) *unused)
 	return ret;
 }
 
+/*
+ * Fetch the contents of the NFSDCLTRACK_CLIENT_MINORVERSION env var. If
+ * it's not set or there is an error converting it to an unsigned int
+ * then return 0 (since the minorversion isn't reliable at that point).
+ */
+static unsigned int
+cltrack_get_minorvers(void)
+{
+	unsigned long minorvers;
+	char *end;
+	char *minorvers_str = getenv("NFSDCLTRACK_CLIENT_MINORVERSION");
+
+	if (!minorvers_str)
+		return 0;
+
+	errno = 0;
+	minorvers = strtoul(minorvers_str, &end, 0);
+	/* Problem converting or value is too large? */
+	if (errno || minorvers > UINT_MAX)
+		return 0;
+
+	return (unsigned int)minorvers;
+}
+
 static int
 cltrack_create(const char *id)
 {

@@ -270,7 +295,7 @@ cltrack_create(const char *id)
 	if (len < 0)
 		return (int)len;
 
-	ret = sqlite_insert_client(blob, len);
+	ret = sqlite_insert_client(blob, len, cltrack_get_minorvers());
 
 	return ret ? -EREMOTEIO : ret;
 }

@@ -323,7 +348,7 @@ cltrack_check_legacy(const unsigned char *blob, const ssize_t len)
 	}
 
 	/* Dir exists, try to insert record into db */
-	ret = sqlite_insert_client(blob, len);
+	ret = sqlite_insert_client(blob, len, 0);
 	if (ret) {
 		xlog(D_GENERAL, "Failed to insert client: %d", ret);
 		return -EREMOTEIO;

index e260e81b1722..01084bb6c4d8 100644
@@ -369,14 +369,22 @@ out_close:
  * Returns a non-zero sqlite error code, or SQLITE_OK (aka 0)
  */
 int
-sqlite_insert_client(const unsigned char *clname, const size_t namelen)
+sqlite_insert_client(const unsigned char *clname, const size_t namelen,
+			const unsigned int minorvers)
 {
 	int ret;
 	sqlite3_stmt *stmt = NULL;
 
-	ret = sqlite3_prepare_v2(dbh, "INSERT OR REPLACE INTO clients VALUES "
-					"(?, strftime('%s', 'now'), 0);", -1,
-					&stmt, NULL);
+	if (minorvers == 0)
+		ret = sqlite3_prepare_v2(dbh, "INSERT OR REPLACE INTO clients "
+				"VALUES (?, strftime('%s', 'now'), 0);", -1,
+				&stmt, NULL);
+	else
+		ret = sqlite3_prepare_v2(dbh, "INSERT OR REPLACE INTO clients "
+				"VALUES (?, strftime('%s', 'now'), "
+				"strftime('%s', 'now'));", -1,
+				&stmt, NULL);
+
 	if (ret != SQLITE_OK) {
 		xlog(L_ERROR, "%s: insert statement prepare failed: %s",
 			__func__, sqlite3_errmsg(dbh));

index 3713bfb66e2f..e9cc84cbb294 100644
@@ -21,7 +21,8 @@
 #define _SQLITE_H_
 
 int sqlite_prepare_dbh(const char *topdir);
-int sqlite_insert_client(const unsigned char *clname, const size_t namelen);
+int sqlite_insert_client(const unsigned char *clname, const size_t namelen,
+				const unsigned int minorvers);
 int sqlite_remove_client(const unsigned char *clname, const size_t namelen);
 int sqlite_check_client(const unsigned char *clname, const size_t namelen);
 int sqlite_remove_unreclaimed(const time_t grace_start);

index cbc98c7edc11..f500216e6021 100644
@@ -50,6 +50,8 @@
 #define CLD_DEFAULT_STORAGEDIR NFS_STATEDIR "/nfsdcltrack"
 #endif
 
+#define NFSD_END_GRACE_FILE "/proc/fs/nfsd/v4_end_grace"
+
 /* defined by RFC 3530 */
 #define NFS4_OPAQUE_LIMIT	1024
 

@@ -211,6 +213,64 @@ cltrack_set_caps(void)
 	return ret;
 }
 
+/* Inform the kernel that it's OK to lift nfsd's grace period */
+static void
+cltrack_lift_grace_period(void)
+{
+	int fd;
+
+	fd = open(NFSD_END_GRACE_FILE, O_WRONLY);
+	if (fd < 0) {
+		/* Don't warn if file isn't present */
+		if (errno != ENOENT)
+			xlog(L_WARNING, "Unable to open %s: %m",
+				NFSD_END_GRACE_FILE);
+		return;
+	}
+
+	if (write(fd, "Y", 1) < 0)
+		xlog(L_WARNING, "Unable to write to %s: %m",
+				NFSD_END_GRACE_FILE);
+
+	close(fd);
+	return;
+}
+
+/*
+ * Fetch the contents of the NFSDCLTRACK_GRACE_START env var. If it's not set
+ * or there's an error converting it to time_t, then return LONG_MAX.
+ */
+static time_t
+cltrack_get_grace_start(void)
+{
+	time_t grace_start;
+	char *end;
+	char *grace_start_str = getenv("NFSDCLTRACK_GRACE_START");
+
+	if (!grace_start_str)
+		return LONG_MAX;
+
+	errno = 0;
+	grace_start = strtol(grace_start_str, &end, 0);
+	/* Problem converting or value is too large? */
+	if (errno)
+		return LONG_MAX;
+
+	return grace_start;
+}
+
+static bool
+cltrack_reclaims_complete(void)
+{
+	time_t grace_start = cltrack_get_grace_start();
+
+	/* Don't query DB if we didn't get a valid boot time */
+	if (grace_start == LONG_MAX)
+		return false;
+
+	return !sqlite_query_reclaiming(grace_start);
+}
+
 static int
 cltrack_init(const char __attribute__((unused)) *unused)
 {

@@ -251,7 +311,11 @@ cltrack_init(const char __attribute__((unused)) *unused)
 		 * stop upcalling until the problem is resolved.
 		 */
 		ret = -EACCES;
+	} else {
+		if (cltrack_reclaims_complete())
+			cltrack_lift_grace_period();
 	}
+
 	return ret;
 }
 

@@ -284,6 +348,7 @@ cltrack_create(const char *id)
 {
 	int ret;
 	ssize_t len;
+	unsigned int minorvers;
 
 	xlog(D_GENERAL, "%s: create client record.", __func__);
 

@@ -295,8 +360,18 @@ cltrack_create(const char *id)
 	if (len < 0)
 		return (int)len;
 
-	ret = sqlite_insert_client(blob, len, cltrack_get_minorvers());
+	minorvers = cltrack_get_minorvers();
+	ret = sqlite_insert_client(blob, len, minorvers);
+	if (ret)
+		goto out;
+
+	/* No point in querying the DB if this client was v4.0 */
+	if (minorvers == 0)
+		goto out;
 
+	if (cltrack_reclaims_complete())
+		cltrack_lift_grace_period();
+out:
 	return ret ? -EREMOTEIO : ret;
 }
 

index 01084bb6c4d8..17437e48ffaa 100644
@@ -547,3 +547,43 @@ sqlite_remove_unreclaimed(time_t grace_start)
 	sqlite3_free(err);
 	return ret;
 }
+
+/*
+ * Are there any clients that are possibly still reclaiming? Return a positive
+ * integer (usually number of clients) if so. If not, then return 0. On any
+ * error, return non-zero.
+ */
+int
+sqlite_query_reclaiming(const time_t grace_start)
+{
+	int ret;
+	sqlite3_stmt *stmt = NULL;
+
+	ret = sqlite3_prepare_v2(dbh, "SELECT count(*) FROM clients WHERE "
+				      "reclaim_complete < ?", -1, &stmt, NULL);
+	if (ret != SQLITE_OK) {
+		xlog(L_ERROR, "%s: unable to prepare select statement: %s",
+				__func__, sqlite3_errstr(ret));
+		return ret;
+	}
+
+	ret = sqlite3_bind_int64(stmt, 1, (sqlite3_int64)grace_start);
+	if (ret != SQLITE_OK) {
+		xlog(L_ERROR, "%s: bind int64 failed: %s",
+				__func__, sqlite3_errstr(ret));
+		return ret;
+	}
+
+	ret = sqlite3_step(stmt);
+	if (ret != SQLITE_ROW) {
+		xlog(L_ERROR, "%s: unexpected return code from select: %s",
+				__func__, sqlite3_errstr(ret));
+		return ret;
+	}
+
+	ret = sqlite3_column_int(stmt, 0);
+	sqlite3_finalize(stmt);
+	xlog(D_GENERAL, "%s: there are %d clients that have not completed "
+			"reclaim", __func__, ret);
+	return ret;
+}

index e9cc84cbb294..0e1939198367 100644
@@ -26,5 +26,6 @@ int sqlite_insert_client(const unsigned char *clname, const size_t namelen,
 int sqlite_remove_client(const unsigned char *clname, const size_t namelen);
 int sqlite_check_client(const unsigned char *clname, const size_t namelen);
 int sqlite_remove_unreclaimed(const time_t grace_start);
+int sqlite_query_reclaiming(const time_t grace_start);
 
 #endif /* _SQLITE_H */