Branch: b1_4

Move the MDC EA/cookie max size initialization later in the mount process,
until we have established our permanent connections.  This avoids needless
a bunch of client->OST connect+disconnect RPCs, and also fixes a bug where
the client didn't get the reply for the first disconnect, and then hit a
sanity check (target handle mismatch) on the server.

This will tie in directly to the bug 5707 fix, as that will additionally
store a "default" EA size to avoid allocating and sending a very large
(and otherwise useless) request buffer, just to have a place to store
the LOV EA for a MDS open replay.    Includes regression test.
b=5317
r=nathan (with minor fix to avoid the storage of lovdesc at all)

diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index 87e32a5..a2946fd 100644 (file)
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -30,6 +30,7 @@ tbd         Cluster File Systems, Inc. <info@clusterfs.com>
        - small ext3 extents cleanups and fixes (5733)
        - improved mballoc code, several small races and bugs fixed (5733, 5638)
        - kernel version 43: fix remove_suid bugs in both 2.4 and 2.6 (5695)
+       - avoid needless client->OST connect, fix handle mismatch (5317)
        * miscellania
        - service request history (4965)
        - put {ll,lov,osc}_async_page structs in a single slab (4699)

diff --git a/lustre/include/linux/lustre_mds.h b/lustre/include/linux/lustre_mds.h
index 0ebde73..20843f5 100644 (file)
--- a/lustre/include/linux/lustre_mds.h
+++ b/lustre/include/linux/lustre_mds.h
@@ -185,10 +185,9 @@ int mdc_enqueue(struct obd_export *exp,
                 void *cb_data);
 
 /* mdc/mdc_request.c */
-int mdc_init_ea_size(struct obd_device *obd, char *lov_name);
+int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp);
 int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
-                      struct obd_export *exp,
-                      struct lustre_md *md);
+                      struct obd_export *exp, struct lustre_md *md);
 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid);
 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
                 unsigned long valid, unsigned int ea_size,

diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index 17665b7..2fb9c66 100644 (file)
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -156,6 +156,7 @@ int client_obd_setup(struct obd_device *obddev, obd_count len, void *buf)
         class_import_put(imp);
 
         cli->cl_import = imp;
+        /* cli->cl_max_mds_{easize,cookiesize} updated by mdc_init_ea_size() */
         cli->cl_max_mds_easize = sizeof(struct lov_mds_md);
         cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie);
         cli->cl_sandev = to_kdev_t(0);

diff --git a/lustre/liblustre/super.c b/lustre/liblustre/super.c
index cc58ec5..1186004 100644 (file)
--- a/lustre/liblustre/super.c
+++ b/lustre/liblustre/super.c
@@ -1345,9 +1345,6 @@ llu_fsswop_mount(const char *source,
                 GOTO(out_free, err = -EINVAL);
         }
 
-        if (mdc_init_ea_size(obd, osc))
-                GOTO(out_free, err = -EINVAL);
-
         /* setup mdc */
         err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
         if (err) {
@@ -1378,6 +1375,8 @@ llu_fsswop_mount(const char *source,
         }
         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 
+        mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
+
         err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
         if (err) {
                 CERROR("cannot mds_connect: rc = %d\n", err);

diff --git a/lustre/llite/dcache.c b/lustre/llite/dcache.c
index c19e289..f29f732 100644 (file)
--- a/lustre/llite/dcache.c
+++ b/lustre/llite/dcache.c
@@ -41,7 +41,7 @@ static void ll_release(struct dentry *de)
         ENTRY;
         LASSERT(de != NULL);
         lld = ll_d2d(de);
-        if (lld == NULL) {
+        if (lld == NULL) { /* NFS copies the de->d_op methods (bug 4655) */
                 EXIT;
                 return;
         }

diff --git a/lustre/llite/llite_lib.c b/lustre/llite/llite_lib.c
index 7c73b6b..9086073 100644 (file)
--- a/lustre/llite/llite_lib.c
+++ b/lustre/llite/llite_lib.c
@@ -121,8 +121,6 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
                         CERROR("could not register mount in /proc/lustre");
         }
 
-        mdc_init_ea_size(obd, osc);
-
         err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
         if (err == -EBUSY) {
                 CERROR("An MDS (mdc %s) is performing recovery, of which this"
@@ -169,6 +167,8 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
         }
         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 
+        mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
+
         if (!ll_async_page_slab) {
                 ll_async_page_slab_size =
                         size_round(sizeof(struct ll_async_page)) +

diff --git a/lustre/mdc/mdc_request.c b/lustre/mdc/mdc_request.c
index 42faae7..ad0f9bd 100644 (file)
--- a/lustre/mdc/mdc_request.c
+++ b/lustre/mdc/mdc_request.c
@@ -648,7 +648,7 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
         case OBD_IOC_LLOG_PRINT: {
                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
                 rc = llog_ioctl(ctxt, cmd, data);
-                
+
                 GOTO(out, rc);
         }
 #endif
@@ -913,39 +913,35 @@ err_rpc_lock:
         RETURN(rc);
 }
 
-int mdc_init_ea_size(struct obd_device *obd, char *lov_name)
+/* Initialize the maximum LOV EA and cookie sizes.  This allows
+ * us to make MDS RPCs with large enough reply buffers to hold the
+ * maximum-sized (= maximum striped) EA and cookie without having to
+ * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
+int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp)
 {
+        struct obd_device *obd = mdc_exp->exp_obd;
         struct client_obd *cli = &obd->u.cli;
-        struct obd_device *lov_obd;
-        struct obd_export *exp;
-        struct lustre_handle conn;
+        struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
         struct lov_desc desc;
-        int valsize;
-        int rc;
-
-        lov_obd = class_name2obd(lov_name);
-        if (!lov_obd) {
-                CERROR("MDC cannot locate LOV %s!\n", lov_name);
-                RETURN(-ENOTCONN);
-        }
+        __u32 valsize = sizeof(desc);
+        int rc, size;
+        ENTRY;
 
-        rc = obd_connect(&conn, lov_obd, &obd->obd_uuid);
-        if (rc) {
-                CERROR("MDC failed connect to LOV %s (%d)\n", lov_name, rc);
+        rc = obd_get_info(lov_exp, strlen("lovdesc") + 1, "lovdesc",
+                          &valsize, &desc);
+        if (rc < 0)
                 RETURN(rc);
-        }
-        exp = class_conn2export(&conn);
 
-        valsize = sizeof(desc);
-        rc = obd_get_info(exp, strlen("lovdesc") + 1, "lovdesc",
-                          &valsize, &desc);
-        if (rc == 0) {
-                cli->cl_max_mds_easize = obd_size_diskmd(exp, NULL);
-                cli->cl_max_mds_cookiesize = desc.ld_tgt_count *
-                        sizeof(struct llog_cookie);
-        }
-        obd_disconnect(exp, 0);
-        RETURN(rc);
+        lsm.lsm_stripe_count = desc.ld_tgt_count;
+        size = obd_size_diskmd(lov_exp, &lsm);
+        if (cli->cl_max_mds_easize < size)
+                cli->cl_max_mds_easize = size;
+
+        size = desc.ld_tgt_count * sizeof(struct llog_cookie);
+        if (cli->cl_max_mds_cookiesize < size)
+                cli->cl_max_mds_cookiesize = size;
+
+        RETURN(0);
 }
 
 static int mdc_precleanup(struct obd_device *obd, int flags)

diff --git a/lustre/tests/replay-ost-single.sh b/lustre/tests/replay-ost-single.sh
index 8088d1b..da1f2a9 100755 (executable)
--- a/lustre/tests/replay-ost-single.sh
+++ b/lustre/tests/replay-ost-single.sh
@@ -56,7 +56,16 @@ setup() {
     start ost --reformat $OSTLCONFARGS
     [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
     start mds --reformat $MDSLCONFARGS
-    grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
+
+    if [ -z "`grep " $MOUNT " /proc/mounts`" ]; then
+       # test "-1" needed during initial client->OST connection
+       log "== test 00: target handle mismatch (bug 5317) === `date +%H:%M:%S`"
+
+       #define OBD_FAIL_OST_ALL_REPLY_NET       0x211
+       do_facet ost "sysctl -w lustre.fail_loc=0x80000211"
+
+       zconf_mount `hostname` $MOUNT && df $MOUNT && pass || error "mount fail"
+    fi
 }
 
 mkdir -p $DIR