dlm: remove shared message stub for recovery
David Teigland [Mon, 4 Apr 2011 20:19:59 +0000 (15:19 -0500)]
kmalloc a stub message struct during recovery instead of sharing the
struct in the lockspace.  This leaves the lockspace stub_ms only for
faking downconvert replies, where it is never modified and sharing
is not a problem.

Also improve the debug messages in the same recovery function.

Signed-off-by: David Teigland <teigland@redhat.com>

fs/dlm/dlm_internal.h
fs/dlm/lock.c

index 6a92478..0262451 100644 (file)
@@ -209,6 +209,7 @@ struct dlm_args {
 #define DLM_IFL_WATCH_TIMEWARN 0x00400000
 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000
 #define DLM_IFL_DEADLOCK_CANCEL        0x01000000
+#define DLM_IFL_STUB_MS                0x02000000 /* magic number for m_flags */
 #define DLM_IFL_USER           0x00000001
 #define DLM_IFL_ORPHAN         0x00000002
 
index e3c8641..8122779 100644 (file)
@@ -1037,10 +1037,10 @@ static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
        struct dlm_ls *ls = lkb->lkb_resource->res_ls;
        int error;
 
-       if (ms != &ls->ls_stub_ms)
+       if (ms->m_flags != DLM_IFL_STUB_MS)
                mutex_lock(&ls->ls_waiters_mutex);
        error = _remove_from_waiters(lkb, ms->m_type, ms);
-       if (ms != &ls->ls_stub_ms)
+       if (ms->m_flags != DLM_IFL_STUB_MS)
                mutex_unlock(&ls->ls_waiters_mutex);
        return error;
 }
@@ -1462,14 +1462,8 @@ static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
    ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
    compatible with other granted locks */
 
-static void munge_demoted(struct dlm_lkb *lkb, struct dlm_message *ms)
+static void munge_demoted(struct dlm_lkb *lkb)
 {
-       if (ms->m_type != DLM_MSG_CONVERT_REPLY) {
-               log_print("munge_demoted %x invalid reply type %d",
-                         lkb->lkb_id, ms->m_type);
-               return;
-       }
-
        if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
                log_print("munge_demoted %x invalid modes gr %d rq %d",
                          lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
@@ -2966,9 +2960,9 @@ static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
        /* down conversions go without a reply from the master */
        if (!error && down_conversion(lkb)) {
                remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
+               r->res_ls->ls_stub_ms.m_flags = DLM_IFL_STUB_MS;
                r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
                r->res_ls->ls_stub_ms.m_result = 0;
-               r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
                __receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
        }
 
@@ -3156,6 +3150,9 @@ static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
 
 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
 {
+       if (ms->m_flags == DLM_IFL_STUB_MS)
+               return;
+
        lkb->lkb_sbflags = ms->m_sbflags;
        lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
                         (ms->m_flags & 0x0000FFFF);
@@ -3698,7 +3695,7 @@ static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
                /* convert was queued on remote master */
                receive_flags_reply(lkb, ms);
                if (is_demoted(lkb))
-                       munge_demoted(lkb, ms);
+                       munge_demoted(lkb);
                del_lkb(r, lkb);
                add_lkb(r, lkb, DLM_LKSTS_CONVERT);
                add_timeout(lkb);
@@ -3708,7 +3705,7 @@ static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
                /* convert was granted on remote master */
                receive_flags_reply(lkb, ms);
                if (is_demoted(lkb))
-                       munge_demoted(lkb, ms);
+                       munge_demoted(lkb);
                grant_lock_pc(r, lkb, ms);
                queue_cast(r, lkb, 0);
                break;
@@ -4082,15 +4079,17 @@ void dlm_receive_buffer(union dlm_packet *p, int nodeid)
        dlm_put_lockspace(ls);
 }
 
-static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
+static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb,
+                                  struct dlm_message *ms_stub)
 {
        if (middle_conversion(lkb)) {
                hold_lkb(lkb);
-               ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
-               ls->ls_stub_ms.m_result = -EINPROGRESS;
-               ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-               ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
-               _receive_convert_reply(lkb, &ls->ls_stub_ms);
+               memset(ms_stub, 0, sizeof(struct dlm_message));
+               ms_stub->m_flags = DLM_IFL_STUB_MS;
+               ms_stub->m_type = DLM_MSG_CONVERT_REPLY;
+               ms_stub->m_result = -EINPROGRESS;
+               ms_stub->m_header.h_nodeid = lkb->lkb_nodeid;
+               _receive_convert_reply(lkb, ms_stub);
 
                /* Same special case as in receive_rcom_lock_args() */
                lkb->lkb_grmode = DLM_LOCK_IV;
@@ -4131,13 +4130,27 @@ static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb)
 void dlm_recover_waiters_pre(struct dlm_ls *ls)
 {
        struct dlm_lkb *lkb, *safe;
+       struct dlm_message *ms_stub;
        int wait_type, stub_unlock_result, stub_cancel_result;
 
+       ms_stub = kmalloc(GFP_KERNEL, sizeof(struct dlm_message));
+       if (!ms_stub) {
+               log_error(ls, "dlm_recover_waiters_pre no mem");
+               return;
+       }
+
        mutex_lock(&ls->ls_waiters_mutex);
 
        list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
-               log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
-                         lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
+
+               /* exclude debug messages about unlocks because there can be so
+                  many and they aren't very interesting */
+
+               if (lkb->lkb_wait_type != DLM_MSG_UNLOCK) {
+                       log_debug(ls, "recover_waiter %x nodeid %d "
+                                 "msg %d to %d", lkb->lkb_id, lkb->lkb_nodeid,
+                                 lkb->lkb_wait_type, lkb->lkb_wait_nodeid);
+               }
 
                /* all outstanding lookups, regardless of destination  will be
                   resent after recovery is done */
@@ -4183,26 +4196,28 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
                        break;
 
                case DLM_MSG_CONVERT:
-                       recover_convert_waiter(ls, lkb);
+                       recover_convert_waiter(ls, lkb, ms_stub);
                        break;
 
                case DLM_MSG_UNLOCK:
                        hold_lkb(lkb);
-                       ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
-                       ls->ls_stub_ms.m_result = stub_unlock_result;
-                       ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-                       ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
-                       _receive_unlock_reply(lkb, &ls->ls_stub_ms);
+                       memset(ms_stub, 0, sizeof(struct dlm_message));
+                       ms_stub->m_flags = DLM_IFL_STUB_MS;
+                       ms_stub->m_type = DLM_MSG_UNLOCK_REPLY;
+                       ms_stub->m_result = stub_unlock_result;
+                       ms_stub->m_header.h_nodeid = lkb->lkb_nodeid;
+                       _receive_unlock_reply(lkb, ms_stub);
                        dlm_put_lkb(lkb);
                        break;
 
                case DLM_MSG_CANCEL:
                        hold_lkb(lkb);
-                       ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
-                       ls->ls_stub_ms.m_result = stub_cancel_result;
-                       ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-                       ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
-                       _receive_cancel_reply(lkb, &ls->ls_stub_ms);
+                       memset(ms_stub, 0, sizeof(struct dlm_message));
+                       ms_stub->m_flags = DLM_IFL_STUB_MS;
+                       ms_stub->m_type = DLM_MSG_CANCEL_REPLY;
+                       ms_stub->m_result = stub_cancel_result;
+                       ms_stub->m_header.h_nodeid = lkb->lkb_nodeid;
+                       _receive_cancel_reply(lkb, ms_stub);
                        dlm_put_lkb(lkb);
                        break;
 
@@ -4213,6 +4228,7 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
                schedule();
        }
        mutex_unlock(&ls->ls_waiters_mutex);
+       kfree(ms_stub);
 }
 
 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
@@ -4277,8 +4293,8 @@ int dlm_recover_waiters_post(struct dlm_ls *ls)
                ou = is_overlap_unlock(lkb);
                err = 0;
 
-               log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
-                         lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
+               log_debug(ls, "recover_waiter %x nodeid %d msg %d r_nodeid %d",
+                         lkb->lkb_id, lkb->lkb_nodeid, mstype, r->res_nodeid);
 
                /* At this point we assume that we won't get a reply to any
                   previous op or overlap op on this lock.  First, do a big