windows-misc-fix-20061005

still tracking down issues when large numbers of temporary files
are created and deleted in the same directory from multiple
machines.

VNOVNODE must return CM_ERROR_BADFD

ensure that all calls to cm_GetCallback are made from cm_SyncOp.

add CM_SCACHESYNC_FORCECB logic to cm_SyncOp so that it can be
removed from cm_GetAccess

don't mix CM_SCACHESYNC_NEEDCALLBACK with CM_SCACHESYNC_STOREDATA
or CM_SCACHESYNC_FETCHDATA

diff --git a/src/WINNT/afsd/cm_access.c b/src/WINNT/afsd/cm_access.c
index e2cdcad..e5a1c12 100644 (file)
--- a/src/WINNT/afsd/cm_access.c
+++ b/src/WINNT/afsd/cm_access.c
@@ -124,9 +124,9 @@ long cm_GetAccessRights(struct cm_scache *scp, struct cm_user *userp,
     /* first, start by finding out whether we have a directory or something
      * else, so we can find what object's ACL we need.
      */
-    if (!cm_HaveCallback(scp)) {
+    if (scp->fileType == CM_SCACHETYPE_DIRECTORY || !cm_HaveCallback(scp)) {
        code = cm_SyncOp(scp, NULL, userp, reqp, 0,
-                     CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
+                        CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS | CM_SCACHESYNC_FORCECB);
        if (code) 
            return code;
 
@@ -152,16 +152,24 @@ long cm_GetAccessRights(struct cm_scache *scp, struct cm_user *userp,
            code = cm_SyncOp(aclScp, NULL, userp, reqp, 0,
                              CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
            if (!code) {
-               code = cm_GetCallback(aclScp, userp, reqp, 1);
-               cm_SyncOpDone(aclScp, NULL, CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
+#if 0
+               /* cm_GetCallback was called by cm_SyncOp */
+               code = cm_GetCallback(aclScp, userp, reqp, 1); 
+#endif
+               cm_SyncOpDone(aclScp, NULL, 
+                             CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS | CM_SCACHESYNC_FORCECB);
            }
            lock_ReleaseMutex(&aclScp->mx);
        }
         cm_ReleaseSCache(aclScp);
         lock_ObtainMutex(&scp->mx);
-    } else if (!got_cb) {
+    }
+#if 0    
+    else if (!got_cb) {
+       /* cm_GetCallback was called by cm_SyncOp */
        code = cm_GetCallback(scp, userp, reqp, 1);
     }
+#endif
 
   _done:
     if (got_cb)

diff --git a/src/WINNT/afsd/cm_callback.c b/src/WINNT/afsd/cm_callback.c
index 36fbbb9..2cf0a17 100644 (file)
--- a/src/WINNT/afsd/cm_callback.c
+++ b/src/WINNT/afsd/cm_callback.c
@@ -1614,7 +1614,7 @@ void cm_EndCallbackGrantingCall(cm_scache_t *scp, cm_callbackRequest_t *cbrp,
 long cm_GetCallback(cm_scache_t *scp, struct cm_user *userp,
                     struct cm_req *reqp, long flags)
 {
-    long code;
+    long code = 0;
     cm_conn_t *connp = NULL;
     AFSFetchStatus afsStatus;
     AFSVolSync volSync;
@@ -1624,6 +1624,7 @@ long cm_GetCallback(cm_scache_t *scp, struct cm_user *userp,
     int mustCall;
     cm_fid_t sfid;
     struct rx_connection * callp = NULL;
+    int syncop_done = 0;
 
     osi_Log4(afsd_logp, "GetCallback scp 0x%p cell %d vol %d flags %lX", 
              scp, scp->fid.cell, scp->fid.volume, flags);
@@ -1670,23 +1671,20 @@ long cm_GetCallback(cm_scache_t *scp, struct cm_user *userp,
     mustCall = (flags & 1);
     cm_AFSFidFromFid(&tfid, &scp->fid);
     while (1) {
-        if (!mustCall && cm_HaveCallback(scp)) {
-            osi_Log3(afsd_logp, "GetCallback Complete scp 0x%p cell %d vol %d", 
-                      scp, scp->fid.cell, scp->fid.volume);
-            return 0;
-        }
+        if (!mustCall && cm_HaveCallback(scp))
+           break;
 
         /* turn off mustCall, since it has now forced us past the check above */
         mustCall = 0;
 
-       /* 20060929 jaltman - We are being called from within cm_SyncOp.
-        * if we call cm_SyncOp again and another thread has attempted
-        * to obtain current status CM_SCACHEFLAG_WAITING will be set
-        * and we will deadlock.  
-        */
         /* otherwise, we have to make an RPC to get the status */
-        cm_SyncOp(scp, NULL, userp, reqp, 0, 
-                 CM_SCACHESYNC_FETCHSTATUS | CM_SCACHESYNC_GETCALLBACK);
+       if (!syncop_done) {
+           code = cm_SyncOp(scp, NULL, userp, reqp, 0, 
+                            CM_SCACHESYNC_FETCHSTATUS | CM_SCACHESYNC_GETCALLBACK);
+           if (code)
+               break;
+           syncop_done = 1;
+       }
         cm_StartCallbackGrantingCall(scp, &cbr);
         sfid = scp->fid;
         lock_ReleaseMutex(&scp->mx);
@@ -1721,17 +1719,25 @@ long cm_GetCallback(cm_scache_t *scp, struct cm_user *userp,
         } else {
             cm_EndCallbackGrantingCall(NULL, &cbr, NULL, 0);
         }
-       /* 20060929 jaltman - don't deadlock */
-        cm_SyncOpDone(scp, NULL, CM_SCACHESYNC_FETCHSTATUS | CM_SCACHESYNC_GETCALLBACK);
-
-        /* now check to see if we got an error */
-        if (code) {
-            osi_Log2(afsd_logp, "GetCallback Failed code 0x%x scp 0x%p -->",code, scp);
-            osi_Log4(afsd_logp, "            cell %u vol %u vn %u uniq %u",
-                     scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique);
-            return code;
-        }
+
+        /* if we got an error, return to caller */
+        if (code)
+           break;
+    }
+
+    if (syncop_done)
+       cm_SyncOpDone(scp, NULL, CM_SCACHESYNC_FETCHSTATUS | CM_SCACHESYNC_GETCALLBACK);
+    
+    if (code) {
+       osi_Log2(afsd_logp, "GetCallback Failed code 0x%x scp 0x%p -->",code, scp);
+       osi_Log4(afsd_logp, "            cell %u vol %u vn %u uniq %u",
+                scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique);
+    } else {
+       osi_Log3(afsd_logp, "GetCallback Complete scp 0x%p cell %d vol %d", 
+                 scp, scp->fid.cell, scp->fid.volume);
     }
+
+    return code;
 }
 
 /* called periodically by cm_daemon to shut down use of expired callbacks */

diff --git a/src/WINNT/afsd/cm_conn.c b/src/WINNT/afsd/cm_conn.c
index 339dcee..b61bf87 100644 (file)
--- a/src/WINNT/afsd/cm_conn.c
+++ b/src/WINNT/afsd/cm_conn.c
@@ -75,23 +75,23 @@ void cm_InitConn(void)
            code = RegQueryValueEx(parmKey, "ConnDeadTimeout", NULL, NULL,
                                    (BYTE *) &dwValue, &dummyLen);
            if (code == ERROR_SUCCESS)
-                ConnDeadtimeout = dwValue;
+                ConnDeadtimeout = (unsigned short)dwValue;
 
            dummyLen = sizeof(DWORD);
            code = RegQueryValueEx(parmKey, "HardDeadTimeout", NULL, NULL,
                                    (BYTE *) &dwValue, &dummyLen);
            if (code == ERROR_SUCCESS)
-                HardDeadtimeout = dwValue;
+                HardDeadtimeout = (unsigned short)dwValue;
            afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
            RegCloseKey(parmKey);
        }
 
        afsi_log("lanmanworkstation : SessTimeout %d", RDRtimeout);
        if (ConnDeadtimeout == 0)
-           ConnDeadtimeout = RDRtimeout / 2;
+           ConnDeadtimeout = (unsigned short) (RDRtimeout / 2);
        afsi_log("ConnDeadTimeout is %d", ConnDeadtimeout);
        if (HardDeadtimeout == 0)
-           HardDeadtimeout = RDRtimeout;
+           HardDeadtimeout = (unsigned short) RDRtimeout;
        afsi_log("HardDeadTimeout is %d", HardDeadtimeout);
 
        osi_EndOnce(&once);

diff --git a/src/WINNT/afsd/cm_dcache.c b/src/WINNT/afsd/cm_dcache.c
index a36e43b..805f1a8 100644 (file)
--- a/src/WINNT/afsd/cm_dcache.c
+++ b/src/WINNT/afsd/cm_dcache.c
@@ -768,8 +768,7 @@ long cm_SetupStoreBIOD(cm_scache_t *scp, osi_hyper_t *inOffsetp, long inSize,
             lock_ObtainMutex(&bufp->mx);
             lock_ObtainMutex(&scp->mx);
 
-            flags = CM_SCACHESYNC_NEEDCALLBACK
-                | CM_SCACHESYNC_GETSTATUS
+            flags = CM_SCACHESYNC_GETSTATUS
                     | CM_SCACHESYNC_STOREDATA
                         | CM_SCACHESYNC_BUFLOCKED;
             code = cm_SyncOp(scp, bufp, userp, reqp, 0, flags); 
@@ -827,8 +826,7 @@ long cm_SetupStoreBIOD(cm_scache_t *scp, osi_hyper_t *inOffsetp, long inSize,
     thyper.HighPart = 0;
     scanEnd = LargeIntegerAdd(scanStart, thyper);
 
-    flags = CM_SCACHESYNC_NEEDCALLBACK
-        | CM_SCACHESYNC_GETSTATUS
+    flags = CM_SCACHESYNC_GETSTATUS
         | CM_SCACHESYNC_STOREDATA
         | CM_SCACHESYNC_BUFLOCKED
         | CM_SCACHESYNC_NOWAIT;
@@ -1111,8 +1109,7 @@ long cm_SetupFetchBIOD(cm_scache_t *scp, osi_hyper_t *offsetp,
             break;
         }
 
-        flags = CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_FETCHDATA
-            | CM_SCACHESYNC_BUFLOCKED;
+        flags = CM_SCACHESYNC_FETCHDATA | CM_SCACHESYNC_BUFLOCKED;
         if (!isFirst) 
             flags |= CM_SCACHESYNC_NOWAIT;
 
@@ -1202,42 +1199,47 @@ void cm_ReleaseBIOD(cm_bulkIO_t *biop, int isStore)
     if (biop->reserved)
         buf_UnreserveBuffers(cm_chunkSize / cm_data.buf_blockSize);
         
-    flags = CM_SCACHESYNC_NEEDCALLBACK;
     if (isStore)
-        flags |= CM_SCACHESYNC_STOREDATA;
+        flags = CM_SCACHESYNC_STOREDATA;
     else
-        flags |= CM_SCACHESYNC_FETCHDATA;
+        flags = CM_SCACHESYNC_FETCHDATA;
 
     scp = biop->scp;
-    for(qdp = biop->bufListp; qdp; qdp = nqdp) {
-        /* lookup next guy first, since we're going to free this one */
-        nqdp = (osi_queueData_t *) osi_QNext(&qdp->q);
+    if (biop->bufListp) {
+       for(qdp = biop->bufListp; qdp; qdp = nqdp) {
+           /* lookup next guy first, since we're going to free this one */
+           nqdp = (osi_queueData_t *) osi_QNext(&qdp->q);
                 
-        /* extract buffer and free queue data */
-        bufp = osi_GetQData(qdp);
-       osi_QRemoveHT((osi_queue_t **) &biop->bufListp,
-                     (osi_queue_t **) &biop->bufListEndp,
-                     &qdp->q);
-        osi_QDFree(qdp);
-
-        /* now, mark I/O as done, unlock the buffer and release it */
-        lock_ObtainMutex(&bufp->mx);
-        lock_ObtainMutex(&scp->mx);
-        cm_SyncOpDone(scp, bufp, flags);
+           /* extract buffer and free queue data */
+           bufp = osi_GetQData(qdp);
+           osi_QRemoveHT((osi_queue_t **) &biop->bufListp,
+                          (osi_queue_t **) &biop->bufListEndp,
+                          &qdp->q);
+           osi_QDFree(qdp);
+
+           /* now, mark I/O as done, unlock the buffer and release it */
+           lock_ObtainMutex(&bufp->mx);
+           lock_ObtainMutex(&scp->mx);
+           cm_SyncOpDone(scp, bufp, flags);
                 
-        /* turn off writing and wakeup users */
-        if (isStore) {
-            if (bufp->flags & CM_BUF_WAITING) {
-                osi_Log2(afsd_logp, "cm_ReleaseBIOD Waking [scp 0x%p] bp 0x%p", scp, bufp);
-                osi_Wakeup((LONG_PTR) bufp);
-            }
-            bufp->flags &= ~(CM_BUF_WRITING | CM_BUF_DIRTY);
-        }
-
-        lock_ReleaseMutex(&scp->mx);
-        lock_ReleaseMutex(&bufp->mx);
-        buf_Release(bufp);
-       bufp = NULL;
+           /* turn off writing and wakeup users */
+           if (isStore) {
+               if (bufp->flags & CM_BUF_WAITING) {
+                   osi_Log2(afsd_logp, "cm_ReleaseBIOD Waking [scp 0x%p] bp 0x%p", scp, bufp);
+                   osi_Wakeup((LONG_PTR) bufp);
+               }
+               bufp->flags &= ~(CM_BUF_WRITING | CM_BUF_DIRTY);
+           }
+
+           lock_ReleaseMutex(&scp->mx);
+           lock_ReleaseMutex(&bufp->mx);
+           buf_Release(bufp);
+           bufp = NULL;
+       }
+    } else {
+       lock_ObtainMutex(&scp->mx);
+       cm_SyncOpDone(scp, NULL, flags);
+       lock_ReleaseMutex(&scp->mx);
     }
 
     /* clean things out */

diff --git a/src/WINNT/afsd/cm_scache.c b/src/WINNT/afsd/cm_scache.c
index 6305431..9765def 100644 (file)
--- a/src/WINNT/afsd/cm_scache.c
+++ b/src/WINNT/afsd/cm_scache.c
@@ -240,7 +240,6 @@ cm_scache_t *cm_GetNewSCache(void)
        /* There were no deleted scache objects that we could use.  Try to find
         * one that simply hasn't been used in a while.
         */
-       while (1) {
            for ( scp = cm_data.scacheLRULastp;
                  scp;
                  scp = (cm_scache_t *) osi_QPrev(&scp->q)) 
@@ -267,17 +266,7 @@ cm_scache_t *cm_GetNewSCache(void)
            }
            osi_Log1(afsd_logp, "GetNewSCache all scache entries in use (retry = %d)", retry);
            
-           /* If get here it means that every scache is either in use or has dirty buffers.
-            * We used to panic.  Now we will give up our lock and wait.
-            */
-           if (++retry < 10) {
-               lock_ReleaseWrite(&cm_scacheLock);
-               Sleep(1000);
-               lock_ObtainWrite(&cm_scacheLock);
-           } else {
                return NULL;
-           }
-       } /* forever */
     }
         
     /* if we get here, we should allocate a new scache entry.  We either are below
@@ -819,6 +808,9 @@ long cm_SyncOp(cm_scache_t *scp, cm_buf_t *bufp, cm_user_t *userp, cm_req_t *req
     cm_buf_t *tbufp;
     afs_uint32 outRights;
     int bufLocked;
+    afs_uint32 sleep_scp_flags = 0;
+    afs_uint32 sleep_buf_cmflags = 0;
+    afs_uint32 sleep_scp_bufs = 0;
 
     /* lookup this first */
     bufLocked = flags & CM_SCACHESYNC_BUFLOCKED;
@@ -994,7 +986,7 @@ long cm_SyncOp(cm_scache_t *scp, cm_buf_t *bufp, cm_user_t *userp, cm_req_t *req
                   cm_fakeDirCallback < 2)
 #endif /* AFS_FREELANCE_CLIENT */
              ) {
-            if (!cm_HaveCallback(scp)) {
+            if ((flags & CM_SCACHESYNC_FORCECB) || !cm_HaveCallback(scp)) {
                 osi_Log1(afsd_logp, "CM SyncOp getting callback on scp 0x%p",
                           scp);
                 if (bufLocked) 
@@ -1007,6 +999,7 @@ long cm_SyncOp(cm_scache_t *scp, cm_buf_t *bufp, cm_user_t *userp, cm_req_t *req
                 }
                 if (code) 
                     return code;
+               flags &= ~CM_SCACHESYNC_FORCECB;        /* only force once */
                 continue;
             }
         }
@@ -1047,6 +1040,10 @@ long cm_SyncOp(cm_scache_t *scp, cm_buf_t *bufp, cm_user_t *userp, cm_req_t *req
         if (flags & CM_SCACHESYNC_NOWAIT) 
             return CM_ERROR_WOULDBLOCK;
 
+       sleep_scp_flags = scp->flags;           /* so we know why we slept */
+       sleep_buf_cmflags = bufp ? bufp->cmFlags : 0;
+       sleep_scp_bufs = (scp->bufReadsp ? 1 : 0) | (scp->bufWritesp ? 2 : 0);
+
         /* wait here, then try again */
         osi_Log1(afsd_logp, "CM SyncOp sleeping scp 0x%p", scp);
         if ( scp->flags & CM_SCACHEFLAG_WAITING ) {
@@ -1144,6 +1141,8 @@ void cm_SyncOpDone(cm_scache_t *scp, cm_buf_t *bufp, afs_uint32 flags)
     osi_queueData_t *qdp;
     cm_buf_t *tbufp;
 
+    lock_AssertMutex(&scp->mx);
+
     /* now, update the recorded state for RPC-type calls */
     if (flags & CM_SCACHESYNC_FETCHSTATUS)
         scp->flags &= ~CM_SCACHEFLAG_FETCHING;

diff --git a/src/WINNT/afsd/cm_utils.c b/src/WINNT/afsd/cm_utils.c
index f85ca04..b7a747c 100644 (file)
--- a/src/WINNT/afsd/cm_utils.c
+++ b/src/WINNT/afsd/cm_utils.c
@@ -259,13 +259,9 @@ long cm_MapRPCError(long error, cm_req_t *reqp)
               || error == 122   /* EDQUOT on Linux */
               || error == 1133) /* EDQUOT on Irix  */
         error = CM_ERROR_QUOTA;
-    else if (error == VNOVNODE) {
-#ifdef COMMENT
+    else if (error == VNOVNODE)
         error = CM_ERROR_BADFD;
-#else
-        error = CM_ERROR_RETRY;
-#endif
-    } else if (error == 21)
+    else if (error == 21)
         return CM_ERROR_ISDIR;
     return error;
 }

diff --git a/src/WINNT/afsd/smb.c b/src/WINNT/afsd/smb.c
index 8bada6f..d90f0fc 100644 (file)
--- a/src/WINNT/afsd/smb.c
+++ b/src/WINNT/afsd/smb.c
@@ -798,6 +798,7 @@ smb_vc_t *smb_FindVC(unsigned short lsn, int flags, int lana)
 {
     smb_vc_t *vcp;
 
+       lock_ObtainWrite(&smb_globalLock);      /* for numVCs */
     lock_ObtainWrite(&smb_rctLock);
     for (vcp = smb_allVCsp; vcp; vcp=vcp->nextp) {
        if (vcp->magic != SMB_VC_MAGIC)
@@ -813,9 +814,7 @@ smb_vc_t *smb_FindVC(unsigned short lsn, int flags, int lana)
     if (!vcp && (flags & SMB_FLAG_CREATE)) {
         vcp = malloc(sizeof(*vcp));
         memset(vcp, 0, sizeof(*vcp));
-       lock_ObtainWrite(&smb_globalLock);
         vcp->vcID = ++numVCs;
-       lock_ReleaseWrite(&smb_globalLock);
        vcp->magic = SMB_VC_MAGIC;
         vcp->refCount = 2;     /* smb_allVCsp and caller */
         vcp->tidCounter = 1;
@@ -858,13 +857,12 @@ smb_vc_t *smb_FindVC(unsigned short lsn, int flags, int lana)
             memset(vcp->encKey, 0, MSV1_0_CHALLENGE_LENGTH);
 
         if (numVCs >= CM_SESSION_RESERVED) {
-           lock_ObtainWrite(&smb_globalLock);
             numVCs = 0;
-           lock_ReleaseWrite(&smb_globalLock);
             osi_Log0(smb_logp, "WARNING: numVCs wrapping around");
         }
     }
     lock_ReleaseWrite(&smb_rctLock);
+       lock_ReleaseWrite(&smb_globalLock);
     return vcp;
 }