Windows: avoid race when writing mountPointString
[openafs.git] / src / WINNT / afsd / cm_vnodeops.c
index 5d23ccd..90f6800 100644 (file)
@@ -7,7 +7,10 @@
  * directory or online at http://www.openafs.org/dl/license10.html
  */
 
+#include <afsconfig.h>
 #include <afs/param.h>
+#include <roken.h>
+
 #include <afs/stds.h>
 
 #include <windows.h>
@@ -147,7 +150,7 @@ long cm_CheckOpen(cm_scache_t *scp, int openMode, int trunc, cm_user_t *userp,
         code = cm_Lock(scp, sLockType, LOffset, LLength, key, 0, userp, reqp, NULL);
 
         if (code == 0) {
-            cm_Unlock(scp, sLockType, LOffset, LLength, key, userp, reqp);
+            cm_Unlock(scp, sLockType, LOffset, LLength, key, 0, userp, reqp);
         } else {
             /* In this case, we allow the file open to go through even
                though we can't enforce mandatory locking on the
@@ -189,17 +192,17 @@ long cm_CheckNTOpen(cm_scache_t *scp, unsigned int desiredAccess,
                    cm_lock_data_t **ldpp)
 {
     long rights;
-    long code;
+    long code = 0;
 
     osi_assertx(ldpp != NULL, "null cm_lock_data_t");
     *ldpp = NULL;
 
     /* Always allow delete; the RPC will tell us if it's OK */
-    if (desiredAccess == DELETE)
-        return 0;
-
     rights = 0;
 
+    if (desiredAccess == DELETE)
+        goto done_2;
+
     if (desiredAccess & (AFS_ACCESS_READ|AFS_ACCESS_EXECUTE))
         rights |= (scp->fileType == CM_SCACHETYPE_DIRECTORY ? PRSFS_LOOKUP : PRSFS_READ);
 
@@ -210,6 +213,9 @@ long cm_CheckNTOpen(cm_scache_t *scp, unsigned int desiredAccess,
     if (desiredAccess & AFS_ACCESS_WRITE)
         rights |= PRSFS_WRITE;
 
+    if (desiredAccess & DELETE)
+        rights |= PRSFS_DELETE;
+
     lock_ObtainWrite(&scp->rw);
 
     code = cm_SyncOp(scp, NULL, userp, reqp, rights,
@@ -291,6 +297,7 @@ long cm_CheckNTOpen(cm_scache_t *scp, unsigned int desiredAccess,
  _done:
     lock_ReleaseWrite(&scp->rw);
 
+ done_2:
     osi_Log3(afsd_logp,"cm_CheckNTOpen scp 0x%p ldp 0x%p code 0x%x", scp, *ldpp, code);
     return code;
 }
@@ -302,7 +309,7 @@ extern long cm_CheckNTOpenDone(cm_scache_t *scp, cm_user_t *userp, cm_req_t *req
     lock_ObtainWrite(&scp->rw);
     if (*ldpp) {
        cm_Unlock(scp, (*ldpp)->sLockType, (*ldpp)->LOffset, (*ldpp)->LLength, 
-                 (*ldpp)->key, userp, reqp);
+                 (*ldpp)->key, 0, userp, reqp);
        free(*ldpp);
        *ldpp = NULL;
     }
@@ -351,7 +358,7 @@ long cm_CheckNTDelete(cm_scache_t *dscp, cm_scache_t *scp, cm_user_t *userp,
         return code;
 
     thyper.HighPart = 0; thyper.LowPart = 0;
-    code = buf_Get(scp, &thyper, &bufferp);
+    code = buf_Get(scp, &thyper, reqp, &bufferp);
     if (code)
         return code;
 
@@ -483,53 +490,57 @@ long cm_ApplyDir(cm_scache_t *scp, cm_DirFuncp_t funcp, void *parmp,
                 return 0;
             }
             sp->caseFold = casefold;
+        }
 
-            /* see if we can find it using the directory hash tables.
-               we can only do exact matches, since the hash is case
-               sensitive. */
-            {
-                cm_dirOp_t dirop;
+        /*
+         * see if we can find it using the directory hash tables.
+         * we can only do exact matches, since the hash is case
+         * sensitive.
+         */
+        if (funcp != (cm_DirFuncp_t)cm_BPlusDirFoo)
+        {
+            cm_dirOp_t dirop;
 #ifdef USE_BPLUS
-                int usedBplus = 0;
+            int usedBplus = 0;
 #endif
 
-                code = ENOENT;
+            code = ENOENT;
 
-                code = cm_BeginDirOp(scp, userp, reqp, CM_DIRLOCK_READ, &dirop);
-                if (code == 0) {
+            code = cm_BeginDirOp(scp, userp, reqp, CM_DIRLOCK_READ,
+                                 CM_DIROP_FLAG_NONE, &dirop);
+            if (code == 0) {
 
 #ifdef USE_BPLUS
-                    code = cm_BPlusDirLookup(&dirop, sp->nsearchNamep, &sp->fid);
-                    if (code != EINVAL)
-                        usedBplus = 1;
-                    else 
+                code = cm_BPlusDirLookup(&dirop, sp->nsearchNamep, &sp->fid);
+                if (code != EINVAL)
+                    usedBplus = 1;
+                else
 #endif
-                        code = cm_DirLookup(&dirop, sp->searchNamep, &sp->fid);
+                    code = cm_DirLookup(&dirop, sp->searchNamep, &sp->fid);
 
-                    cm_EndDirOp(&dirop);
-                }
+                cm_EndDirOp(&dirop);
+            }
 
-                if (code == 0) {
+            if (code == 0) {
+                /* found it */
+                sp->found = TRUE;
+                sp->ExactFound = TRUE;
+                *retscp = NULL; /* force caller to call cm_GetSCache() */
+                return 0;
+            }
+#ifdef USE_BPLUS
+            if (usedBplus) {
+                if (sp->caseFold && code == CM_ERROR_INEXACT_MATCH) {
                     /* found it */
                     sp->found = TRUE;
-                    sp->ExactFound = TRUE;
+                    sp->ExactFound = FALSE;
                     *retscp = NULL; /* force caller to call cm_GetSCache() */
                     return 0;
                 }
-#ifdef USE_BPLUS
-                if (usedBplus) {
-                    if (sp->caseFold && code == CM_ERROR_INEXACT_MATCH) {
-                        /* found it */
-                        sp->found = TRUE;
-                        sp->ExactFound = FALSE;
-                        *retscp = NULL; /* force caller to call cm_GetSCache() */
-                        return 0;
-                    }
-                    
-                    return CM_ERROR_BPLUS_NOMATCH;
-                }
-#endif 
+
+                return CM_ERROR_BPLUS_NOMATCH;
             }
+#endif
         }
     }  
 
@@ -590,7 +601,7 @@ long cm_ApplyDir(cm_scache_t *scp, cm_DirFuncp_t funcp, void *parmp,
                 bufferp = NULL;
             }
 
-            code = buf_Get(scp, &thyper, &bufferp);
+            code = buf_Get(scp, &thyper, reqp, &bufferp);
             if (code) {
                 /* if buf_Get() fails we do not have a buffer object to lock */
                 bufferp = NULL;
@@ -667,6 +678,21 @@ long cm_ApplyDir(cm_scache_t *scp, cm_DirFuncp_t funcp, void *parmp,
         tp = bufferp->datap + entryInBuffer;
         dep = (cm_dirEntry_t *) tp;    /* now points to AFS3 dir entry */
 
+        /*
+         * here are some consistency checks
+         */
+        if (dep->flag != CM_DIR_FFIRST ||
+            strlen(dep->name) > 256) {
+            code = CM_ERROR_INVAL;
+            osi_Log2(afsd_logp,
+                     "cm_ApplyDir invalid directory entry for scp %p bufp %p",
+                     scp, bufferp);
+            osi_Log4(afsd_logp,"... cell %u vol %u vnode %u uniq %u",
+                     scp->fid.cell, scp->fid.volume, scp->fid.vnode, scp->fid.unique);
+            bufferp->dataVersion = CM_BUF_VERSION_BAD;
+            break;
+        }
+
         /* while we're here, compute the next entry's location, too,
          * since we'll need it when writing out the cookie into the
          * dir listing stream.
@@ -808,9 +834,6 @@ long cm_LookupSearchProc(cm_scache_t *scp, cm_dirEntry_t *dep, void *rockp,
 long cm_ReadMountPoint(cm_scache_t *scp, cm_user_t *userp, cm_req_t *reqp)
 {
     long code;
-    cm_buf_t *bufp = NULL;
-    osi_hyper_t thyper;
-    int tlen;
 
     if (scp->mountPointStringp[0]) 
         return 0;
@@ -825,62 +848,30 @@ long cm_ReadMountPoint(cm_scache_t *scp, cm_user_t *userp, cm_req_t *reqp)
     } else 
 #endif /* AFS_FREELANCE_CLIENT */        
     {
-        /* otherwise, we have to read it in */
-        lock_ReleaseWrite(&scp->rw);
+        char temp[MOUNTPOINTLEN];
+        osi_hyper_t thyper;
 
+        /* otherwise, we have to read it in */
         thyper.LowPart = thyper.HighPart = 0;
-        code = buf_Get(scp, &thyper, &bufp);
-
-        lock_ObtainWrite(&scp->rw);
+        code = cm_GetData(scp, &thyper, temp, MOUNTPOINTLEN, userp, reqp);
         if (code)
             return code;
 
-        while (1) {
-            code = cm_SyncOp(scp, bufp, userp, reqp, 0,
-                              CM_SCACHESYNC_READ | CM_SCACHESYNC_NEEDCALLBACK);
-            if (code)
-                goto done;
-
-            cm_SyncOpDone(scp, bufp, CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_READ);
-
-            if (cm_HaveBuffer(scp, bufp, 0)) 
-                break;
-
-            /* otherwise load buffer */
-            code = cm_GetBuffer(scp, bufp, NULL, userp, reqp);
-            if (code)
-                goto done;
-        }
-        /* locked, has callback, has valid data in buffer */
-        if ((tlen = scp->length.LowPart) > MOUNTPOINTLEN - 1) 
-            return CM_ERROR_TOOBIG;
-        if (tlen <= 0) {
-            code = CM_ERROR_INVAL;
-            goto done;
-        }
-
-        /* someone else did the work while we were out */
-        if (scp->mountPointStringp[0]) {
-            code = 0;
-            goto done;
-        }
-
-        /* otherwise, copy out the link */
-        memcpy(scp->mountPointStringp, bufp->datap, tlen);
-
-        /* now make it null-terminated.  Note that the original contents of a
-         * link that is a mount point is "#volname." where "." is there just to
-         * be turned into a null.  That is, we can trash the last char of the
-         * link without damaging the vol name.  This is a stupid convention,
-         * but that's the protocol.
+        /*
+         * scp->length is the actual length of the mount point string.
+         * It is current because cm_GetData merged the most up to date
+         * status info into scp and has not dropped the rwlock since.
          */
-        scp->mountPointStringp[tlen-1] = 0;
-        code = 0;
+        if (scp->length.LowPart > MOUNTPOINTLEN - 1)
+            return CM_ERROR_TOOBIG;
+        if (scp->length.LowPart == 0)
+            return CM_ERROR_INVAL;
 
-      done:
-        if (bufp) 
-            buf_Release(bufp);
+        /* convert the terminating dot to a NUL */
+        temp[thyper.LowPart - 1] = 0;
+        memcpy(scp->mountPointStringp, temp, thyper.LowPart);
     }
+
     return code;
 }
 
@@ -893,7 +884,6 @@ long cm_FollowMountPoint(cm_scache_t *scp, cm_scache_t *dscp, cm_user_t *userp,
 {
     fschar_t *cellNamep = NULL;
     fschar_t *volNamep = NULL;
-    int tlen;
     afs_uint32 code;
     fschar_t *cp;
     fschar_t *mpNamep;
@@ -918,7 +908,6 @@ long cm_FollowMountPoint(cm_scache_t *scp, cm_scache_t *dscp, cm_user_t *userp,
     mpNamep = scp->mountPointStringp;
     if (!mpNamep[0])
        return CM_ERROR_NOSUCHPATH;
-    tlen = cm_FsStrLen(scp->mountPointStringp);
     mtType = *scp->mountPointStringp;
 
     cp = cm_FsStrChr(mpNamep, _FS(':'));
@@ -936,7 +925,21 @@ long cm_FollowMountPoint(cm_scache_t *scp, cm_scache_t *dscp, cm_user_t *userp,
         /* normal mt pt */
         volNamep = cm_FsStrDup(mpNamep + 1);
 
-        cellp = cm_FindCellByID(scp->fid.cell, 0);
+#ifdef AFS_FREELANCE_CLIENT
+        /* 
+         * Mount points in the Freelance cell should default
+         * to the workstation cell.
+         */
+        if (cm_freelanceEnabled &&
+             scp->fid.cell==AFS_FAKE_ROOT_CELL_ID &&
+             scp->fid.volume==AFS_FAKE_ROOT_VOL_ID )
+        {       
+            fschar_t rootCellName[256]="";
+            cm_GetRootCellName(rootCellName);
+            cellp = cm_GetCell(rootCellName, 0);
+        } else 
+#endif /* AFS_FREELANCE_CLIENT */        
+            cellp = cm_FindCellByID(scp->fid.cell, 0);
     }
 
     if (!cellp) {
@@ -1038,6 +1041,7 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
     int getroot;
     normchar_t *nnamep = NULL;
     fschar_t *fnamep = NULL;
+    size_t fnlen;
 
     *outScpp = NULL;
 
@@ -1065,6 +1069,7 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
         goto done;
     }
 
+retry_lookup:
     if (flags & CM_FLAG_NOMOUNTCHASE) {
         /* In this case, we should go and call cm_Dir* functions
            directly since the following cm_ApplyDir() function will
@@ -1075,7 +1080,8 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
         int usedBplus = 0;
 #endif
 
-        code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_READ, &dirop);
+        code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_READ,
+                             CM_DIROP_FLAG_NONE, &dirop);
         if (code == 0) {
 #ifdef USE_BPLUS
             code = cm_BPlusDirLookup(&dirop, nnamep, &rock.fid);
@@ -1102,7 +1108,8 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
                 goto haveFid;
             }
             
-            return CM_ERROR_BPLUS_NOMATCH;
+            code = CM_ERROR_BPLUS_NOMATCH;
+            goto notfound;
         }
 #endif
     }
@@ -1122,7 +1129,7 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
      * that we stopped early, probably because we found the entry we're
      * looking for.  Any other non-zero code is an error.
      */
-    if (code && code != CM_ERROR_STOPNOW) {
+    if (code && code != CM_ERROR_STOPNOW && code != CM_ERROR_BPLUS_NOMATCH) {
         /* if the cm_scache_t we are searching in is not a directory 
          * we must return path not found because the error 
          * is to describe the final component not an intermediary
@@ -1136,6 +1143,7 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
         goto done;
     }
 
+notfound:
     getroot = (dscp==cm_data.rootSCachep) ;
     if (!rock.found) {
         if (!cm_freelanceEnabled || !getroot) {
@@ -1154,6 +1162,7 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
             /* nonexistent dir on freelance root, so add it */
             fschar_t fullname[CELL_MAXNAMELEN + 1] = ".";  /* +1 so that when we skip the . the size is still CELL_MAXNAMELEN */
             int  found = 0;
+            int  retry = 0;
 
             osi_Log1(afsd_logp,"cm_Lookup adding mount for non-existent directory: %S", 
                      osi_LogSaveClientString(afsd_logp,cnamep));
@@ -1166,28 +1175,71 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
              */
 
             code = -1;
+            fnlen = strlen(fnamep);
+            if ( fnamep[fnlen-1] == '.') {
+                fnamep[fnlen-1] = '\0';
+                fnlen--;
+                retry = 1;
+            }
+
             if (cnamep[0] == '.') {
                 if (cm_GetCell_Gen(&fnamep[1], &fullname[1], CM_FLAG_CREATE)) {
                     found = 1;
-                    if (!cm_FreelanceMountPointExists(fullname, 0))
-                        code = cm_FreelanceAddMount(fullname, &fullname[1], "root.cell.",
-                                                    1, &rock.fid);
-                    if ( cm_FsStrCmpI(&fnamep[1], &fullname[1]) && 
-                         !cm_FreelanceMountPointExists(fnamep, flags & CM_FLAG_DFS_REFERRAL ? 1 : 0) &&
-                         !cm_FreelanceSymlinkExists(fnamep, flags & CM_FLAG_DFS_REFERRAL ? 1 : 0))
-                        code = cm_FreelanceAddSymlink(fnamep, fullname, &rock.fid);
+                    code = cm_FreelanceAddMount(fullname, &fullname[1], "root.cell", 1, &rock.fid);
+                    if ( cm_FsStrCmpI(&fnamep[1], &fullname[1])) {
+                        /*
+                         * Do not permit symlinks that are one of:
+                         *  . the cellname followed by a dot
+                         *  . the cellname minus a single character
+                         *  . a substring of the cellname that does not consist of full components
+                         */
+                        if ( cm_strnicmp_utf8(&fnamep[1], fullname, (int)fnlen-1) == 0 &&
+                             (fnlen-1 == strlen(fullname)-1 || fullname[fnlen-1] != '.'))
+                        {
+                            /* do not add; substitute fullname for the search */
+                            free(fnamep);
+                            fnamep = malloc(strlen(fullname)+2);
+                            fnamep[0] = '.';
+                            strncpy(&fnamep[1], fullname, strlen(fullname)+1);
+                            retry = 1;
+                        } else {
+                            code = cm_FreelanceAddSymlink(fnamep, fullname, &rock.fid);
+                        }
+                    }
                 }
             } else {
                 if (cm_GetCell_Gen(fnamep, fullname, CM_FLAG_CREATE)) {
                     found = 1;
-                    if (!cm_FreelanceMountPointExists(fullname, 0))
-                        code = cm_FreelanceAddMount(fullname, fullname, "root.cell.", 0, &rock.fid);
-                    if ( cm_FsStrCmpI(fnamep, fullname) && 
-                         !cm_FreelanceMountPointExists(fnamep, flags & CM_FLAG_DFS_REFERRAL ? 1 : 0) &&
-                         !cm_FreelanceSymlinkExists(fnamep, flags & CM_FLAG_DFS_REFERRAL ? 1 : 0))
-                        code = cm_FreelanceAddSymlink(fnamep, fullname, &rock.fid);
+                    code = cm_FreelanceAddMount(fullname, fullname, "root.cell", 0, &rock.fid);
+                    if ( cm_FsStrCmpI(fnamep, fullname)) {
+                        /*
+                         * Do not permit symlinks that are one of:
+                         *  . the cellname followed by a dot
+                         *  . the cellname minus a single character
+                         *  . a substring of the cellname that does not consist of full components
+                         */
+                        if ( cm_strnicmp_utf8(fnamep, fullname, (int)fnlen-1) == 0 &&
+                             (fnlen == strlen(fullname)-1 || fullname[fnlen] != '.'))
+                        {
+                            /* do not add; substitute fullname for the search */
+                                free(fnamep);
+                                fnamep = strdup(fullname);
+                                code = 0;
+                                retry = 1;
+                        } else {
+                            code = cm_FreelanceAddSymlink(fnamep, fullname, &rock.fid);
+                        }
+                    }
                 }
             }
+
+            if (retry) {
+                if (nnamep)
+                    free(nnamep);
+                nnamep = cm_FsStringToNormStringAlloc(fnamep, -1, NULL);
+                goto retry_lookup;
+            }
+
             if (!found || code) {   /* add mount point failed, so give up */
                 if (flags & CM_FLAG_CHECKPATH)
                     code = CM_ERROR_NOSUCHPATH;
@@ -1205,7 +1257,7 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
         }
     }
 
-  haveFid:       
+  haveFid:
     if ( !tscp )    /* we did not find it in the dnlc */
     {
         dnlcHit = 0; 
@@ -1216,14 +1268,26 @@ long cm_LookupInternal(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_u
     /* tscp is now held */
 
     lock_ObtainWrite(&tscp->rw);
-    code = cm_SyncOp(tscp, NULL, userp, reqp, 0,
-                      CM_SCACHESYNC_GETSTATUS | CM_SCACHESYNC_NEEDCALLBACK);
-    if (code) { 
-        lock_ReleaseWrite(&tscp->rw);
-        cm_ReleaseSCache(tscp);
-        goto done;
+
+    /*
+     * Do not get status if we do not already have a callback.
+     * The process of reading the mount point string will obtain status information
+     * in a single RPC.  No reason to add a second round trip.
+     *
+     * If we do have a callback, use cm_SyncOp to get status in case the
+     * current cm_user_t is not the same as the one that obtained the
+     * mount point string contents.
+     */
+    if (cm_HaveCallback(tscp)) {
+        code = cm_SyncOp(tscp, NULL, userp, reqp, 0,
+                          CM_SCACHESYNC_GETSTATUS | CM_SCACHESYNC_NEEDCALLBACK);
+        if (code) {
+            lock_ReleaseWrite(&tscp->rw);
+            cm_ReleaseSCache(tscp);
+            goto done;
+        }
+        cm_SyncOpDone(tscp, NULL, CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
     }
-    cm_SyncOpDone(tscp, NULL, CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
     /* tscp is now locked */
 
     if (!(flags & CM_FLAG_NOMOUNTCHASE)
@@ -1524,10 +1588,13 @@ long cm_Unlink(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t * cnamep,
     cm_scache_t *scp = NULL;
     int free_fnamep = FALSE;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     if (fnamep == NULL) {
         code = -1;
 #ifdef USE_BPLUS
-        code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_READ, &dirop);
+        code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_READ,
+                             CM_DIROP_FLAG_NONE, &dirop);
         if (code == 0) {
             code = cm_BPlusDirLookupOriginalName(&dirop, cnamep, &fnamep);
             if (code == 0)
@@ -1548,9 +1615,18 @@ long cm_Unlink(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t * cnamep,
 #endif  
 
     code = cm_Lookup(dscp, cnamep, CM_FLAG_NOMOUNTCHASE, userp, reqp, &scp);
+    if (code)
+        goto done;
+
+    /* Check for RO volume */
+    if (dscp->flags & CM_SCACHEFLAG_RO) {
+        code = CM_ERROR_READONLY;
+        goto done;
+    }
 
     /* make sure we don't screw up the dir status during the merge */
-    code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, &dirop);
+    code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE,
+                         CM_DIROP_FLAG_NONE, &dirop);
 
     lock_ObtainWrite(&dscp->rw);
     sflags = CM_SCACHESYNC_STOREDATA;
@@ -1593,7 +1669,7 @@ long cm_Unlink(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t * cnamep,
     cm_dnlcRemove(dscp, cnamep);
     cm_SyncOpDone(dscp, NULL, sflags);
     if (code == 0) {
-        cm_MergeStatus(NULL, dscp, &newDirStatus, &volSync, userp, CM_MERGEFLAG_DIROP);
+        cm_MergeStatus(NULL, dscp, &newDirStatus, &volSync, userp, reqp, CM_MERGEFLAG_DIROP);
     } else if (code == CM_ERROR_NOSUCHFILE) {
        /* windows would not have allowed the request to delete the file 
         * if it did not believe the file existed.  therefore, we must 
@@ -1615,7 +1691,9 @@ long cm_Unlink(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t * cnamep,
         cm_ReleaseSCache(scp);
         if (code == 0) {
            lock_ObtainWrite(&scp->rw);
-            scp->flags |= CM_SCACHEFLAG_DELETED;
+            if (--scp->linkCount == 0)
+                scp->flags |= CM_SCACHEFLAG_DELETED;
+            cm_DiscardSCache(scp);
            lock_ReleaseWrite(&scp->rw);
         }
     }
@@ -1632,10 +1710,7 @@ long cm_Unlink(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t * cnamep,
  */
 long cm_HandleLink(cm_scache_t *linkScp, cm_user_t *userp, cm_req_t *reqp)
 {
-    long code;
-    cm_buf_t *bufp;
-    long temp;
-    osi_hyper_t thyper;
+    long code = 0;
 
     lock_AssertWrite(&linkScp->rw);
     if (!linkScp->mountPointStringp[0]) {
@@ -1650,56 +1725,36 @@ long cm_HandleLink(cm_scache_t *linkScp, cm_user_t *userp, cm_req_t *reqp)
         } else 
 #endif /* AFS_FREELANCE_CLIENT */        
         {
-            /* read the link data from the file server*/
-            lock_ReleaseWrite(&linkScp->rw);
+            char temp[MOUNTPOINTLEN];
+            osi_hyper_t thyper;
+
+            /* read the link data from the file server */
             thyper.LowPart = thyper.HighPart = 0;
-            code = buf_Get(linkScp, &thyper, &bufp);
-            lock_ObtainWrite(&linkScp->rw);
-            if (code) 
+            code = cm_GetData(linkScp, &thyper, temp, MOUNTPOINTLEN, userp, reqp);
+            if (code)
                 return code;
-            while (1) {
-                code = cm_SyncOp(linkScp, bufp, userp, reqp, 0,
-                                  CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_READ);
-                if (code) {
-                    buf_Release(bufp);
-                    return code;
-                }
-                cm_SyncOpDone(linkScp, bufp, CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_READ);
 
-                if (cm_HaveBuffer(linkScp, bufp, 0)) 
-                    break;
-
-                code = cm_GetBuffer(linkScp, bufp, NULL, userp, reqp);
-                if (code) {
-                    buf_Release(bufp);
-                    return code;
-                }
-            } /* while loop to get the data */
-
-            /* now if we still have no link read in,
-             * copy the data from the buffer */
-            if ((temp = linkScp->length.LowPart) >= MOUNTPOINTLEN) {
-                buf_Release(bufp);
+            /*
+             * linkScp->length is the actual length of the symlink target string.
+             * It is current because cm_GetData merged the most up to date
+             * status info into scp and has not dropped the rwlock since.
+             */
+            if (linkScp->length.LowPart > MOUNTPOINTLEN - 1)
                 return CM_ERROR_TOOBIG;
-            }       
+            if (linkScp->length.LowPart == 0)
+                return CM_ERROR_INVAL;
 
-            /* otherwise, it fits; make sure it is still null (could have
-             * lost race with someone else referencing this link above),
-             * and if so, copy in the data.
-             */
-            if (!linkScp->mountPointStringp[0]) {
-                strncpy(linkScp->mountPointStringp, bufp->datap, temp);
-                linkScp->mountPointStringp[temp] = 0;  /* null terminate */
-            }
-            buf_Release(bufp);
+            /* convert the terminating dot to a NUL */
+            temp[thyper.LowPart - 1] = 0;
+            memcpy(linkScp->mountPointStringp, temp, thyper.LowPart);
         }
         
         if ( !strnicmp(linkScp->mountPointStringp, "msdfs:", strlen("msdfs:")) )
             linkScp->fileType = CM_SCACHETYPE_DFSLINK;
 
-    }  /* don't have sym link contents cached */
+    }  /* don't have symlink contents cached */
 
-    return 0;
+    return code;
 }       
 
 /* called with a held vnode and a path suffix, with the held vnode being a
@@ -1721,6 +1776,25 @@ long cm_AssembleLink(cm_scache_t *linkScp, fschar_t *pathSuffixp,
     *newSpaceBufferp = NULL;
 
     lock_ObtainWrite(&linkScp->rw);
+    /*
+     * Do not get status if we do not already have a callback.
+     * The process of reading the symlink string will obtain status information
+     * in a single RPC.  No reason to add a second round trip.
+     *
+     * If we do have a callback, use cm_SyncOp to get status in case the
+     * current cm_user_t is not the same as the one that obtained the
+     * symlink string contents.
+     */
+    if (cm_HaveCallback(linkScp)) {
+        code = cm_SyncOp(linkScp, NULL, userp, reqp, 0,
+                          CM_SCACHESYNC_GETSTATUS | CM_SCACHESYNC_NEEDCALLBACK);
+        if (code) {
+            lock_ReleaseWrite(&linkScp->rw);
+            cm_ReleaseSCache(linkScp);
+            goto done;
+        }
+        cm_SyncOpDone(linkScp, NULL, CM_SCACHESYNC_NEEDCALLBACK | CM_SCACHESYNC_GETSTATUS);
+    }
     code = cm_HandleLink(linkScp, userp, reqp);
     if (code)
         goto done;
@@ -1742,8 +1816,8 @@ long cm_AssembleLink(cm_scache_t *linkScp, fschar_t *pathSuffixp,
             StringCbCopyA((char *) tsp->data, sizeof(tsp->data), linkp+cm_mountRootLen+1);
         else
             tsp->data[0] = 0;
-        *newRootScpp = cm_data.rootSCachep;
-        cm_HoldSCache(cm_data.rootSCachep);
+        *newRootScpp = cm_RootSCachep(userp, reqp);
+        cm_HoldSCache(*newRootScpp);
     } else if (linkp[0] == '\\' && linkp[1] == '\\') {
         if (!strnicmp(&linkp[2], cm_NetbiosName, (len = (long)strlen(cm_NetbiosName)))) 
         {
@@ -1756,8 +1830,8 @@ long cm_AssembleLink(cm_scache_t *linkScp, fschar_t *pathSuffixp,
                 if (*p == '\\')
                     *p = '/';
             }
-            *newRootScpp = cm_data.rootSCachep;
-            cm_HoldSCache(cm_data.rootSCachep);
+            *newRootScpp = cm_RootSCachep(userp, reqp);
+            cm_HoldSCache(*newRootScpp);
         } else {
             linkScp->fileType = CM_SCACHETYPE_DFSLINK;
             StringCchCopyA(tsp->data,lengthof(tsp->data), linkp);
@@ -1774,8 +1848,8 @@ long cm_AssembleLink(cm_scache_t *linkScp, fschar_t *pathSuffixp,
          * but this seems to create problems.  instead, we will just
          * reject the link */
         StringCchCopyA(tsp->data,lengthof(tsp->data), linkp+1);
-        *newRootScpp = cm_data.rootSCachep;
-        cm_HoldSCache(cm_data.rootSCachep);
+        *newRootScpp = cm_RootSCachep(userp, reqp);
+        cm_HoldSCache(*newRootScpp);
 #else
         /* we still copy the link data into the response so that 
          * the user can see what the link points to
@@ -1932,17 +2006,13 @@ long cm_NameI(cm_scache_t *rootSCachep, clientchar_t *pathp, long flags,
                         fid_count = i+1;
                     } else {
                         /* add the new fid to the list */
-                        for ( i=0; i<fid_count; i++) {
-                            if ( !cm_FidCmp(&nscp->fid, &fids[i]) ) {
-                                code = CM_ERROR_TOO_MANY_SYMLINKS;
-                                cm_ReleaseSCache(nscp);
-                                nscp = NULL;
-                                break;
-                            }
-                        }
-                        if (i == fid_count && fid_count < MAX_FID_COUNT) {
-                            fids[fid_count++] = nscp->fid;
-                        }
+                        if (fid_count == MAX_FID_COUNT) {
+                            code = CM_ERROR_TOO_MANY_SYMLINKS;
+                            cm_ReleaseSCache(nscp);
+                            nscp = NULL;
+                            break;
+                        }       
+                        fids[fid_count++] = nscp->fid;
                     }
                 }
 
@@ -2026,9 +2096,10 @@ long cm_NameI(cm_scache_t *rootSCachep, clientchar_t *pathp, long flags,
                     }
 
                     if (code == 0 && linkScp != NULL) {
-                        if (linkScp == cm_data.rootSCachep) 
+                        if (linkScp == cm_data.rootSCachep) {
                             fid_count = 0;
-                        else {
+                            i = 0;
+                        } else {
                             for ( i=0; i<fid_count; i++) {
                                 if ( !cm_FidCmp(&linkScp->fid, &fids[i]) ) {
                                     code = CM_ERROR_TOO_MANY_SYMLINKS;
@@ -2282,8 +2353,10 @@ cm_TryBulkStatRPC(cm_scache_t *dscp, cm_bulkStat_t *bbp, cm_user_t *userp, cm_re
     cm_scache_t *scp;
     cm_fid_t tfid;
     struct rx_connection * rxconnp;
-    int inlinebulk = 0;                /* Did we use InlineBulkStatus RPC or not? */
-        
+    int inlinebulk;            /* Did we use InlineBulkStatus RPC or not? */
+
+    memset(&volSync, 0, sizeof(volSync));
+
     /* otherwise, we may have one or more bulk stat's worth of stuff in bb;
      * make the calls to create the entries.  Handle AFSCBMAX files at a
      * time.
@@ -2301,15 +2374,26 @@ cm_TryBulkStatRPC(cm_scache_t *dscp, cm_bulkStat_t *bbp, cm_user_t *userp, cm_re
         callbackStruct.AFSCBs_val = &bbp->callbacks[filex];
         cm_StartCallbackGrantingCall(NULL, &cbReq);
         osi_Log1(afsd_logp, "CALL BulkStatus, %d entries", filesThisCall);
+
+        /*
+         * Whenever cm_Analyze is called for a RXAFS_ RPC there must
+         * be a FID provided.  However, the error code from RXAFS_BulkStatus
+         * or RXAFS_InlinkBulkStatus does not apply to any FID.  Therefore,
+         * we generate an invalid FID to match with the RPC error.
+         */
+        cm_SetFid(&tfid, dscp->fid.cell, dscp->fid.volume, 0, 0);
+
         do {
-            code = cm_ConnFromFID(&dscp->fid, userp, reqp, &connp);
+            inlinebulk = 0;
+
+            code = cm_ConnFromFID(&tfid, userp, reqp, &connp);
             if (code) 
                 continue;
 
             rxconnp = cm_GetRxConn(connp);
            if (!(connp->serverp->flags & CM_SERVERFLAG_NOINLINEBULK)) {
                code = RXAFS_InlineBulkStatus(rxconnp, &fidStruct,
-                                     &statStruct, &callbackStruct, &volSync);
+                                              &statStruct, &callbackStruct, &volSync);
                if (code == RXGEN_OPCODE) {
                    cm_SetServerNoInlineBulk(connp->serverp, 0);
                } else {
@@ -2322,72 +2406,117 @@ cm_TryBulkStatRPC(cm_scache_t *dscp, cm_bulkStat_t *bbp, cm_user_t *userp, cm_re
            }
             rx_PutConnection(rxconnp);
 
-        } while (cm_Analyze(connp, userp, reqp, &dscp->fid,
-                             &volSync, NULL, &cbReq, code));
+            /*
+             * If InlineBulk RPC was called and it succeeded,
+             * then pull out the return code from the status info
+             * and use it for cm_Analyze so that we can failover to other
+             * .readonly volume instances.  But only do it for errors that
+             * are volume global.
+             */
+            if (inlinebulk && code == 0 && (&bbp->stats[0])->errorCode) {
+                osi_Log1(afsd_logp, "cm_TryBulkStat inline-bulk stat error: %d",
+                          (&bbp->stats[0])->errorCode);
+                switch ((&bbp->stats[0])->errorCode) {
+                case VBUSY:
+                case VRESTARTING:
+                case VNOVOL:
+                case VMOVED:
+                case VOFFLINE:
+                case VSALVAGE:
+                case VNOSERVICE:
+                case VIO:
+                    code = (&bbp->stats[0])->errorCode;
+                    break;
+                default:
+                    /* Rx and Rxkad errors are volume global */
+                    if ( (&bbp->stats[0])->errorCode >= -64 && (&bbp->stats[0])->errorCode < 0 ||
+                         (&bbp->stats[0])->errorCode >= ERROR_TABLE_BASE_RXK && (&bbp->stats[0])->errorCode < ERROR_TABLE_BASE_RXK + 256)
+                        code = (&bbp->stats[0])->errorCode;
+                }
+            }
+        } while (cm_Analyze(connp, userp, reqp, &tfid, &volSync, NULL, &cbReq, code));
         code = cm_MapRPCError(code, reqp);
-        if (code)
-            osi_Log2(afsd_logp, "CALL %sBulkStatus FAILURE code 0x%x", 
-                     inlinebulk ? "Inline" : "", code);
-        else
-            osi_Log1(afsd_logp, "CALL %sBulkStatus SUCCESS", inlinebulk ? "Inline" : "");
 
-        /* may as well quit on an error, since we're not going to do
+        /*
+         * might as well quit on an error, since we're not going to do
          * much better on the next immediate call, either.
          */
         if (code) {
-            cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, 0);
+            osi_Log2(afsd_logp, "CALL %sBulkStatus FAILURE code 0x%x",
+                     inlinebulk ? "Inline" : "", code);
+            cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, NULL, 0);
             break;
         }
 
-        /* otherwise, we should do the merges */
+        /*
+         * The bulk RPC has succeeded or at least not failed with a
+         * volume global error result.  For items that have inlineBulk
+         * errors we must call cm_Analyze in order to perform required
+         * logging of errors.
+         *
+         * If the RPC was not inline bulk or the entry either has no error
+         * the status must be merged.
+         */
+        osi_Log1(afsd_logp, "CALL %sBulkStatus SUCCESS", inlinebulk ? "Inline" : "");
+
         for (i = 0; i<filesThisCall; i++) {
             j = filex + i;
             cm_SetFid(&tfid, dscp->fid.cell, bbp->fids[j].Volume, bbp->fids[j].Vnode, bbp->fids[j].Unique);
-            code = cm_GetSCache(&tfid, &scp, userp, reqp);
-            if (code != 0) 
-                continue;
 
-            /* otherwise, if this entry has no callback info, 
-             * merge in this.
-             */
-            lock_ObtainWrite(&scp->rw);
-            /* now, we have to be extra paranoid on merging in this
-             * information, since we didn't use cm_SyncOp before
-             * starting the fetch to make sure that no bad races
-             * were occurring.  Specifically, we need to make sure
-             * we don't obliterate any newer information in the
-             * vnode than have here.
-             *
-             * Right now, be pretty conservative: if there's a
-             * callback or a pending call, skip it.
-             */
-            if ((scp->cbServerp == NULL || (scp->flags & CM_SCACHEFLAG_EACCESS))
-                 && !(scp->flags &
-                       (CM_SCACHEFLAG_FETCHING
-                         | CM_SCACHEFLAG_STORING
-                         | CM_SCACHEFLAG_SIZESTORING))) {
-                cm_EndCallbackGrantingCall(scp, &cbReq,
-                                            &bbp->callbacks[j],
-                                            CM_CALLBACK_MAINTAINCOUNT);
-                cm_MergeStatus(dscp, scp, &bbp->stats[j], &volSync, userp, 0);
-            }       
-            lock_ReleaseWrite(&scp->rw);
-            cm_ReleaseSCache(scp);
+            if (inlinebulk && (&bbp->stats[j])->errorCode) {
+                cm_req_t treq = *reqp;
+                cm_Analyze(NULL, userp, &treq, &tfid, &volSync, NULL, &cbReq, (&bbp->stats[j])->errorCode);
+            } else {
+                code = cm_GetSCache(&tfid, &scp, userp, reqp);
+                if (code != 0)
+                    continue;
+
+                /*
+                 * otherwise, if this entry has no callback info,
+                 * merge in this.  If there is existing callback info
+                 * we skip the merge because the existing data must be
+                 * current (we have a callback) and the response from
+                 * a non-inline bulk rpc might actually be wrong.
+                 *
+                 * now, we have to be extra paranoid on merging in this
+                 * information, since we didn't use cm_SyncOp before
+                 * starting the fetch to make sure that no bad races
+                 * were occurring.  Specifically, we need to make sure
+                 * we don't obliterate any newer information in the
+                 * vnode than have here.
+                 *
+                 * Right now, be pretty conservative: if there's a
+                 * callback or a pending call, skip it.
+                 * However, if the prior attempt to obtain status
+                 * was refused access or the volume is .readonly,
+                 * take the data in any case since we have nothing
+                 * better for the in flight directory enumeration that
+                 * resulted in this function being called.
+                 */
+                lock_ObtainRead(&scp->rw);
+                if ((scp->cbServerp == NULL &&
+                     !(scp->flags & (CM_SCACHEFLAG_FETCHING | CM_SCACHEFLAG_STORING | CM_SCACHEFLAG_SIZESTORING))) ||
+                     (scp->flags & CM_SCACHEFLAG_PURERO) ||
+                     (scp->flags & CM_SCACHEFLAG_EACCESS))
+                {
+                    lock_ConvertRToW(&scp->rw);
+                    cm_EndCallbackGrantingCall(scp, &cbReq,
+                                               &bbp->callbacks[j],
+                                               &volSync,
+                                               CM_CALLBACK_MAINTAINCOUNT);
+                    cm_MergeStatus(dscp, scp, &bbp->stats[j], &volSync, userp, reqp, 0);
+                    lock_ReleaseWrite(&scp->rw);
+                } else {
+                    lock_ReleaseRead(&scp->rw);
+                }
+                cm_ReleaseSCache(scp);
+            }
         } /* all files in the response */
         /* now tell it to drop the count,
          * after doing the vnode processing above */
-        cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, 0);
+        cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, NULL, 0);
     }  /* while there are still more files to process */
 
-    /* If we did the InlineBulk RPC pull out the return code and log it */
-    if (inlinebulk) {
-       if ((&bbp->stats[0])->errorCode) {
-           osi_Log1(afsd_logp, "cm_TryBulkStat bulk stat error: %d", 
-                    (&bbp->stats[0])->errorCode);
-            code = (&bbp->stats[0])->errorCode;
-       }
-    }
-
     return code;
 }
 
@@ -2579,11 +2708,20 @@ long cm_SetAttr(cm_scache_t *scp, cm_attr_t *attrp, cm_user_t *userp,
     AFSStoreStatus afsInStatus;
     struct rx_connection * rxconnp;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     /* handle file length setting */
     if (attrp->mask & CM_ATTRMASK_LENGTH)
         return cm_SetLength(scp, &attrp->length, userp, reqp);
 
     lock_ObtainWrite(&scp->rw);
+    /* Check for RO volume */
+    if (scp->flags & CM_SCACHEFLAG_RO) {
+        code = CM_ERROR_READONLY;
+       lock_ReleaseWrite(&scp->rw);
+        return code;
+    }
+
     /* otherwise, we have to make an RPC to get the status */
     code = cm_SyncOp(scp, NULL, userp, reqp, 0, CM_SCACHESYNC_STORESTATUS);
     if (code) {
@@ -2624,7 +2762,7 @@ long cm_SetAttr(cm_scache_t *scp, cm_attr_t *attrp, cm_user_t *userp,
     lock_ObtainWrite(&scp->rw);
     cm_SyncOpDone(scp, NULL, CM_SCACHESYNC_STORESTATUS);
     if (code == 0)
-        cm_MergeStatus(NULL, scp, &afsOutStatus, &volSync, userp,
+        cm_MergeStatus(NULL, scp, &afsOutStatus, &volSync, userp, reqp,
                         CM_MERGEFLAG_FORCE|CM_MERGEFLAG_STOREDATA);
        
     /* if we're changing the mode bits, discard the ACL cache, 
@@ -2656,6 +2794,8 @@ long cm_Create(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *a
     cm_dirOp_t dirop;
     fschar_t * fnamep = NULL;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     /* can't create names with @sys in them; must expand it manually first.
      * return "invalid request" if they try.
      */
@@ -2673,11 +2813,16 @@ long cm_Create(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *a
     }
 #endif /* AFS_FREELANCE_CLIENT */
 
+    /* Check for RO volume */
+    if (dscp->flags & CM_SCACHEFLAG_RO)
+        return CM_ERROR_READONLY;
+
     /* before starting the RPC, mark that we're changing the file data, so
      * that someone who does a chmod will know to wait until our call
      * completes.
      */
-    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, &dirop);
+    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, CM_DIROP_FLAG_NONE,
+                  &dirop);
     lock_ObtainWrite(&dscp->rw);
     code = cm_SyncOp(dscp, NULL, userp, reqp, 0, CM_SCACHESYNC_STOREDATA);
     lock_ReleaseWrite(&dscp->rw);
@@ -2729,7 +2874,7 @@ long cm_Create(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *a
     lock_ObtainWrite(&dscp->rw);
     cm_SyncOpDone(dscp, NULL, CM_SCACHESYNC_STOREDATA);
     if (code == 0) {
-        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, CM_MERGEFLAG_DIROP);
+        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, reqp, CM_MERGEFLAG_DIROP);
     }
     lock_ReleaseWrite(&dscp->rw);
 
@@ -2745,10 +2890,10 @@ long cm_Create(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *a
             lock_ObtainWrite(&scp->rw);
            scp->creator = userp;               /* remember who created it */
             if (!cm_HaveCallback(scp)) {
-                cm_MergeStatus(dscp, scp, &newFileStatus, &volSync,
-                               userp, 0);
                 cm_EndCallbackGrantingCall(scp, &cbReq,
-                                           &newFileCallback, 0);
+                                           &newFileCallback, &volSync, 0);
+                cm_MergeStatus(dscp, scp, &newFileStatus, &volSync,
+                               userp, reqp, 0);
                 didEnd = 1;     
             }       
             lock_ReleaseWrite(&scp->rw);
@@ -2757,7 +2902,7 @@ long cm_Create(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *a
 
     /* make sure we end things properly */
     if (!didEnd)
-        cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, 0);
+        cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, NULL, 0);
 
     if (scp && cm_CheckDirOpForSingleChange(&dirop)) {
         cm_DirCreateEntry(&dirop, fnamep, &newFid);
@@ -2779,10 +2924,16 @@ long cm_Create(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *a
     return code;
 }       
 
-long cm_FSync(cm_scache_t *scp, cm_user_t *userp, cm_req_t *reqp)
+/*
+ * locked if TRUE means write-locked
+ * else the cm_scache_t rw must not be held
+ */
+long cm_FSync(cm_scache_t *scp, cm_user_t *userp, cm_req_t *reqp, afs_uint32 locked)
 {
     long code;
 
+    if (locked)
+        lock_ReleaseWrite(&scp->rw);
     code = buf_CleanVnode(scp, userp, reqp);
     if (code == 0) {
         lock_ObtainWrite(&scp->rw);
@@ -2797,7 +2948,10 @@ long cm_FSync(cm_scache_t *scp, cm_user_t *userp, cm_req_t *reqp)
            scp->flags &= ~(CM_SCACHEFLAG_OVERQUOTA | CM_SCACHEFLAG_OUTOFSPACE);
        }
 
-        lock_ReleaseWrite(&scp->rw);
+        if (!locked)
+            lock_ReleaseWrite(&scp->rw);
+    } else if (locked) {
+        lock_ObtainWrite(&scp->rw);
     }
     return code;
 }
@@ -2822,6 +2976,8 @@ long cm_MakeDir(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *
     cm_dirOp_t dirop;
     fschar_t * fnamep = NULL;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     /* can't create names with @sys in them; must expand it manually first.
      * return "invalid request" if they try.
      */
@@ -2839,11 +2995,16 @@ long cm_MakeDir(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *
     }
 #endif /* AFS_FREELANCE_CLIENT */
 
+    /* Check for RO volume */
+    if (dscp->flags & CM_SCACHEFLAG_RO)
+        return CM_ERROR_READONLY;
+
     /* before starting the RPC, mark that we're changing the directory
      * data, so that someone who does a chmod on the dir will wait until
      * our call completes.
      */
-    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, &dirop);
+    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, CM_DIROP_FLAG_NONE,
+                  &dirop);
     lock_ObtainWrite(&dscp->rw);
     code = cm_SyncOp(dscp, NULL, userp, reqp, 0, CM_SCACHESYNC_STOREDATA);
     lock_ReleaseWrite(&dscp->rw);
@@ -2894,7 +3055,7 @@ long cm_MakeDir(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *
     lock_ObtainWrite(&dscp->rw);
     cm_SyncOpDone(dscp, NULL, CM_SCACHESYNC_STOREDATA);
     if (code == 0) {
-        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, CM_MERGEFLAG_DIROP);
+        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, reqp, CM_MERGEFLAG_DIROP);
     }
     lock_ReleaseWrite(&dscp->rw);
 
@@ -2909,10 +3070,10 @@ long cm_MakeDir(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *
         if (code == 0) {
             lock_ObtainWrite(&scp->rw);
             if (!cm_HaveCallback(scp)) {
-                cm_MergeStatus(dscp, scp, &newDirStatus, &volSync,
-                                userp, 0);
                 cm_EndCallbackGrantingCall(scp, &cbReq,
-                                            &newDirCallback, 0);
+                                            &newDirCallback, &volSync, 0);
+                cm_MergeStatus(dscp, scp, &newDirStatus, &volSync,
+                                userp, reqp, 0);
                 didEnd = 1;             
             }
             lock_ReleaseWrite(&scp->rw);
@@ -2921,7 +3082,7 @@ long cm_MakeDir(cm_scache_t *dscp, clientchar_t *cnamep, long flags, cm_attr_t *
 
     /* make sure we end things properly */
     if (!didEnd)
-        cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, 0);
+        cm_EndCallbackGrantingCall(NULL, &cbReq, NULL, NULL, 0);
 
     if (scp && cm_CheckDirOpForSingleChange(&dirop)) {
         cm_DirCreateEntry(&dirop, fnamep, &newFid);
@@ -2958,12 +3119,19 @@ long cm_Link(cm_scache_t *dscp, clientchar_t *cnamep, cm_scache_t *sscp, long fl
     cm_dirOp_t dirop;
     fschar_t * fnamep = NULL;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     if (dscp->fid.cell != sscp->fid.cell ||
         dscp->fid.volume != sscp->fid.volume) {
         return CM_ERROR_CROSSDEVLINK;
     }
 
-    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, &dirop);
+    /* Check for RO volume */
+    if (dscp->flags & CM_SCACHEFLAG_RO)
+        return CM_ERROR_READONLY;
+
+    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, CM_DIROP_FLAG_NONE,
+                  &dirop);
     lock_ObtainWrite(&dscp->rw);
     code = cm_SyncOp(dscp, NULL, userp, reqp, 0, CM_SCACHESYNC_STOREDATA);
     lock_ReleaseWrite(&dscp->rw);
@@ -3012,7 +3180,7 @@ long cm_Link(cm_scache_t *dscp, clientchar_t *cnamep, cm_scache_t *sscp, long fl
     lock_ObtainWrite(&dscp->rw);
     cm_SyncOpDone(dscp, NULL, CM_SCACHESYNC_STOREDATA);
     if (code == 0) {
-        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, CM_MERGEFLAG_DIROP);
+        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, reqp, CM_MERGEFLAG_DIROP);
     }
     lock_ReleaseWrite(&dscp->rw);
 
@@ -3026,6 +3194,13 @@ long cm_Link(cm_scache_t *dscp, clientchar_t *cnamep, cm_scache_t *sscp, long fl
     }
     cm_EndDirOp(&dirop);
 
+    /* Update the linked object status */
+    if (code == 0) {
+        lock_ObtainWrite(&sscp->rw);
+        cm_MergeStatus(NULL, sscp, &newLinkStatus, &volSync, userp, reqp, 0);
+        lock_ReleaseWrite(&sscp->rw);
+    }
+
     free(fnamep);
 
     return code;
@@ -3048,11 +3223,18 @@ long cm_SymLink(cm_scache_t *dscp, clientchar_t *cnamep, fschar_t *contentsp, lo
     cm_dirOp_t dirop;
     fschar_t *fnamep = NULL;
 
+    /* Check for RO volume */
+    if (dscp->flags & CM_SCACHEFLAG_RO)
+        return CM_ERROR_READONLY;
+
+    memset(&volSync, 0, sizeof(volSync));
+
     /* before starting the RPC, mark that we're changing the directory data,
      * so that someone who does a chmod on the dir will wait until our
      * call completes.
      */
-    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, &dirop);
+    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, CM_DIROP_FLAG_NONE,
+                  &dirop);
     lock_ObtainWrite(&dscp->rw);
     code = cm_SyncOp(dscp, NULL, userp, reqp, 0, CM_SCACHESYNC_STOREDATA);
     lock_ReleaseWrite(&dscp->rw);
@@ -3099,7 +3281,7 @@ long cm_SymLink(cm_scache_t *dscp, clientchar_t *cnamep, fschar_t *contentsp, lo
     lock_ObtainWrite(&dscp->rw);
     cm_SyncOpDone(dscp, NULL, CM_SCACHESYNC_STOREDATA);
     if (code == 0) {
-        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, CM_MERGEFLAG_DIROP);
+        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, reqp, CM_MERGEFLAG_DIROP);
     }
     lock_ReleaseWrite(&dscp->rw);
 
@@ -3127,7 +3309,7 @@ long cm_SymLink(cm_scache_t *dscp, clientchar_t *cnamep, fschar_t *contentsp, lo
             lock_ObtainWrite(&scp->rw);
             if (!cm_HaveCallback(scp)) {
                 cm_MergeStatus(dscp, scp, &newLinkStatus, &volSync,
-                                userp, 0);
+                                userp, reqp, 0);
             }       
             lock_ReleaseWrite(&scp->rw);
             cm_ReleaseSCache(scp);
@@ -3172,10 +3354,13 @@ long cm_RemoveDir(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t *cnamep, cm_
     cm_scache_t *scp = NULL;
     int free_fnamep = FALSE;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     if (fnamep == NULL) {
         code = -1;
 #ifdef USE_BPLUS
-        code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_READ, &dirop);
+        code = cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_READ,
+                             CM_DIROP_FLAG_NONE, &dirop);
         if (code == 0) {
             code = cm_BPlusDirLookupOriginalName(&dirop, cnamep, &fnamep);
             if (code == 0)
@@ -3191,11 +3376,18 @@ long cm_RemoveDir(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t *cnamep, cm_
     if (code)
         goto done;
 
+    /* Check for RO volume */
+    if (dscp->flags & CM_SCACHEFLAG_RO) {
+        code = CM_ERROR_READONLY;
+        goto done;
+    }
+
     /* before starting the RPC, mark that we're changing the directory data,
      * so that someone who does a chmod on the dir will wait until our
      * call completes.
      */
-    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, &dirop);
+    cm_BeginDirOp(dscp, userp, reqp, CM_DIRLOCK_NONE, CM_DIROP_FLAG_NONE,
+                  &dirop);
     lock_ObtainWrite(&dscp->rw);
     code = cm_SyncOp(dscp, NULL, userp, reqp, 0, CM_SCACHESYNC_STOREDATA);
     lock_ReleaseWrite(&dscp->rw);
@@ -3238,7 +3430,7 @@ long cm_RemoveDir(cm_scache_t *dscp, fschar_t *fnamep, clientchar_t *cnamep, cm_
     cm_SyncOpDone(dscp, NULL, CM_SCACHESYNC_STOREDATA);
     if (code == 0) {
         cm_dnlcRemove(dscp, cnamep); 
-        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, CM_MERGEFLAG_DIROP);
+        cm_MergeStatus(NULL, dscp, &updatedDirStatus, &volSync, userp, reqp, CM_MERGEFLAG_DIROP);
     }
     lock_ReleaseWrite(&dscp->rw);
 
@@ -3333,6 +3525,8 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
     int free_oldNamep = FALSE;
     cm_scache_t *oldScp = NULL, *newScp = NULL;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     if (cOldNamep == NULL || cNewNamep == NULL ||
         cm_ClientStrLen(cOldNamep) == 0 ||
         cm_ClientStrLen(cNewNamep) == 0)
@@ -3366,13 +3560,21 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
     } else {
         code = 0;
     }
+
+    /* Check for RO volume */
+    if (code == 0 &&
+        (oldDscp->flags & CM_SCACHEFLAG_RO) || (newDscp->flags & CM_SCACHEFLAG_RO)) {
+        code = CM_ERROR_READONLY;
+    }
+
     if (code) 
         goto done;
 
     if (oldNamep == NULL) {
         code = -1;
 #ifdef USE_BPLUS
-        code = cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_READ, &oldDirOp);
+        code = cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_READ,
+                             CM_DIROP_FLAG_NONE, &oldDirOp);
         if (code == 0) {
             code = cm_BPlusDirLookupOriginalName(&oldDirOp, cOldNamep, &oldNamep);
             if (code == 0)
@@ -3403,7 +3605,8 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
         }
 
         oneDir = 1;
-        cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_NONE, &oldDirOp);
+        cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_NONE,
+                      CM_DIROP_FLAG_NONE, &oldDirOp);
         lock_ObtainWrite(&oldDscp->rw);
         cm_dnlcRemove(oldDscp, cOldNamep);
         cm_dnlcRemove(oldDscp, cNewNamep);
@@ -3437,7 +3640,8 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
         }
 
         if (oldDscp->fid.vnode < newDscp->fid.vnode) {
-            cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_NONE, &oldDirOp);
+            cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_NONE,
+                          CM_DIROP_FLAG_NONE, &oldDirOp);
             lock_ObtainWrite(&oldDscp->rw);
             cm_dnlcRemove(oldDscp, cOldNamep);
             code = cm_SyncOp(oldDscp, NULL, userp, reqp, 0,
@@ -3446,7 +3650,8 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
             if (code != 0)
                 cm_EndDirOp(&oldDirOp);
             if (code == 0) {
-                cm_BeginDirOp(newDscp, userp, reqp, CM_DIRLOCK_NONE, &newDirOp);
+                cm_BeginDirOp(newDscp, userp, reqp, CM_DIRLOCK_NONE,
+                              CM_DIROP_FLAG_NONE, &newDirOp);
                 lock_ObtainWrite(&newDscp->rw);
                 cm_dnlcRemove(newDscp, cNewNamep);
                 code = cm_SyncOp(newDscp, NULL, userp, reqp, 0,
@@ -3466,7 +3671,8 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
         }
         else {
             /* lock the new vnode entry first */
-            cm_BeginDirOp(newDscp, userp, reqp, CM_DIRLOCK_NONE, &newDirOp);
+            cm_BeginDirOp(newDscp, userp, reqp, CM_DIRLOCK_NONE,
+                          CM_DIROP_FLAG_NONE, &newDirOp);
             lock_ObtainWrite(&newDscp->rw);
             cm_dnlcRemove(newDscp, cNewNamep);
             code = cm_SyncOp(newDscp, NULL, userp, reqp, 0,
@@ -3475,7 +3681,8 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
             if (code != 0)
                 cm_EndDirOp(&newDirOp);
             if (code == 0) {
-                cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_NONE, &oldDirOp);
+                cm_BeginDirOp(oldDscp, userp, reqp, CM_DIRLOCK_NONE,
+                              CM_DIROP_FLAG_NONE, &oldDirOp);
                 lock_ObtainWrite(&oldDscp->rw);
                 cm_dnlcRemove(oldDscp, cOldNamep);
                 code = cm_SyncOp(oldDscp, NULL, userp, reqp, 0,
@@ -3543,7 +3750,7 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
 
     if (code == 0)
         cm_MergeStatus(NULL, oldDscp, &updatedOldDirStatus, &volSync,
-                       userp, CM_MERGEFLAG_DIROP);
+                       userp, reqp, CM_MERGEFLAG_DIROP);
     lock_ReleaseWrite(&oldDscp->rw);
 
     if (code == 0 && cm_CheckDirOpForSingleChange(&oldDirOp)) {
@@ -3583,7 +3790,7 @@ long cm_Rename(cm_scache_t *oldDscp, fschar_t *oldNamep, clientchar_t *cOldNamep
         cm_SyncOpDone(newDscp, NULL, CM_SCACHESYNC_STOREDATA);
         if (code == 0)
             cm_MergeStatus(NULL, newDscp, &updatedNewDirStatus, &volSync,
-                            userp, CM_MERGEFLAG_DIROP);
+                            userp, reqp, CM_MERGEFLAG_DIROP);
         lock_ReleaseWrite(&newDscp->rw);
 
 #if 0 
@@ -4155,6 +4362,8 @@ long cm_IntSetLock(cm_scache_t * scp, cm_user_t * userp, int lockType,
     AFSVolSync volSync;
     afs_uint32 reqflags = reqp->flags;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     tfid.Volume = scp->fid.volume;
     tfid.Vnode = scp->fid.vnode;
     tfid.Unique = scp->fid.unique;
@@ -4200,6 +4409,8 @@ long cm_IntReleaseLock(cm_scache_t * scp, cm_user_t * userp,
     struct rx_connection * rxconnp;
     AFSVolSync volSync;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     tfid.Volume = scp->fid.volume;
     tfid.Vnode = scp->fid.vnode;
     tfid.Unique = scp->fid.unique;
@@ -4681,6 +4892,10 @@ long cm_Lock(cm_scache_t *scp, unsigned char sLockType,
                  "cm_Lock Rejecting lock (code = 0x%x)", code);
     }
 
+    /* Convert from would block to lock not granted */
+    if (code == CM_ERROR_WOULDBLOCK)
+        code = CM_ERROR_LOCK_NOT_GRANTED;
+
     return code;
 }
 
@@ -4797,10 +5012,12 @@ long cm_UnlockByKey(cm_scache_t * scp,
     if (scp->serverLock == LockWrite &&
         scp->exclusiveLocks == 0 &&
         scp->sharedLocks > 0) {
-
         /* The serverLock should be downgraded to LockRead */
         osi_Log0(afsd_logp, "  DOWNGRADE lock from LockWrite to LockRead");
 
+        /* Make sure there are no dirty buffers left. */
+        code = cm_FSync(scp, userp, reqp, TRUE);
+
         /* since scp->serverLock looked sane, we are going to assume
            that we have a valid server lock. */
         scp->lockDataVersion = scp->dataVersion;
@@ -4852,6 +5069,11 @@ long cm_UnlockByKey(cm_scache_t * scp,
               scp->sharedLocks == 0) {
         /* The serverLock should be released entirely */
 
+        if (scp->serverLock == LockWrite) {
+            /* Make sure there are no dirty buffers left. */
+            code = cm_FSync(scp, userp, reqp, TRUE);
+        }
+
         code = cm_IntReleaseLock(scp, userp, reqp);
 
         if (code == 0)
@@ -4872,6 +5094,7 @@ long cm_Unlock(cm_scache_t *scp,
                unsigned char sLockType,
                LARGE_INTEGER LOffset, LARGE_INTEGER LLength,
                cm_key_t key, 
+               afs_uint32 flags,
                cm_user_t *userp, 
                cm_req_t *reqp)
 {
@@ -4880,12 +5103,19 @@ long cm_Unlock(cm_scache_t *scp,
     cm_file_lock_t *fileLock;
     osi_queue_t *q;
     int release_userp = FALSE;
+    int exact_match = !(flags & CM_UNLOCK_FLAG_MATCH_RANGE);
+    int lock_found  = 0;
+    LARGE_INTEGER RangeEnd;
 
     osi_Log4(afsd_logp, "cm_Unlock scp 0x%p type 0x%x offset %d length %d",
              scp, sLockType, (unsigned long)LOffset.QuadPart, (unsigned long)LLength.QuadPart);
-    osi_Log3(afsd_logp, "... key <0x%x,0x%x,0x%x>",
-             key.process_id, key.session_id, key.file_id);
+    osi_Log4(afsd_logp, "... key <0x%x,0x%x,0x%x> flags 0x%x",
+             key.process_id, key.session_id, key.file_id, flags);
 
+    if (!exact_match)
+        RangeEnd.QuadPart = LOffset.QuadPart + LLength.QuadPart;
+
+  try_again:
     lock_ObtainRead(&cm_scacheLock);
 
     for (q = scp->fileLocksH; q; q = osi_QNext(q)) {
@@ -4908,21 +5138,39 @@ long cm_Unlock(cm_scache_t *scp,
             osi_assertx(FALSE, "invalid fid value");
         }
 #endif
-        if (!IS_LOCK_DELETED(fileLock) &&
-            cm_KeyEquals(&fileLock->key, &key, 0) &&
-            fileLock->range.offset == LOffset.QuadPart &&
-            fileLock->range.length == LLength.QuadPart) {
-            break;
+        if (exact_match) {
+            if (!IS_LOCK_DELETED(fileLock) &&
+                 cm_KeyEquals(&fileLock->key, &key, 0) &&
+                 fileLock->range.offset == LOffset.QuadPart &&
+                 fileLock->range.length == LLength.QuadPart) {
+                lock_found = 1;
+                break;
+            }
+        } else {
+
+            if (!IS_LOCK_DELETED(fileLock) &&
+                 cm_KeyEquals(&fileLock->key, &key, 0) &&
+                 fileLock->range.offset >= LOffset.QuadPart &&
+                 fileLock->range.offset < RangeEnd.QuadPart &&
+                 (fileLock->range.offset + fileLock->range.length) <= RangeEnd.QuadPart) {
+                lock_found = 1;
+                break;
+            }
         }
     }
 
     if (!q) {
-        osi_Log0(afsd_logp, "cm_Unlock lock not found; failure");
-        
         lock_ReleaseRead(&cm_scacheLock);
 
-        /* The lock didn't exist anyway. *shrug* */
-        return CM_ERROR_RANGE_NOT_LOCKED;
+        if (lock_found && !exact_match) {
+            code = 0;
+            goto done;
+        } else {
+            osi_Log0(afsd_logp, "cm_Unlock lock not found; failure");
+        
+            /* The lock didn't exist anyway. *shrug* */
+            return CM_ERROR_RANGE_NOT_LOCKED;
+        }
     }
 
     /* discard lock record */
@@ -4976,6 +5224,9 @@ long cm_Unlock(cm_scache_t *scp,
         /* The serverLock should be downgraded to LockRead */
         osi_Log0(afsd_logp, "  DOWNGRADE lock from LockWrite to LockRead");
 
+        /* Make sure there are no dirty buffers left. */
+        code = cm_FSync(scp, userp, reqp, TRUE);
+
         /* Since we already had a lock, we assume that there is a
            valid server lock. */
         scp->lockDataVersion = scp->dataVersion;
@@ -5040,6 +5291,11 @@ long cm_Unlock(cm_scache_t *scp,
               scp->sharedLocks == 0) {
         /* The serverLock should be released entirely */
 
+        if (scp->serverLock == LockWrite) {
+            /* Make sure there are no dirty buffers left. */
+            code = cm_FSync(scp, userp, reqp, TRUE);
+        }
+
         code = cm_IntReleaseLock(scp, userp, reqp);
 
         if (code == 0) {
@@ -5047,8 +5303,13 @@ long cm_Unlock(cm_scache_t *scp,
         }
     }
 
-    if (release_userp)
+    if (release_userp) {
         cm_ReleaseUser(userp);
+        release_userp = FALSE;
+    }
+
+    if (!exact_match)
+        goto try_again;         /* might be more than one lock in the range */
 
  done:
 
@@ -5104,6 +5365,8 @@ void cm_CheckLocks()
     struct rx_connection * rxconnp;
     cm_scache_t * scp;
 
+    memset(&volSync, 0, sizeof(volSync));
+
     cm_InitReq(&req);
 
     lock_ObtainWrite(&cm_scacheLock);