[OpenAFS-devel] lots of vmalloc failures lately...

[chas williams (contractor)]

In message <5C51DC2B8353AB4BA2CD04B34F2EE79C3EFF20@umr-umail1.umr.edu>,"Neuling
er, Nathan" writes:
>I've been seeing a lot more vmalloc failures on a couple of my boxes
>recently... running with a 350MB disk cache.
>
>Unfortunately, the result is almost always that AFS on the box becomes
>completely unusable, and the load usually shoots through the roof over
>the next few minutes.

i have a patch you could try --

Index: src/afs/afs.h
===================================================================
RCS file: /cvs/openafs/src/afs/afs.h,v
retrieving revision 1.45
diff -u -u -r1.45 afs.h
--- src/afs/afs.h	5 Apr 2004 22:39:51 -0000	1.45
+++ src/afs/afs.h	16 Apr 2004 21:18:59 -0000
@@ -926,6 +926,7 @@
 /* kept in memory */
 struct dcache {
     struct afs_q lruq;		/* Free queue for in-memory images */
+    struct afs_q dirty;		/* Queue of dirty entries that need written */
     afs_rwlock_t lock;		/* Protects validPos, some f */
     afs_rwlock_t tlock;		/* Atomizes updates to refCount */
     afs_rwlock_t mflock;	/* Atomizes accesses/updates to mflags */
Index: src/afs/afs_dcache.c
===================================================================
RCS file: /cvs/openafs/src/afs/afs_dcache.c,v
retrieving revision 1.40
diff -u -u -r1.40 afs_dcache.c
--- src/afs/afs_dcache.c	15 Jul 2003 23:14:12 -0000	1.40
+++ src/afs/afs_dcache.c	16 Apr 2004 21:18:59 -0000
@@ -2582,8 +2582,8 @@
 {
     register struct dcache *tdc;
     register afs_int32 i, touchedit = 0;
-    struct dcache **ents;
-    int entmax, entcount;
+
+    struct afs_q DirtyQ, *tq;
 
     AFS_STATCNT(afs_WriteThroughDSlots);
 
@@ -2593,9 +2593,7 @@
      * for every dcache entry, and exit xdcache.
      */
     MObtainWriteLock(&afs_xdcache, 283);
-    entmax = afs_cacheFiles;
-    ents = afs_osi_Alloc(entmax * sizeof(struct dcache *));
-    entcount = 0;
+    QInit(&DirtyQ);
     for (i = 0; i < afs_cacheFiles; i++) {
 	tdc = afs_indexTable[i];
 
@@ -2605,7 +2603,7 @@
 	    tdc->refCount++;
 	    ReleaseWriteLock(&tdc->tlock);
 
-	    ents[entcount++] = tdc;
+	    QAdd(&DirtyQ, &tdc->dirty);
 	}
     }
     MReleaseWriteLock(&afs_xdcache);
@@ -2616,9 +2614,11 @@
      * afs_cacheInodep, and flush it.  Don't forget to put back
      * the refcounts.
      */
-    for (i = 0; i < entcount; i++) {
-	tdc = ents[i];
 
+#define DQTODC(q)	((struct dcache *)(((char *) (q)) - sizeof(struct afs_q)))
+
+    for (tq = DirtyQ.prev; tq != &DirtyQ; tq = QPrev(tq)) {
+        tdc = DQTODC(tq);
 	if (tdc->dflags & DFEntryMod) {
 	    int wrLock;
 
@@ -2638,7 +2638,6 @@
 
 	afs_PutDCache(tdc);
     }
-    afs_osi_Free(ents, entmax * sizeof(struct dcache *));
 
     MObtainWriteLock(&afs_xdcache, 617);
     if (!touchedit && (cacheDiskType != AFS_FCACHE_TYPE_MEM)) {