From: Mark Vitale <mvitale@sinenomine.net>
Date: Mon, 29 Oct 2018 20:48:14 +0000 (-0400)
Subject: afs: avoid afs_GetDownDSlot panic on afs_WriteDCache failure
X-Git-Tag: openafs-devel-1_9_0~392
X-Git-Url: https://git.openafs.org/?p=openafs.git;a=commitdiff_plain;h=d6f52d11c358f71b2c4357cb135e898de7c6277b

afs: avoid afs_GetDownDSlot panic on afs_WriteDCache failure

If afs_GetDownDSlot() finds insuffcient free slots in the
afs_freeDSList, it will walk the afs_DLRU attempting to flush and free
eligible dcaches.  However, if an error occurs during the flush to
CacheItems (afs_WriteDCache()), e.g., -EINTR, afs_GetDownDSlot() will
assert.

However, a panic in this case is overkill, since afs_GetDownDSlot() is a
best-effort attempt to free dslots.  The caller (afs_UFSGetDSlot()) will
allocate more dcaches if needed.

Instead:
- Refactor afs_GetDownDSlot() by moving the QRemove() call to after the
afs_WriteDCache logic, so it accompanies the logic that puts the dcache
back on the freelist.  This is safe because we hold the afs_xdcache W
lock for the duration of the routine.
- If afs_WriteDCache() returns an error, return early and let the caller
handle any recovery.

Change-Id: Ifd0d56120095c9792998ff935776bbd339a76c8a
Reviewed-on: https://gerrit.openafs.org/13364
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Tested-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
---

diff --git a/src/afs/afs_dcache.c b/src/afs/afs_dcache.c
index 4b93915..15e63ff 100644
--- a/src/afs/afs_dcache.c
+++ b/src/afs/afs_dcache.c
@@ -1282,8 +1282,6 @@ afs_GetDownDSlot(int anumber)
 	if (tdc->refCount == 0) {
 	    if ((ix = tdc->index) == NULLIDX)
 		osi_Panic("getdowndslot");
-	    /* pull the entry out of the lruq and put it on the free list */
-	    QRemove(&tdc->lruq);
 
 	    /* write-through if modified */
 	    if (tdc->dflags & DFEntryMod) {
@@ -1305,12 +1303,23 @@ afs_GetDownDSlot(int anumber)
 		    AFS_GLOCK();
 		}
 #else
+		int code;
+
+		code = afs_WriteDCache(tdc, 1);
+		if (code) {
+		    /*
+		     * We couldn't flush it at this time; return early because
+		     * if afs_WriteDCache() failed once it is likely to
+		     * continue failing for subsequent dcaches.
+		     */
+		    return;
+		}
 		tdc->dflags &= ~DFEntryMod;
-		osi_Assert(afs_WriteDCache(tdc, 1) == 0);
 #endif
 	    }
 
-	    /* finally put the entry in the free list */
+	    /* pull the entry out of the lruq and put it on the free list */
+	    QRemove(&tdc->lruq);
 	    afs_indexTable[ix] = NULL;
 	    afs_indexFlags[ix] &= ~IFEverUsed;
 	    tdc->index = NULLIDX;