Linux: Restrict # of cbrs we allocate at once

With commit a309e274632993c5aeec04c6e090f5ac95837a40, we changed the
number of CBRs that we allocate in a chunk from 300 to 1024. However,
this change takes the amount of memory requried to allocate a chunk
of CBRs above PAGE_SIZE on Linux. This changes the allocator that we
use from kmalloc to vmalloc. Whilst we can, and do, prevent kmalloc
from flushing filesystem pages when we invoke it, we don't have a
similar level of control over vmalloc.

In one reported case, clients deadlock whilst attempting to allocate
this memory, in a call stack that looks something like:

  afs_Daemon -> afs_ShakeLooseVCaches -> osi_TryEvictVCache
      -> afs_FlushVCache -> afs_QueueVCB -> afs_AllocCBR
      -> osi_linux_alloc -> ... -> __vmalloc_node -> ...
      -> try_to_free_pages

This is probably because we end up deadlocking in the writeback
invoked by try_to_free_pages, likely due to locks which are held
by ShakeLooseVCaches.

As a quick fix for the problem, make sure that the memory we
allocate always fits within a page, so we always use kmalloc for
it.

FIXES 129751

Change-Id: I8b5be06ed62764b67f99792df66e9cb0f8941fd6
Reviewed-on: http://gerrit.openafs.org/4510
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Jeffrey Altman <jaltman@openafs.org>
Reviewed-by: Derrick Brashear <shadow@dementia.org>

diff --git a/src/afs/afs.h b/src/afs/afs.h
index c135a1b..13614cb 100644 (file)
--- a/src/afs/afs.h
+++ b/src/afs/afs.h
@@ -85,7 +85,9 @@ extern int afs_shuttingdown;
 #define MAXNUMSYSNAMES 32      /* max that current constants allow */
 #define        NOTOKTIMEOUT    (2*3600)        /* time after which to timeout conns sans tokens */
 #define        NOPAG           0xffffffff
-#define AFS_NCBRS      1024    /* max # of call back return entries */
+
+
+
 #define AFS_MAXCBRSCALL        32      /* max to return in a given call (must be <= AFSCBMAX) */
 #define        AFS_SALLOC_LOW_WATER    250     /* Min free blocks before allocating more */
 #define        AFS_LRALLOCSIZ  4096    /* "Large" allocated size */
@@ -261,6 +263,16 @@ struct afs_cbr {
     unsigned int dynalloc:1;
 };
 
+#ifdef AFS_LINUX22_ENV
+/* On Linux, we have to be able to allocate the storage for this using
+ * kmalloc, as otherwise we may deadlock. So, it needs to be able to fit
+ * in a single page
+ */
+# define AFS_NCBRS     PAGE_SIZE/sizeof(struct afs_cbr)
+#else
+# define AFS_NCBRS     1024    /* max # of call back return entries */
+#endif
+
 /* cellinfo file magic number */
 #define AFS_CELLINFO_MAGIC     0xf32817cd