nfsstat printing overflow

nfsstat printing overflow

[Michael Griffith]

[-- Attachment #1: Type: text/plain, Size: 131 bytes --]

A small patch to handle when the RPC call counts are > 2^32/100.

-- 
Michael A. Griffith
<grif@acm.org>
408-907-2151
650-292-1516

[-- Attachment #2: nfsstat.diff --]
[-- Type: text/plain, Size: 2042 bytes --]



Without this patch, nfsstat reports the wrong percentage for
any RPC which has more than 2^32/100 calls.  For example
my machine running 1.0.1 reports:

l-sim-208-11:diag/testgen/batch_0/nv40_curie_ogtest/000001> nfsstat at
-c
Client rpc stats:
calls      retrans    authrefrsh
61096094   181        0
Client nfs v2:
null       getattr    setattr    root       lookup     readlink
0       0% 0       0% 0       0% 0       0% 0       0% 0       0%
read       wrcache    write      create     remove     rename
0       0% 0       0% 0       0% 0       0% 0       0% 0       0%
link       symlink    mkdir      rmdir      readdir    fsstat
0       0% 0       0% 0       0% 0       0% 0       0% 0       0%

Client nfs v3:
null       getattr    setattr    lookup     access     readlink
0       0% 2661431  4% 76995   0% 4938464  8% 3254610  5% 131     0%
read       write      create     mkdir      symlink    mknod
138409  0% 259382  0% 185867  0% 3570    0% 49      0% 0       0%
remove     rmdir      rename     link       readdir    readdirplus
4035088  6% 318094  0% 157     0% 0       0% 1626490  2% 43597144  1%
fsstat     fsinfo     pathconf   commit
18      0% 195     0% 0       0% 0       0%


Note that readdir plus should be 71%, not 1%. 



--- 1.0.1/nfsstat.c	Sun May  5 16:33:30 2002
+++ nfsstat.c	Thu May 29 16:10:21 2003
@@ -307,7 +307,8 @@
 print_callstats(const char *hdr, const char **names,
 				 unsigned int *info, unsigned int nr)
 {
-	unsigned int	total;
+	unsigned long long	total;
+        unsigned long long      pct;
 	int		i, j;
 
 	fputs(hdr, stdout);
@@ -319,9 +320,10 @@
 		for (j = 0; j < 6 && i + j < nr; j++)
 			printf("%-11s", names[i+j]);
 		printf("\n");
-		for (j = 0; j < 6 && i + j < nr; j++)
-			printf("%-6d %2d%% ",
-				info[i+j], 100 * info[i+j] / total);
+		for (j = 0; j < 6 && i + j < nr; j++) {
+                        pct = ((unsigned long long) info[i+j]*100)/total;
+                        printf("%-6d %2d%% ", info[i+j], pct);
+                }
 		printf("\n");
 	}
 	printf("\n");

Re: nfsstat printing overflow

[Neil Brown]

On Thursday May 29, grif@michaelgriffith.com wrote:
> A small patch to handle when the RPC call counts are > 2^32/100.
> 
Thanks.
However:

> +        unsigned long long      pct;

> +		for (j = 0; j < 6 && i + j < nr; j++) {
> +                        pct = ((unsigned long long) info[i+j]*100)/total;
> +                        printf("%-6d %2d%% ", info[i+j], pct);
> +                }

if pct is an "unsigned long long" we should use "%llu" to print it.

The following patch has been commited to the nfs-utils CVS and will be
in the next release, which will probably be 1.1.0

NeilBrown

Index: ChangeLog
===================================================================
RCS file: /cvsroot/nfs/nfs-utils/ChangeLog,v
retrieving revision 1.202
diff -u -r1.202 ChangeLog
--- ChangeLog	30 May 2003 05:16:52 -0000	1.202
+++ ChangeLog	30 May 2003 05:38:44 -0000
@@ -1,3 +1,9 @@
+2003-05-30 Michael Griffith <grif@michaelgriffith.com>
+	NeilBrown <neilb@cse.unsw.edu.au>
+
+	utils/nfsstat/nfsstat.c(print_callstats): use unsigned
+	long long to avoid overflow when printing stats.
+	
 2003-05-30  NeilBrown <neilb@cse.unsw.edu.au>
 
 	* support/export/export.c, support/include/nfslib.h,
Index: utils/nfsstat/nfsstat.c
===================================================================
RCS file: /cvsroot/nfs/nfs-utils/utils/nfsstat/nfsstat.c,v
retrieving revision 1.4
diff -u -r1.4 nfsstat.c
--- utils/nfsstat/nfsstat.c	5 May 2002 23:33:30 -0000	1.4
+++ utils/nfsstat/nfsstat.c	30 May 2003 05:38:46 -0000
@@ -307,7 +307,8 @@
 print_callstats(const char *hdr, const char **names,
 				 unsigned int *info, unsigned int nr)
 {
-	unsigned int	total;
+	unsigned long long	total;
+        unsigned long long      pct;
 	int		i, j;
 
 	fputs(hdr, stdout);
@@ -319,9 +320,10 @@
 		for (j = 0; j < 6 && i + j < nr; j++)
 			printf("%-11s", names[i+j]);
 		printf("\n");
-		for (j = 0; j < 6 && i + j < nr; j++)
-			printf("%-6d %2d%% ",
-				info[i+j], 100 * info[i+j] / total);
+		for (j = 0; j < 6 && i + j < nr; j++) {
+                        pct = ((unsigned long long) info[i+j]*100)/total;
+                        printf("%-6d %2llu%% ", info[i+j], pct);
+                }
 		printf("\n");
 	}
 	printf("\n");


-------------------------------------------------------
This SF.net email is sponsored by: eBay
Get office equipment for less on eBay!
http://adfarm.mediaplex.com/ad/ck/711-11697-6916-5
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: Re: nfsstat printing overflow

[Greg Lindahl]

On Fri, May 30, 2003 at 03:43:00PM +1000, Neil Brown wrote:
> On Thursday May 29, grif@michaelgriffith.com wrote:
> > A small patch to handle when the RPC call counts are > 2^32/100.
> > 
> Thanks.
> However:
> 
> > +        unsigned long long      pct;
> 
> > +		for (j = 0; j < 6 && i + j < nr; j++) {
> > +                        pct = ((unsigned long long) info[i+j]*100)/total;
> > +                        printf("%-6d %2d%% ", info[i+j], pct);
> > +                }

It feels like overkill to use long long for this when you can get
just as good of behavior with:

	unsigned long toto100;

	if (total >= 100)
	   toto100 = total / 100;
	else
	   toto100 = 1;

	for (j = 0; j < 6 && i + j < nr; j++)
              printf("%-6d %2d%% ", info[i+j], info[i+j]/toto100);

In both pieces of code, if total overflows you get the wrong answer. So you
didn't get much of a benefit from the long long.

If you care about getting the %s reasonable when total is small, you can
move the if() inside the for().

Doesn't gcc inline code on archs without real 64-bit ops? Yech.

greg



-------------------------------------------------------
This SF.net email is sponsored by: eBay
Get office equipment for less on eBay!
http://adfarm.mediaplex.com/ad/ck/711-11697-6916-5
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs