Missing some ZFS values in sysctl output

[Willem Jan Withagen]

On 22-3-2015 23:44, Willem Jan Withagen wrote:
> Hoi,
> 
> There is this zfs-snmp script that allows one the graph certain things
> from a ZFS setup.
> The original uses solaris' kstat, but just about all values are in
> Sysctl with FreeBSD.
> 
> However I seem to overlook the values for:
> 
> Original code:
> ====
> def zfs_read(oid):
>     return ('counter', kstat("unix:0:vopstats_zfs:read_bytes") / 1024 %
> 2**32) # 32 bit KB counter
> 
> def zfs_readdir(oid):
>     return ('counter', kstat("unix:0:vopstats_zfs:readdir_bytes") / 1024
> % 2**32) # 32 bit KB counter
> 
> def zfs_write(oid):
>     return ('counter', kstat("unix:0:vopstats_zfs:write_bytes") / 1024 %
> 2**32) # 32 bit KB counter
> =======
> 
> Read and write are available for L2 arc, but for the global values I
> seem to miss them.
> 
> Are they really not in ZFS?
> And is it possible to add them?

Right,

Just to answer my own question....

I looked into the illumos code, and it looks like the vopstats are stats
that are kept per filesystem-type on solaris, and these are the ZFS
variants....

typedef struct vopstats {
        kstat_named_t   nopen;          /* VOP_OPEN */
        kstat_named_t   nclose;         /* VOP_CLOSE */
        kstat_named_t   nread;          /* VOP_READ */
        kstat_named_t   read_bytes;
        kstat_named_t   nwrite;         /* VOP_WRITE */
        kstat_named_t   write_bytes;
        kstat_named_t   nioctl;         /* VOP_IOCTL */
        kstat_named_t   nsetfl;         /* VOP_SETFL */
        kstat_named_t   ngetattr;       /* VOP_GETATTR */
        kstat_named_t   nsetattr;       /* VOP_SETATTR */
        kstat_named_t   naccess;        /* VOP_ACCESS */
        kstat_named_t   nlookup;        /* VOP_LOOKUP */
        kstat_named_t   ncreate;        /* VOP_CREATE */
        kstat_named_t   nremove;        /* VOP_REMOVE */
        kstat_named_t   nlink;          /* VOP_LINK */
        kstat_named_t   nrename;        /* VOP_RENAME */
        kstat_named_t   nmkdir;         /* VOP_MKDIR */
        kstat_named_t   nrmdir;         /* VOP_RMDIR */
        kstat_named_t   nreaddir;       /* VOP_READDIR */
        kstat_named_t   readdir_bytes;
        kstat_named_t   nsymlink;       /* VOP_SYMLINK */
        kstat_named_t   nreadlink;      /* VOP_READLINK */
        kstat_named_t   nfsync;         /* VOP_FSYNC */
        kstat_named_t   ninactive;      /* VOP_INACTIVE */
        kstat_named_t   nfid;           /* VOP_FID */
        kstat_named_t   nrwlock;        /* VOP_RWLOCK */
        kstat_named_t   nrwunlock;      /* VOP_RWUNLOCK */
        kstat_named_t   nseek;          /* VOP_SEEK */
        kstat_named_t   ncmp;           /* VOP_CMP */
        kstat_named_t   nfrlock;        /* VOP_FRLOCK */
        kstat_named_t   nspace;         /* VOP_SPACE */
        kstat_named_t   nrealvp;        /* VOP_REALVP */
        kstat_named_t   ngetpage;       /* VOP_GETPAGE */
        kstat_named_t   nputpage;       /* VOP_PUTPAGE */
        kstat_named_t   nmap;           /* VOP_MAP */
        kstat_named_t   naddmap;        /* VOP_ADDMAP */
        kstat_named_t   ndelmap;        /* VOP_DELMAP */
        kstat_named_t   npoll;          /* VOP_POLL */
        kstat_named_t   ndump;          /* VOP_DUMP */
        kstat_named_t   npathconf;      /* VOP_PATHCONF */
        kstat_named_t   npageio;        /* VOP_PAGEIO */
        kstat_named_t   ndumpctl;       /* VOP_DUMPCTL */
        kstat_named_t   ndispose;       /* VOP_DISPOSE */
        kstat_named_t   nsetsecattr;    /* VOP_SETSECATTR */
        kstat_named_t   ngetsecattr;    /* VOP_GETSECATTR */
        kstat_named_t   nshrlock;       /* VOP_SHRLOCK */
        kstat_named_t   nvnevent;       /* VOP_VNEVENT */
        kstat_named_t   nreqzcbuf;      /* VOP_REQZCBUF */
        kstat_named_t   nretzcbuf;      /* VOP_RETZCBUF */
} vopstats_t;

As far as I can tell we do not have these structures in vnode.{c,h} and
as such do not keep tracks of these statistics. Nothing equal to this
jumps out from the sysctl-tree. But I happily stand corrected.

Tinkering this in just the ZFS code would be a serious undertaking. At
least for me. And all this just to some snmp-stats.

--WjW