Using fstatfs on a ZFS disk

[Willem Jan Withagen]

On 19-2-2018 19:33, Rodney W. Grimes wrote:
>> Hi,
>>
>>> On 19 Feb 2018, at 15:50, Willem Jan Withagen <wjw at digiware.nl> wrote:
>>>
>>> On 19-2-2018 16:00, Bob Bishop wrote:
>>>> Hi,
>>>>> On 19 Feb 2018, at 14:48, Willem Jan Withagen <wjw at digiware.nl> wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>> I'm trying to find the values of the returned f_type for ZFS
>>>>> in the fstatfs call when a file is on ZFS....
>>>>>
>>>>> But I have not yet found the definitions of the ENUMS that
>>>>> would fill that value... Let alone the value for ZFS.
>>>> I chased this particular wild goose myself recently. It?s FS_... in  /usr/include/sys/disklabel,h that you want.
>>>
>>> Hi Bob,
>>>
>>> I grepped on MAGIC and FS, but the combo did not deliver anything useful. So this is already more that I found.
>>> I did get:
>>> /usr/include/ufs/ffs/fs.h:#define       FS_UFS1_MAGIC   0x011954 /* UFS1 fast filesystem magic number */
>>> /usr/include/ufs/ffs/fs.h:#define       FS_UFS2_MAGIC   0x19540119 /* UFS2 fast filesystem magic number */
>>> /usr/include/ufs/ffs/fs.h:#define       FS_BAD_MAGIC    0x19960408 /* UFS incomplete newfs magic number */
>>
>> Those I believe are magic numbers for UFS superblocks...
>>
>>> So I was looking for something like: FS_ZFS_MAGIC
>>
>> ... so you won?t find that.
>>
>>> disklabel.h contains:
>>> #ifdef  FSTYPENAMES
>>> static const char *fstypenames[] = {
>>>
>>> And further search:
>>> /usr/include/sys/disk/bsd.h:#define     FS_ZFS  27    /* Sun's ZFS */
>>>
>>> Running:
>>> #include "stdio.h"
>>>
>>> #include <sys/param.h>
>>> #include <sys/mount.h>
>>>
>>> int main() {
>>>         struct statfs fstr;
>>>         char * str;
>>>
>>>         str = "/tmp";
>>>         statfs(str, &fstr);
>>>         printf("%s, ftype: 0x%x.\n", str, fstr.f_type);
>>> }
>>> results in:
>>> 	/tmp, ftype: 0xde.
>>>
>>> Now 0xde != 27, so the question is, where is this 0xde specified.
>>> And more important is this f_type constant over all FreeBSD ZFS filesystems?
>>
>> You got me. And a quick look at sys/kern/vfs_syscalls.c doesn?t help except to imply that the type is set when the filesystem is mounted. I have no idea where 0xde comes from.
> 
> Could that 0xde be the start of 0xdeadcode?
> 
> 0xde is 222 decimal, that does not ring a bell for me either.

Well the VSTAT node gets setup during vfs_mount_alloc() which copies 
data from:
/*
  * Filesystem configuration information. One of these exists for each
  * type of filesystem supported by the kernel. These are searched at
  * mount time to identify the requested filesystem.
  *
  * XXX: Never change the first two arguments!
  */
struct vfsconf {
   u_int   vfc_version;            /* ABI version number */
   char    vfc_name[MFSNAMELEN];   /* filesystem type name */
   struct  vfsops *vfc_vfsops;     /* filesystem operations vector */
   int     vfc_typenum;            /* historic filesystem type number */
   int     vfc_refcount;           /* number mounted of this type */
   int     vfc_flags;              /* permanent flags */
   struct  vfsoptdecl *vfc_opts;   /* mount options */
   TAILQ_ENTRY(vfsconf) vfc_list;  /* list of vfscons */
};

And the "historic filesystem type number" is a worrying comment.

As it turns out the type is calculated/hashed in vfs_init:
====
     if (vfs_typenumhash != 0) {
        /*
         * Calculate a hash on vfc_name to use for vfc_typenum. Unless
         * all of 1<->255 are assigned, it is limited to 8bits since
         * that is what ZFS uses from vfc_typenum and is also the
         * preferred range for vfs_getnewfsid().
         */
        hashval = fnv_32_str(vfc->vfc_name, FNV1_32_INIT);
        hashval &= 0xff;
        secondpass = 0;
        do {
             /* Look for and fix any collision. */
             TAILQ_FOREACH(tvfc, &vfsconf, vfc_list) {
                 if (hashval == tvfc->vfc_typenum) {
                     if (hashval == 255 && secondpass == 0) {
                          hashval = 1;
                          secondpass = 1;
                      } else
                          hashval++;
                      break;
                  }
             }
        } while (tvfc != NULL);
        vfc->vfc_typenum = hashval;
        if (vfc->vfc_typenum >= maxvfsconf)
            maxvfsconf = vfc->vfc_typenum + 1;

So the f_type is sort of stable over time, unless een new type creates a 
collision and things start moving.

Made a small test program:
/tmp, ftype: 0xde, name: zfs.
/home/wjw, ftype: 0x3a, name: nfs.
/home/wjw/.tcshrc, ftype: 0x3a, name: nfs.
/dev/random, ftype: 0x71, name: devfs.

So I guess I'm going to '#define FS_ZFS_TYPE 0xde'
--WjW