Re: VOP_CREATE hanging

[Antti Kantee <pooka%cs.hut.fi@localhost>]

On Wed Aug 13 2008 at 04:16:03 -0700, Adam Burkepile wrote:
> >It is not legal to call VOP_CREATE() without calling namei (ultimately
> >VOP_LOOKUP()), since this call sets up a lot of state in the file  
> >system
> >to prepare for the actual creation operation.
> 
> Ok, I have made the changes you mentioned (I think), but I'm still  
> having the hanging problem.
> 
>                 printf("subfile not found, needs to be created\n");
>                 NDINIT(&sfnd, CREATE, FOLLOW | TRYEMULROOT,  
> UIO_USERSPACE, SCARG(uap, sub_path));
> 
>                 /* set namedata with namei */
>                 error = namei(&sfnd);
>                 printf("namei error = %d\n",error);
> 
>                 sfnd.ni_cnd.cn_nameiop = CREATE;
>                 sfnd.ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF;

These locking flags affect only namei(), so setting them afterwards
makes no difference.

>                 if ((SCARG(uap, flags) & O_EXCL) == 0 &&
>                                 ((SCARG(uap, flags) & O_NOFOLLOW) ==  
> 0))
>                         sfnd.ni_cnd.cn_flags |= FOLLOW;
>
>                 //lock subfile directory
>                 vn_lock(sdvp, LK_EXCLUSIVE | LK_RETRY);
> 
>                 /* set the attr of the new file */
>                 VATTR_NULL(&sfva);
>                 sfva.va_type = VREG;
>                 sfva.va_rdev = VNOVAL;
>                 sfva.va_mode = (bva.va_mode & ALLPERMS) &~ p->p_cwdi- 
> >cwdi_cmask;
>                 if (SCARG(uap, flags) & O_EXCL)
>                         sfva.va_vaflags |= VA_EXCLUSIVE;
> 
>                 error = VOP_LOOKUP(sdvp,&sfvp,&(sfnd.ni_cnd));
>                 printf("VOP_LOOKUP error = %d\n",error);
> 
>                 error = VOP_CREATE(sdvp,&sfvp,&(sfnd.ni_cnd),&sfva);
>                 printf("VOP_CREATE error = %d\n",error);
> 
>                 if (error == 0)
>                         vput(sfvp);
>                 goto out;
> 
> 
> If it helps, from what I've been able to tell this is the functions it  
> goes through before it hangs:
> sys_subfile_open()
> VOP_CREATE()
> ufs_makeinode()
> ufs_direnter()
> UFS_BALLOC()

Why don't you just give the exact place where it is hanging?  Getting a
backtrace or even just the program counter is quite trivial (but varies
in detail depending on your setup ;).

> I had a few other questions:
> 1. Is VERIEXEC_PATH_GET / VERIEXEC_PATH_PUT necessary for VOP_CREATE?
> 2. When you said "calling ... ultimately VOP_LOOKUP()", did you mean  
> me calling it (like I am trying to do).

No, I meant namei() (or actually lookup() which is called by namei())
is supposed to call VOP_LOOKUP.  What you are doing is ... unorthodox.
I have no idea what horrors will happen if you try to do that.  But I
suspect namei() will already have exhausted the entire pathname leaving
your VOP_LOOKUP as a NOP.  If you're really interested, single-step
through the code, no need for guessing.

> 3.
> >Or, if you're working just with ffs, you could hack
> >lookup into pieces so you can call the parts involved with the  
> >necessary
> >setup for create.
> 
> If VOP_LOOKUP is returning but ultimately messing up, could this be  
> causing the problems that I'm having with VOP_CREATE? I ask because I  
> am having a problem with VOP_LOOKUP not releasing some sort of lock on  
> the fs. It is not letting me umount my test fs and manifest a some  
> other weird errors (not being able to find files). The deadline is  
> coming up so I was inclined to skip the problem and come back to it  
> since it didn't seem like it was hanging the rest of the program and I  
> want to have some functionality to show.

I have no idea without seeing all of your code.  If the regular system
calls still work and you can unmount, the problem is most likely your
subfile syscalls.  I suggest to adapt sys_mknod() for the VOP_CREATE()
case and see if that gets things rolling.