Re: vn_fullpath: 0xc85e24a0 is not locked but should be

From: Don Lewis <truckman_at_FreeBSD.org>
Date: Fri, 12 Dec 2003 10:35:35 -0800 (PST)
On 12 Dec, Jun Kuriyama wrote:
> At Thu, 11 Dec 2003 23:14:50 -0500 (EST),
> Robert Watson wrote:
>> Ah, you're still runing with the VFS lock debugging :-).  Indeed, it looks
>> like a vn_lock() and unlock of p->p_textvp is missing in
>> procfs_doprocfile(), even though that likely would violate the VFS lock
>> order.  The attached (untested) patch might well fix it, but might not be
>> right -- I'm not sure that curthread holds a valid reference to
>> p->p_textvp that can't evaporate during these operations.  I'm not sure
>> the proc reference stuff protects us properly here, but John would know
>> (CC'd). 
>> 
>> Index: procfs.c
>> ===================================================================
>> RCS file: /home/ncvs/src/sys/fs/procfs/procfs.c,v
>> retrieving revision 1.9
>> diff -u -r1.9 procfs.c
>> --- procfs.c	17 Apr 2003 22:12:12 -0000	1.9
>> +++ procfs.c	12 Dec 2003 04:13:10 -0000
>> _at__at_ -70,7 +70,9 _at__at_
>>  	char *fullpath = "unknown";
>>  	char *freepath = NULL;
>>  
>> +	vn_lock(p->p_textvp, LK_EXCLUSIVE | LK_RETRY, td);
>>  	vn_fullpath(td, p->p_textvp, &fullpath, &freepath);
>> +	VOP_UNLOCK(p->p_textvp, 0, td);
>>  	sbuf_printf(sb, "%s", fullpath);
>>  	if (freepath)
>>  		free(freepath, M_TEMP);
> 
> Okay, I'll wait without DEBUG_VFS_LOCKS until fix is committed.

DEBUG_VFS_LOCKS is quite usable as long as you don't run find, tar, etc.
that traverses procfs.
Received on Fri Dec 12 2003 - 09:35:48 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:33 UTC