[Ocfs2-devel] Failed to upgrade ocfs2-tools to v1.8.4

[piaojun]

Hi Gang and all,

	Sorry for the bug in the patch of debugfs.ocfs2, I have fix it in patch V2.
	
	Please help review my new patch.	

	Thanks
	Piao

在 2015/4/28 17:44, Gang He 写道:
> Hi Goldwyn and all,
> 
> I think that I have got the root cause, which let the test case run_backup_super() failed.
> the test case failure was caused by execute the command "debugfs.ocfs2 /dev/mapper/cluster--vg1-big--lv -s 1"
> 
> + /usr/bin/sudo -u root /usr/sbin/debugfs.ocfs2 /dev/mapper/cluster--vg1-big--lv -s 1
> open: Device name specified was not found Can't get the blocksize from the device by the num 1
> 
> Now, let us look at the related backtrace under ocfs2-tools v1.8.3 (this version is good)
> (gdb) bt
> #0  get_blocksize (dev=0x65feb0 "/dev/mapper/cluster--vg1-big--lv", offset=1073741824, blocksize=0x7fffffffdf88, super_no=1) at commands.c:476  <<== pls note dev argument is OK
> #1  0x0000000000404ea9 in process_open_args (args=0x65fe80, superblock=0x7fffffffe028, blocksize=0x7fffffffe020) at commands.c:567
> #2  0x0000000000405230 in do_open (args=0x65fe80) at commands.c:679
> #3  0x000000000040476f in do_command (cmd=0x65f0f0 "open /dev/mapper/cluster--vg1-big--lv -s 1") at commands.c:346
> #4  0x00000000004043dc in main (argc=4, argv=0x7fffffffe218) at main.c:491
> 
> look at the related backtrace under ocfs2-tools v1.8.4
> (gdb) bt
> #0  get_blocksize (dev=0x65feb0 "-s", offset=1073741824, blocksize=0x7fffffffdff8, super_no=1) at commands.c:476  <<== pls note dev argument is not a device name, this is why the command will fail
> #1  0x0000000000404f0d in process_open_args (args=0x65fe80, superblock=0x7fffffffe098, blocksize=0x7fffffffe090) at commands.c:567
> #2  0x00000000004052a0 in do_open (args=0x65fe80) at commands.c:680
> #3  0x00000000004047d3 in do_command (cmd=0x65f0f0 "open /dev/mapper/cluster--vg1-big--lv -s 1") at commands.c:346
> #4  0x0000000000404440 in main (argc=4, argv=0x7fffffffe288) at main.c:492
> 
> then, look at the latest code changes from git log
> commit 9233fb7eca586de1cad82488ef4a60dbf245f034
> Author: Srinivas Eeda <srinivas.eeda at oracle.com>
> Date:   Fri Jan 30 12:51:45 2015 -0800
>     tools: Up version to 1.8.4
>     Signed-off-by: Srinivas Eeda <srinivas.eeda at oracle.com>
> commit 9693851641bfcd0f2bab226e9f03d9ab05cb7edf
> Author: piaojun <piaojun at huawei.com>
> Date:   Sun Feb 15 08:51:45 2015 +0800
>     debugfs.ocfs2: Fix memory leak problem in process_open_args() & main()  <<== here, the related code is modified.
> 
> 526 static int process_open_args(char **args,
>  527                  uint64_t *superblock, uint64_t *blocksize)
>  528 {
>  529     errcode_t ret = 0;
>  530     uint32_t s = 0;
>  531     char *ptr, *dev;
>  532     uint64_t byte_off[OCFS2_MAX_BACKUP_SUPERBLOCKS];
>  533     uint64_t blksize = 0;
>  534     int num, argc, c;
>  535
>  536     for (argc = 0; (args[argc]); ++argc);
>  537     optind = 0;
>  538     while ((c = getopt(argc, args, "is:")) != EOF) {
>  539         switch (c) {
>  540             case 'i':
>  541                 gbls.imagefile = 1;
>  542                 break;
>  543             case 's':
>  544                 s = strtoul(optarg, &ptr, 0);
>  545                 break;
>  546             default:
>  547                 return 1;
>  548                 break;
>  549         }
>  550     }
>  551
>  552     if (!s)
>  553         return 0;
>  554
>  555     num = ocfs2_get_backup_super_offsets(NULL, byte_off,
>  556                          ARRAY_SIZE(byte_off));
>  557     if (!num)
>  558         return -1;
>  559
>  560     if (s < 1 || s > num) {
>  561         fprintf(stderr, "Backup super block is outside of valid range"
>  562             "(between 1 and %d)\n", num);
>  563         return -1;
>  564     }
>  565
>  566     dev = strdup(args[1]);     <<== pls note this line is moved here from line 537, but actually the args[] is changed by code (getopt()?), then the fix bring this issue, pls move this line back to line 537 and consider the memory leak problem again.     
>  567     ret = get_blocksize(dev, byte_off[s-1], &blksize, s);
>  568     if (ret) {
>  569         com_err(args[0],ret, "Can't get the blocksize from the device"
>  570             " by the num %u\n", s);
>  571         goto bail;
>  572     }
>  573
>  574     *blocksize = blksize;
>  575     *superblock = byte_off[s-1]/blksize;
> 
> 
> Thanks
> Gang
> 
> 
>>>>
>> Hi Goldwyn,
>>
>> I did a upgrade ocfs2-tools to v1.8.4, the code change is here,
>> https://api.opensuse.org/package/rdiff/home:ganghe:branches:network:ha-clust 
>> ering:Factory/ocfs2-tools?opackage=ocfs2-tools&oproject=network%3Aha-clusteri
>> ng%3AFactory&rev=3
>>
>> But, the new build cannot pass the run_backup_super() test case, the 
>> previous build v1.8.3 is OK.
>> if you have time, please take a glance, I also continue to look at why this 
>> case failed.
>> The run_back_super() test failure log is attached.
>> the code diff between v.1.8.3 and v1.8.4 is attached.
>> the code diff between upstream and v1.8.4 is attached.
>> tools/test packages are attached
>>
>>
>>
>>  
>>
>>
>> Thanks
>> Gang
>>
>>