On 9/16/13 5:55 PM, Dave Chinner wrote:
> On Mon, Sep 16, 2013 at 04:55:28PM -0500, Eric Sandeen wrote:
>> This patch adds a "-D" switch to fsstress so that every time
>> we call readdir, we stat the dentry & compare it's st_mode
>> to the d_type.
>>
>> If -D is specified only once, it ignores DT_UNKNOWN. If specified
>> twice, it considers DT_UNKNOWN to be an error.
>
> Hmmmm. DT_UNKNOWN is actually a valid type on disk right through to the
> userspace interface. I can't think of why we'd want to consider it
> invalid, especially as right now xfs_repair siply zeros the field
> when recreating directory entries...
>
> Cheers,
>
> Dave.
no fair signing off w/ more text below ;)
>> +void test_d_type(int opno, pathname_t *f, struct dirent64 *de)
>> +{
>> + struct stat64 sb;
>> + char path[PATH_MAX];
>> +
>> + snprintf(path, PATH_MAX, "%s/%s", f->path, de->d_name);
>> +
>> + /* Don't check ./. or ./.. */
>> + if (!strncmp(path, "./.", 3))
>> + return;
>
> . and .. should have the values of DT_UNKNOWN or DT_DIR. They are
> the only valid values for these entries.
Hm let me look at something, I saw something that prompted this
but now that I think about it maybe it's a bug.
>> +
>> + if (lstat64(path, &sb)) {
>> + printf("%d/%d: getdents - can't stat %s\n",
>> + procid, opno, path);
>> + } else {
>> + int bad_d_type = 0;
>> +
>> + switch (de->d_type) {
>> + case DT_BLK:
>> + if (!S_ISBLK(sb.st_mode))
>> + bad_d_type++;
>> + break;
>> + case DT_CHR:
>> + if (!S_ISCHR(sb.st_mode))
>> + bad_d_type++;
>> + break;
>> + case DT_DIR:
>> + if (!S_ISDIR(sb.st_mode))
>> + bad_d_type++;
>> + break;
>> + case DT_FIFO:
>> + if (!S_ISFIFO(sb.st_mode))
>> + bad_d_type++;
>> + break;
>> + case DT_LNK:
>> + if (!S_ISLNK(sb.st_mode))
>> + bad_d_type++;
>> + break;
>> + case DT_REG:
>> + if (!S_ISREG(sb.st_mode))
>> + bad_d_type++;
>> + break;
>> + case DT_SOCK:
>> + if (!S_ISSOCK(sb.st_mode))
>> + bad_d_type++;
>> + break;
>> + case DT_UNKNOWN:
>> + if (verify_d_type > 1)
>> + bad_d_type++;
>> + break;
>> + }
>
> And DT_WHT? That's defined on disk and in the user interface ;)
but fsstress won't create it, will it?
> i.e. this will not do the right thing with an unknown de->d_type.
... but we know what fsstress can possibly create, right, so testing
those created types should be safe, I'd think.
I'll give this some thought & send V2.
-Eric
> Cheers,
>
> Dave.
>
|