On Saturday 07 October 2006 01:04, Nic James Ferrier wrote:
> Blaisorblade <[EMAIL PROTECTED]> writes:
> >> Is there a reason that can't be done?
> >
> > No, it can be done. Produce a working patch and then that will be ok, and
> > I guess it will be appreciated (at least by me). It isn't IMHO that much
> > trivial however - the code will be probably be similar but not identical
> > to the userspace equivalent (using getdents()).
>
> Here's my current attempt. It doesn't work reliably yet and I'm not
> sure why:
That is not a bad attempt, and you show having some uncommon experience with
kernel code. I'm not reviewing it line-by-line - I should check the APIs of
VFS calls and usage examples, but I haven't the time.
I can only suggest reading the code of sys_getdents as a starting point. I
think you could use without problems vfs_readdir instead of ->readdir since
it accepts a filldir_t callback and does the needed locking you do not do.
You very likely race with process creation and exit - as /proc/<pid> folders
appear and disappear you need some kind of locking.
In general, as you can see, we tried writing mconsole_proc() with direct
access to VFS, but it is commented out and we now use sys_open() and friends,
because that one was not reliable; you could try that for a first version do
that too (save this version for future use), even it if looks as _this_
operation is simpler to reimplement (if you call vfs_readdir it is).
To show you what was happening there, after some time, Al Viro suggested that
we were using them wrongly, for refcount issues:
nd.mnt = NULL;
should be replaced by something as nd.mnt=mntget(proc_mountpoint) - in fact
there is a corresponding mntput() done by the function we called, and that
mntput() at the end freed the proc mountpoint (there was a similar issue with
nd.dentry IIRC). The result was that after about 5 calls to mconsole proc
(i.e. after removing all references to the proc mountpoint and freeing it),
UML crashed.
PS: this is a recollection of ideas about the crash, but the fix was never
completed so consider it as a reasonable approximation of actual problems.
> @@ -209,19 +209,49 @@
> }
> #endif
>
> +int mconsole_proc_filldir(void * __buf, const char * name, int namlen,
> loff_t offset, + ino_t ino, unsigned int d_type)
> +{
> + struct mc_request *req = (struct mc_request *) __buf;
> + char buf[namlen+1];
> + memcpy(buf, name, namlen);
> + buf[namlen] = '\n';
> + mconsole_reply_len(req, buf, namlen +1, 0, 1);
> + return 0;
> +}
> +
> void mconsole_proc(struct mc_request *req)
> {
> char path[64];
> - char *buf;
> + char *buf = NULL;
> int len;
> int fd;
> int first_chunk = 1;
> + struct file *file;
> char *ptr = req->request.data;
>
> ptr += strlen("proc");
> while(isspace(*ptr)) ptr++;
> snprintf(path, sizeof(path), "/proc/%s", ptr);
>
> + printk("path: %s\n", path);
> +
> + file = filp_open(path, 0, 0);
> +
> + if (*file->f_op->readdir != NULL) {
> + int dir_result;
> + mconsole_reply(req, "\n", 0, 1);
> +
> + do {
> + dir_result = (*file->f_op->readdir)(file, req,
> mconsole_proc_filldir); + }
> + while(dir_result > 0);
> + printk("dir_result: %d\n", dir_result);
> +
> + mconsole_reply(req, "", 0, 0);
> + goto out;
> + }
> +
> fd = sys_open(path, 0, 0);
> if (fd < 0) {
> mconsole_reply(req, "Failed to open file", 1, 0);
> @@ -257,11 +287,12 @@
> }
>
> out_free:
> - kfree(buf);
> + kfree(buf);
> out_close:
> sys_close(fd);
> out:
> - /* nothing */;
> + ;
> + //filp_close(file, NULL);
> }
>
> I have yet to establish wny it doesn't work reliably.
--
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade
http://www.user-mode-linux.org/~blaisorblade
Chiacchiera con i tuoi amici in tempo reale!
http://it.yahoo.com/mail_it/foot/*http://it.messenger.yahoo.com
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
User-mode-linux-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
