* [PATCH] Take 2: cdev_unmap()
@ 2004-02-05 18:01 Jonathan Corbet
2004-02-05 18:15 ` Greg KH
0 siblings, 1 reply; 2+ messages in thread
From: Jonathan Corbet @ 2004-02-05 18:01 UTC (permalink / raw)
To: linux-kernel; +Cc: akpm, viro
I didn't get any responses to my last posting on cdev_unmap(); I take it
that means nobody objects :)
To recap my argument: the current cdev implementation keeps an uncounted
reference to every cdev in cdev_map. Creators of cdevs must know to call
cdev_unmap() with the same arguments they passed to cdev_add() before
releasing the device, or that reference will remain and will oops the
kernel should user space attempt to open the (missing) device. It's an
easy mistake to make, and, IMO, entirely unnecessary; the cdev code should
be able to do its own bookkeeping.
So, does anybody have a reason why this shouldn't go in? Al, have I missed
something?
[This is against 2.6.2-mm1].
jon
diff -urNp -X dontdiff 2.6.2-mm1-v/drivers/char/tty_io.c 2.6.2-mm1/drivers/char/tty_io.c
--- 2.6.2-mm1-v/drivers/char/tty_io.c 2004-02-06 00:53:14.000000000 -0700
+++ 2.6.2-mm1/drivers/char/tty_io.c 2004-02-06 01:18:38.000000000 -0700
@@ -2264,7 +2264,6 @@ int tty_unregister_driver(struct tty_dri
if (driver->refcount)
return -EBUSY;
- cdev_unmap(MKDEV(driver->major, driver->minor_start), driver->num);
unregister_chrdev_region(MKDEV(driver->major, driver->minor_start),
driver->num);
diff -urNp -X dontdiff 2.6.2-mm1-v/drivers/ieee1394/amdtp.c 2.6.2-mm1/drivers/ieee1394/amdtp.c
--- 2.6.2-mm1-v/drivers/ieee1394/amdtp.c 2004-02-06 00:51:44.000000000 -0700
+++ 2.6.2-mm1/drivers/ieee1394/amdtp.c 2004-02-06 01:18:38.000000000 -0700
@@ -1308,7 +1308,6 @@ static void __exit amdtp_exit_module (vo
hpsb_unregister_highlevel(&amdtp_highlevel);
devfs_remove("amdtp");
- cdev_unmap(IEEE1394_AMDTP_DEV, 16);
cdev_del(&amdtp_cdev);
HPSB_INFO("Unloaded AMDTP driver");
diff -urNp -X dontdiff 2.6.2-mm1-v/drivers/ieee1394/dv1394.c 2.6.2-mm1/drivers/ieee1394/dv1394.c
--- 2.6.2-mm1-v/drivers/ieee1394/dv1394.c 2004-02-06 00:51:44.000000000 -0700
+++ 2.6.2-mm1/drivers/ieee1394/dv1394.c 2004-02-06 01:18:38.000000000 -0700
@@ -2609,7 +2609,6 @@ static void __exit dv1394_exit_module(vo
hpsb_unregister_protocol(&dv1394_driver);
hpsb_unregister_highlevel(&dv1394_highlevel);
- cdev_unmap(IEEE1394_DV1394_DEV, 16);
cdev_del(&dv1394_cdev);
devfs_remove("ieee1394/dv");
}
diff -urNp -X dontdiff 2.6.2-mm1-v/drivers/ieee1394/raw1394.c 2.6.2-mm1/drivers/ieee1394/raw1394.c
--- 2.6.2-mm1-v/drivers/ieee1394/raw1394.c 2004-02-06 00:51:44.000000000 -0700
+++ 2.6.2-mm1/drivers/ieee1394/raw1394.c 2004-02-06 01:18:38.000000000 -0700
@@ -2682,7 +2682,6 @@ static int __init init_raw1394(void)
static void __exit cleanup_raw1394(void)
{
hpsb_unregister_protocol(&raw1394_driver);
- cdev_unmap(IEEE1394_RAW1394_DEV, 1);
cdev_del(&raw1394_cdev);
devfs_remove(RAW1394_DEVICE_NAME);
hpsb_unregister_highlevel(&raw1394_highlevel);
diff -urNp -X dontdiff 2.6.2-mm1-v/drivers/ieee1394/video1394.c 2.6.2-mm1/drivers/ieee1394/video1394.c
--- 2.6.2-mm1-v/drivers/ieee1394/video1394.c 2004-02-06 00:51:44.000000000 -0700
+++ 2.6.2-mm1/drivers/ieee1394/video1394.c 2004-02-06 01:18:38.000000000 -0700
@@ -1447,7 +1447,6 @@ static void __exit video1394_exit_module
hpsb_unregister_highlevel(&video1394_highlevel);
devfs_remove(VIDEO1394_DRIVER_NAME);
- cdev_unmap(IEEE1394_VIDEO1394_DEV, 16);
cdev_del(&video1394_cdev);
PRINT_G(KERN_INFO, "Removed " VIDEO1394_DRIVER_NAME " module");
diff -urNp -X dontdiff 2.6.2-mm1-v/drivers/scsi/sg.c 2.6.2-mm1/drivers/scsi/sg.c
--- 2.6.2-mm1-v/drivers/scsi/sg.c 2004-02-06 00:53:19.000000000 -0700
+++ 2.6.2-mm1/drivers/scsi/sg.c 2004-02-06 01:21:23.000000000 -0700
@@ -1521,7 +1521,6 @@ sg_remove(struct class_device *cl_dev)
if (sdp) {
sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
class_simple_device_remove(MKDEV(SCSI_GENERIC_MAJOR, k));
- cdev_unmap(MKDEV(SCSI_GENERIC_MAJOR, k), 1);
cdev_del(sdp->cdev);
sdp->cdev = NULL;
devfs_remove("%s/generic", scsidp->devfs_name);
diff -urNp -X dontdiff 2.6.2-mm1-v/drivers/scsi/st.c 2.6.2-mm1/drivers/scsi/st.c
--- 2.6.2-mm1-v/drivers/scsi/st.c 2004-02-06 00:53:19.000000000 -0700
+++ 2.6.2-mm1/drivers/scsi/st.c 2004-02-06 01:21:37.000000000 -0700
@@ -3946,8 +3946,6 @@ out_free_tape:
if (cdev == STm->cdevs[j])
cdev = NULL;
sysfs_remove_link(&STm->cdevs[j]->kobj, "device");
- cdev_unmap(MKDEV(SCSI_TAPE_MAJOR,
- TAPE_MINOR(dev_num, mode, j)), 1);
cdev_del(STm->cdevs[j]);
}
}
@@ -3990,8 +3988,6 @@ static int st_remove(struct device *dev)
for (j=0; j < 2; j++) {
sysfs_remove_link(&tpnt->modes[mode].cdevs[j]->kobj,
"device");
- cdev_unmap(MKDEV(SCSI_TAPE_MAJOR,
- TAPE_MINOR(i, mode, j)), 1);
cdev_del(tpnt->modes[mode].cdevs[j]);
tpnt->modes[mode].cdevs[j] = NULL;
}
diff -urNp -X dontdiff 2.6.2-mm1-v/fs/char_dev.c 2.6.2-mm1/fs/char_dev.c
--- 2.6.2-mm1-v/fs/char_dev.c 2004-02-06 00:51:25.000000000 -0700
+++ 2.6.2-mm1/fs/char_dev.c 2004-02-06 01:22:02.000000000 -0700
@@ -240,7 +240,6 @@ void unregister_chrdev_region(dev_t from
int unregister_chrdev(unsigned int major, const char *name)
{
struct char_device_struct *cd;
- cdev_unmap(MKDEV(major, 0), 256);
cd = __unregister_chrdev_region(major, 0, 256);
if (cd && cd->cdev)
cdev_del(cd->cdev);
@@ -347,16 +346,19 @@ int cdev_add(struct cdev *p, dev_t dev,
err = kobj_map(cdev_map, dev, count, NULL, exact_match, exact_lock, p);
if (err)
kobject_del(&p->kobj);
+ p->dev = dev;
+ p->count = count;
return err;
}
-void cdev_unmap(dev_t dev, unsigned count)
+static void cdev_unmap(dev_t dev, unsigned count)
{
kobj_unmap(cdev_map, dev, count);
}
void cdev_del(struct cdev *p)
{
+ cdev_unmap(p->dev, p->count);
kobject_del(&p->kobj);
kobject_put(&p->kobj);
}
@@ -458,6 +460,5 @@ EXPORT_SYMBOL(cdev_get);
EXPORT_SYMBOL(cdev_put);
EXPORT_SYMBOL(cdev_del);
EXPORT_SYMBOL(cdev_add);
-EXPORT_SYMBOL(cdev_unmap);
EXPORT_SYMBOL(register_chrdev);
EXPORT_SYMBOL(unregister_chrdev);
diff -urNp -X dontdiff 2.6.2-mm1-v/include/linux/cdev.h 2.6.2-mm1/include/linux/cdev.h
--- 2.6.2-mm1-v/include/linux/cdev.h 2003-09-08 13:50:09.000000000 -0600
+++ 2.6.2-mm1/include/linux/cdev.h 2004-02-06 01:22:02.000000000 -0700
@@ -7,6 +7,8 @@ struct cdev {
struct module *owner;
struct file_operations *ops;
struct list_head list;
+ dev_t dev;
+ unsigned int count;
};
void cdev_init(struct cdev *, struct file_operations *);
@@ -21,8 +23,6 @@ int cdev_add(struct cdev *, dev_t, unsig
void cdev_del(struct cdev *);
-void cdev_unmap(dev_t, unsigned);
-
void cd_forget(struct inode *);
#endif
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Take 2: cdev_unmap()
2004-02-05 18:01 [PATCH] Take 2: cdev_unmap() Jonathan Corbet
@ 2004-02-05 18:15 ` Greg KH
0 siblings, 0 replies; 2+ messages in thread
From: Greg KH @ 2004-02-05 18:15 UTC (permalink / raw)
To: Jonathan Corbet; +Cc: linux-kernel, akpm, viro
On Thu, Feb 05, 2004 at 11:01:03AM -0700, Jonathan Corbet wrote:
> I didn't get any responses to my last posting on cdev_unmap(); I take it
> that means nobody objects :)
>
> To recap my argument: the current cdev implementation keeps an uncounted
> reference to every cdev in cdev_map. Creators of cdevs must know to call
> cdev_unmap() with the same arguments they passed to cdev_add() before
> releasing the device, or that reference will remain and will oops the
> kernel should user space attempt to open the (missing) device. It's an
> easy mistake to make, and, IMO, entirely unnecessary; the cdev code should
> be able to do its own bookkeeping.
>
> So, does anybody have a reason why this shouldn't go in? Al, have I missed
> something?
I sure like it, it starts to make the cdev interface easier to use.
Al, unless you object, I'll add this to my driver bk tree which will get
sucked into -mm and then send it off to Linus after a bit of testing.
thanks,
greg k-h
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2004-02-05 18:15 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-02-05 18:01 [PATCH] Take 2: cdev_unmap() Jonathan Corbet
2004-02-05 18:15 ` Greg KH
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox