Skip to contentSkip to conversation

s390/tape: iterator after loop end in tape_assign_minor?

PATCH New Maoyi Xie 2 participants 3 messages May 19 – Jun 1 s390

lore
Share & tools
lore

Maoyi Xie

author
#0 lore raw

Hi all,

While reading drivers/s390/char/tape_core.c I noticed something that looks like a past the end iterator pattern. I would appreciate it if you could take a look and let me know whether this is a real issue, and whether it is worth fixing.

The site is tape_assign_minor() (linux-7.1-rc1, around line 339):

    list_for_each_entry(tmp, &tape_device_list, node) {
            if (minor < tmp->first_minor)
                    break;
            minor += TAPE_MINORS_PER_DEV;
    }
    if (minor >= 256) {
            write_unlock(&tape_device_lock);
            return -ENODEV;
    }
    device->first_minor = minor;
    list_add_tail(&device->node, &tmp->node);

When the loop walks all entries without break (every existing entry has first_minor at or below the candidate minor), tmp is past the end. tmp->node then aliases tape_device_list (the list head) via container_of offset cancellation. list_add_tail(&device->node, &tape_device_list) inserts the new device at the tail of the list. That is the intended behaviour for a sorted insertion where the new device has the largest first_minor. The dereference of the past the end iterator is undefined per C11.

Jakob Koschel cleaned up many such sites in 2022, for example commits 99d8ae4ec8a (tracing: Remove usage of list iterator variable after the loop), 2966a9918df (clockevents:
Use dedicated list iterator variable) and dc1acd5c946 (dlm:
replace usage of found with dedicated list iterator variable). This site was not covered.

A candidate fix would track an explicit insertion target. Declare struct list_head *insert_before = &tape_device_list before the loop. Overwrite it to &tmp->node only when the loop breaks early. The final list_add_tail then reads insert_before. On break that points to the entry right after the insertion position. On fall-through it stays at the list head, so list_add_tail appends at the tail. The behaviour is unchanged in all cases, including an empty list and a list with one entry.

If this is intentional or already known, please disregard. Otherwise, I am happy to send a [PATCH] or to leave the fix to you. Thank you for your time, and sorry for the noise if this is not actually worth fixing or has already been spotted.

Thanks,
Maoyi Xie
https://maoyixie.com/

maintainer
#1 lore raw
Quoted 7 lines · Maoyi Xie

Hi all,

While reading drivers/s390/char/tape_core.c I noticed something that looks like a past the end iterator pattern. I would appreciate it if you could take a look and let me know whether this is a real issue, and whether it is worth fixing.

Thanks for reporting. This something that Jan should look into. Full quote below.

Quoted 52 lines

The site is tape_assign_minor() (linux-7.1-rc1, around line 339):

    list_for_each_entry(tmp, &tape_device_list, node) {
            if (minor < tmp->first_minor)
                    break;
            minor += TAPE_MINORS_PER_DEV;
    }
    if (minor >= 256) {
            write_unlock(&tape_device_lock);
            return -ENODEV;
    }
    device->first_minor = minor;
    list_add_tail(&device->node, &tmp->node);

When the loop walks all entries without break (every existing entry has first_minor at or below the candidate minor), tmp is past the end. tmp->node then aliases tape_device_list (the list head) via container_of offset cancellation. list_add_tail(&device->node, &tape_device_list) inserts the new device at the tail of the list. That is the intended behaviour for a sorted insertion where the new device has the largest first_minor. The dereference of the past the end iterator is undefined per C11.

Jakob Koschel cleaned up many such sites in 2022, for example commits 99d8ae4ec8a (tracing: Remove usage of list iterator variable after the loop), 2966a9918df (clockevents:
Use dedicated list iterator variable) and dc1acd5c946 (dlm:
replace usage of found with dedicated list iterator variable). This site was not covered.

A candidate fix would track an explicit insertion target. Declare struct list_head *insert_before = &tape_device_list before the loop. Overwrite it to &tmp->node only when the loop breaks early. The final list_add_tail then reads insert_before. On break that points to the entry right after the insertion position. On fall-through it stays at the list head, so list_add_tail appends at the tail. The behaviour is unchanged in all cases, including an empty list and a list with one entry.

If this is intentional or already known, please disregard. Otherwise, I am happy to send a [PATCH] or to leave the fix to you. Thank you for your time, and sorry for the noise if this is not actually worth fixing or has already been spotted.

Thanks,
Maoyi Xie
https://maoyixie.com/

author
#2 lore raw

PATCH 1/1 s390/tape: avoid past-the-end iterator in tape_assign_minor()

In reply to Maoyi Xie

tape_assign_minor() walks tape_device_list to find the sorted insertion point, then does list_add_tail(&device->node, &tmp->node). When the loop runs to the end without break, tmp is past the end and &tmp->node aliases the list head via container_of. list_add_tail then appends at the tail, which is the intended result, but the iterator is dereferenced past the end, which is undefined per the C standard.

Track the insertion point explicitly. insert_before starts at the list head and is set to &tmp->node only when the loop breaks early. The list_add_tail uses insert_before, so the behaviour is unchanged in every case including an empty list.

---
 drivers/s390/char/tape_core.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/s390/char/tape_core.c +5 -2
index bd8e3deb1199..361184c05940 100644
@@ -330,14 +330,17 @@ __tape_cancel_io(struct tape_device *device, struct tape_request *request)
 static int
 tape_assign_minor(struct tape_device *device)
 {
+	struct list_head *insert_before = &tape_device_list;
 	struct tape_device *tmp;
 	int minor;
 
 	minor = 0;
 	write_lock(&tape_device_lock);
 	list_for_each_entry(tmp, &tape_device_list, node) {
-		if (minor < tmp->first_minor)
+		if (minor < tmp->first_minor) {
+			insert_before = &tmp->node;
 			break;
+		}
 		minor += TAPE_MINORS_PER_DEV;
 	}
 	if (minor >= 256) {
@@ -345,7 +348,7 @@ tape_assign_minor(struct tape_device *device)
 		return -ENODEV;
 	}
 	device->first_minor = minor;
-	list_add_tail(&device->node, &tmp->node);
+	list_add_tail(&device->node, insert_before);
 	write_unlock(&tape_device_lock);
 	return 0;
 }
Signature · 1 line
2.34.1