From: Hou Tao <houtao1@huawei.com>
To: Yufen Yu <yuyufen@huawei.com>, <axboe@kernel.dk>
Cc: <linux-block@vger.kernel.org>, <ming.lei@redhat.com>,
<hch@lst.de>, <zhengchuan@huawei.com>, <yi.zhang@huawei.com>,
<paulmck@kernel.org>, <joel@joelfernandes.org>,
<rcu@vger.kernel.org>
Subject: Re: [PATCH] block: make sure last_lookup set as NULL after part deleted
Date: Tue, 31 Dec 2019 22:55:47 +0800 [thread overview]
Message-ID: <a9ce86d6-dadb-9301-7d76-8cef81d782fd@huawei.com> (raw)
In-Reply-To: <20191231110945.10857-1-yuyufen@huawei.com>
Hi,
On 2019/12/31 19:09, Yufen Yu wrote:
> When delete partition executes concurrently with IOs issue,
> it may cause use-after-free on part in disk_map_sector_rcu()
> as following:
snip
>
> diff --git a/block/genhd.c b/block/genhd.c
> index ff6268970ddc..39fa8999905f 100644
> --- a/block/genhd.c
> +++ b/block/genhd.c
> @@ -293,7 +293,23 @@ struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
> part = rcu_dereference(ptbl->part[i]);
>
> if (part && sector_in_part(part, sector)) {
snip
> rcu_assign_pointer(ptbl->last_lookup, part);
> + part = rcu_dereference(ptbl->part[i]);
> + if (part == NULL) {
> + rcu_assign_pointer(ptbl->last_lookup, NULL);
> + break;
> + }
> return part;
> }
> }
Not ensure whether the re-read can handle the following case or not:
process A process B process C
disk_map_sector_rcu(): delete_partition(): disk_map_sector_rcu():
rcu_read_lock
// need to iterate partition table
part[i] != NULL (1) part[i] = NULL (2)
smp_mb()
last_lookup = NULL (3)
call_rcu() (4)
last_lookup = part[i] (5)
rcu_read_lock()
read last_lookup return part[i] (6)
sector_in_part() is OK (7)
return part[i] (8)
part[i] == NULL (9)
last_lookup = NULL (10)
rcu_read_unlock() (11)
one RCU grace period completes
__delete_partition() (12)
free hd_partition (13)
// use-after-free
hd_struct_try_get(part[i]) (14)
* the number in the parenthesis is the sequence of events.
Maybe RCU experts can shed some light on this problem, so cc +paulmck@kernel.org, +joel@joelfernandes.org and +RCU maillist.
If the above case is possible, maybe we can fix the problem by pinning last_lookup through increasing its ref-count
(the following patch is only compile tested):
diff --git a/block/genhd.c b/block/genhd.c
index 6e8543ca6912..179e0056fae1 100644
--- a/block/genhd.c
+++ b/block/genhd.c
@@ -279,7 +279,14 @@ struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
part = rcu_dereference(ptbl->part[i]);
if (part && sector_in_part(part, sector)) {
- rcu_assign_pointer(ptbl->last_lookup, part);
+ struct hd_struct *old;
+
+ if (!hd_struct_try_get(part))
+ break;
+
+ old = xchg(&ptbl->last_lookup, part);
+ if (old)
+ hd_struct_put(old);
return part;
}
}
@@ -1231,7 +1238,11 @@ static void disk_replace_part_tbl(struct gendisk *disk,
rcu_assign_pointer(disk->part_tbl, new_ptbl);
if (old_ptbl) {
- rcu_assign_pointer(old_ptbl->last_lookup, NULL);
+ struct hd_struct *part;
+
+ part = xchg(&old_ptbl->last_lookup, NULL);
+ if (part)
+ hd_struct_put(part);
kfree_rcu(old_ptbl, rcu_head);
}
}
diff --git a/block/partition-generic.c b/block/partition-generic.c
index 98d60a59b843..441c1c591c04 100644
--- a/block/partition-generic.c
+++ b/block/partition-generic.c
@@ -285,7 +285,8 @@ void delete_partition(struct gendisk *disk, int partno)
return;
rcu_assign_pointer(ptbl->part[partno], NULL);
- rcu_assign_pointer(ptbl->last_lookup, NULL);
+ if (cmpxchg(&ptbl->last_lookup, part, NULL) == part)
+ hd_struct_put(part);
kobject_put(part->holder_dir);
device_del(part_to_dev(part));
--
2.22.0
Regards,
Tao
> diff --git a/block/partition-generic.c b/block/partition-generic.c
> index 1d20c9cf213f..1e0065ed6f02 100644
> --- a/block/partition-generic.c
> +++ b/block/partition-generic.c
> @@ -284,6 +284,13 @@ void delete_partition(struct gendisk *disk, int partno)
> return;
>
> rcu_assign_pointer(ptbl->part[partno], NULL);
> + /*
> + * Without the memory barrier, disk_map_sector_rcu()
> + * may read the old value after overwriting the
> + * last_lookup. Then it can not clear last_lookup,
> + * which may cause use-after-free.
> + */
> + smp_mb();
> rcu_assign_pointer(ptbl->last_lookup, NULL);
> kobject_put(part->holder_dir);
> device_del(part_to_dev(part));
>
next parent reply other threads:[~2019-12-31 14:56 UTC|newest]
Thread overview: 20+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <20191231110945.10857-1-yuyufen@huawei.com>
2019-12-31 14:55 ` Hou Tao [this message]
2019-12-31 23:11 ` [PATCH] block: make sure last_lookup set as NULL after part deleted Paul E. McKenney
2020-01-01 2:33 ` htbegin
2020-01-01 3:39 ` htbegin
2020-01-03 23:45 ` Joel Fernandes
2020-01-04 9:16 ` Hou Tao
2020-01-02 1:23 ` Ming Lei
2020-01-03 3:06 ` Hou Tao
2020-01-03 4:18 ` Ming Lei
2020-01-03 7:35 ` Hou Tao
2020-01-03 8:17 ` Ming Lei
2020-01-03 12:03 ` Yufen Yu
2020-01-03 15:16 ` Ming Lei
2020-01-06 7:39 ` Yufen Yu
2020-01-06 8:11 ` Ming Lei
2020-01-06 9:41 ` Hou Tao
2020-01-06 10:05 ` Ming Lei
2020-01-07 11:40 ` Hou Tao
2020-01-08 3:19 ` Ming Lei
2020-01-03 12:43 ` Yufen Yu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a9ce86d6-dadb-9301-7d76-8cef81d782fd@huawei.com \
--to=houtao1@huawei.com \
--cc=axboe@kernel.dk \
--cc=hch@lst.de \
--cc=joel@joelfernandes.org \
--cc=linux-block@vger.kernel.org \
--cc=ming.lei@redhat.com \
--cc=paulmck@kernel.org \
--cc=rcu@vger.kernel.org \
--cc=yi.zhang@huawei.com \
--cc=yuyufen@huawei.com \
--cc=zhengchuan@huawei.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).