rcu.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
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));
> 


       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).