public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
From: Yufan Chen <yufan.chen@linux.dev>
To: Joseph Qi <joseph.qi@linux.alibaba.com>, ocfs2-devel@lists.linux.dev
Cc: Mark Fasheh <mark@fasheh.com>, Joel Becker <jlbec@evilplan.org>,
	linux-kernel@vger.kernel.org
Subject: [PATCH v2] ocfs2/heartbeat: fix slot mapping rollback leaks on error paths
Date: Mon, 30 Mar 2026 23:34:28 +0800	[thread overview]
Message-ID: <20260330153428.19586-1-yufan.chen@linux.dev> (raw)

From: Yufan Chen <ericterminal@gmail.com>

o2hb_map_slot_data() allocates hr_tmp_block, hr_slots,
hr_slot_data, and pages in stages. If a later allocation fails, the
current code returns without unwinding the earlier allocations.

o2hb_region_dev_store() also leaves slot mapping resources behind when
setup aborts, and it keeps hr_aborted_start/hr_node_deleted set across
retries. That leaves stale state behind after a failed start.

Factor the slot cleanup into o2hb_unmap_slot_data(), use it from both
o2hb_map_slot_data() and o2hb_region_release(), and call it from the
dev_store() rollback after stopping a started heartbeat thread. While
freeing pages, clear each hr_slot_data entry as it is released, and
reset the start state before each new setup attempt.

This closes the slot mapping leak on allocation/setup failure paths
and keeps failed setup attempts retryable.

Signed-off-by: Yufan Chen <ericterminal@gmail.com>
---
 fs/ocfs2/cluster/heartbeat.c | 83 ++++++++++++++++++++++++------------
 1 file changed, 56 insertions(+), 27 deletions(-)

diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
index fe1949578..d12784aaa 100644
--- a/fs/ocfs2/cluster/heartbeat.c
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -1488,33 +1488,45 @@ static struct o2hb_region *to_o2hb_region(struct config_item *item)
 	return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
 }
 
-/* drop_item only drops its ref after killing the thread, nothing should
- * be using the region anymore.  this has to clean up any state that
- * attributes might have built up. */
-static void o2hb_region_release(struct config_item *item)
+static void o2hb_unmap_slot_data(struct o2hb_region *reg)
 {
 	int i;
 	struct page *page;
-	struct o2hb_region *reg = to_o2hb_region(item);
-
-	mlog(ML_HEARTBEAT, "hb region release (%pg)\n", reg_bdev(reg));
-
-	kfree(reg->hr_tmp_block);
 
 	if (reg->hr_slot_data) {
 		for (i = 0; i < reg->hr_num_pages; i++) {
 			page = reg->hr_slot_data[i];
-			if (page)
+			if (page) {
 				__free_page(page);
+				reg->hr_slot_data[i] = NULL;
+			}
 		}
 		kfree(reg->hr_slot_data);
+		reg->hr_slot_data = NULL;
 	}
 
+	kfree(reg->hr_slots);
+	reg->hr_slots = NULL;
+
+	kfree(reg->hr_tmp_block);
+	reg->hr_tmp_block = NULL;
+}
+
+/* drop_item only drops its ref after killing the thread, nothing should
+ * be using the region anymore.  this has to clean up any state that
+ * attributes might have built up.
+ */
+static void o2hb_region_release(struct config_item *item)
+{
+	struct o2hb_region *reg = to_o2hb_region(item);
+
+	mlog(ML_HEARTBEAT, "hb region release (%pg)\n", reg_bdev(reg));
+
+	o2hb_unmap_slot_data(reg);
+
 	if (reg->hr_bdev_file)
 		fput(reg->hr_bdev_file);
 
-	kfree(reg->hr_slots);
-
 	debugfs_remove_recursive(reg->hr_debug_dir);
 	kfree(reg->hr_db_livenodes);
 	kfree(reg->hr_db_regnum);
@@ -1667,6 +1679,7 @@ static void o2hb_init_region_params(struct o2hb_region *reg)
 static int o2hb_map_slot_data(struct o2hb_region *reg)
 {
 	int i, j;
+	int ret = -ENOMEM;
 	unsigned int last_slot;
 	unsigned int spp = reg->hr_slots_per_page;
 	struct page *page;
@@ -1674,14 +1687,14 @@ static int o2hb_map_slot_data(struct o2hb_region *reg)
 	struct o2hb_disk_slot *slot;
 
 	reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
-	if (reg->hr_tmp_block == NULL)
-		return -ENOMEM;
+	if (!reg->hr_tmp_block)
+		goto out;
 
 	reg->hr_slots = kzalloc_objs(struct o2hb_disk_slot, reg->hr_blocks);
-	if (reg->hr_slots == NULL)
-		return -ENOMEM;
+	if (!reg->hr_slots)
+		goto out;
 
-	for(i = 0; i < reg->hr_blocks; i++) {
+	for (i = 0; i < reg->hr_blocks; i++) {
 		slot = &reg->hr_slots[i];
 		slot->ds_node_num = i;
 		INIT_LIST_HEAD(&slot->ds_live_item);
@@ -1695,12 +1708,12 @@ static int o2hb_map_slot_data(struct o2hb_region *reg)
 
 	reg->hr_slot_data = kzalloc_objs(struct page *, reg->hr_num_pages);
 	if (!reg->hr_slot_data)
-		return -ENOMEM;
+		goto out;
 
-	for(i = 0; i < reg->hr_num_pages; i++) {
+	for (i = 0; i < reg->hr_num_pages; i++) {
 		page = alloc_page(GFP_KERNEL);
 		if (!page)
-			return -ENOMEM;
+			goto out;
 
 		reg->hr_slot_data[i] = page;
 
@@ -1720,6 +1733,10 @@ static int o2hb_map_slot_data(struct o2hb_region *reg)
 	}
 
 	return 0;
+
+out:
+	o2hb_unmap_slot_data(reg);
+	return ret;
 }
 
 /* Read in all the slots available and populate the tracking
@@ -1809,9 +1826,11 @@ static ssize_t o2hb_region_dev_store(struct config_item *item,
 		     "blocksize %u incorrect for device, expected %d",
 		     reg->hr_block_bytes, sectsize);
 		ret = -EINVAL;
-		goto out3;
+		goto out;
 	}
 
+	reg->hr_aborted_start = 0;
+	reg->hr_node_deleted = 0;
 	o2hb_init_region_params(reg);
 
 	/* Generation of zero is invalid */
@@ -1823,13 +1842,13 @@ static ssize_t o2hb_region_dev_store(struct config_item *item,
 	ret = o2hb_map_slot_data(reg);
 	if (ret) {
 		mlog_errno(ret);
-		goto out3;
+		goto out;
 	}
 
 	ret = o2hb_populate_slot_data(reg);
 	if (ret) {
 		mlog_errno(ret);
-		goto out3;
+		goto out;
 	}
 
 	INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
@@ -1860,7 +1879,7 @@ static ssize_t o2hb_region_dev_store(struct config_item *item,
 	if (IS_ERR(hb_task)) {
 		ret = PTR_ERR(hb_task);
 		mlog_errno(ret);
-		goto out3;
+		goto out;
 	}
 
 	spin_lock(&o2hb_live_lock);
@@ -1877,12 +1896,12 @@ static ssize_t o2hb_region_dev_store(struct config_item *item,
 
 	if (reg->hr_aborted_start) {
 		ret = -EIO;
-		goto out3;
+		goto out;
 	}
 
 	if (reg->hr_node_deleted) {
 		ret = -EINVAL;
-		goto out3;
+		goto out;
 	}
 
 	/* Ok, we were woken.  Make sure it wasn't by drop_item() */
@@ -1901,8 +1920,18 @@ static ssize_t o2hb_region_dev_store(struct config_item *item,
 		printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%pg)\n",
 		       config_item_name(&reg->hr_item), reg_bdev(reg));
 
-out3:
+out:
 	if (ret < 0) {
+		spin_lock(&o2hb_live_lock);
+		hb_task = reg->hr_task;
+		reg->hr_task = NULL;
+		spin_unlock(&o2hb_live_lock);
+
+		if (hb_task)
+			kthread_stop(hb_task);
+
+		o2hb_unmap_slot_data(reg);
+
 		fput(reg->hr_bdev_file);
 		reg->hr_bdev_file = NULL;
 	}
-- 
2.47.3


             reply	other threads:[~2026-03-30 15:35 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-03-30 15:34 Yufan Chen [this message]
2026-03-31  5:47 ` [PATCH v2] ocfs2/heartbeat: fix slot mapping rollback leaks on error paths Joseph Qi

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=20260330153428.19586-1-yufan.chen@linux.dev \
    --to=yufan.chen@linux.dev \
    --cc=jlbec@evilplan.org \
    --cc=joseph.qi@linux.alibaba.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mark@fasheh.com \
    --cc=ocfs2-devel@lists.linux.dev \
    /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