[PATCH v2 0/2] mm/swapfile: release swap info when swap device is unpluged

[PATCH v2 0/2] mm/swapfile: release swap info when swap device is unpluged

[liubo]

This series patch try to fix the problem that the resources
corresponding to swap are not effectively released when
the swap partition disk is hot-pluged.

1/2: Extract operations of resource release in the swapoff process.
2/2: release swap info when swap device is unpluged

This is a second version, which addressed some comments:
1. turn first patch into a two patch series. The first does the
code movement with no functional changes and the second does
the actual fix.

---
For v1, see: https://lore.kernel.org/linux-mm/20220528084941.28391-1-liubo254@huawei.com/
---

liubo (2):
  mm/swapfile: Extract operations of resource release in the swapoff
    process
  mm/swapfile: release swap info when swap device is unpluged

 mm/swapfile.c | 261 +++++++++++++++++++++++++++++++++++---------------
 1 file changed, 183 insertions(+), 78 deletions(-)

-- 
2.27.0

[PATCH v2 1/2] mm/swapfile: Extract operations of resource release in the swapoff process

[liubo]

In order to realize the interface reuse in the swapoff resource
release process, some of the operations are abstracted
into separate interfaces.

Only code movement, no functional modifications and changes.

del_useless_swap_info():
Remove specific swap_info_struct from swap_active_head and
update total_swap_pages.

release_swap_info_memory():
Clear the corresponding resources of swap_info_struct.

Signed-off-by: liubo <liubo254@huawei.com>
---
v2:
Only code movement, no functional modifications and changes.
Former: https://lore.kernel.org/linux-mm/20220528084941.28391-1-liubo254@huawei.com/
---
 mm/swapfile.c | 168 +++++++++++++++++++++++++++-----------------------
 1 file changed, 91 insertions(+), 77 deletions(-)

diff --git a/mm/swapfile.c b/mm/swapfile.c
index 3fa26f6971e9..2ef5e7b4918e 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -2386,18 +2386,103 @@ bool has_usable_swap(void)
 	return ret;
 }
 
-SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
+static void release_swap_info_memory(struct swap_info_struct *p)
 {
-	struct swap_info_struct *p = NULL;
 	unsigned char *swap_map;
 	struct swap_cluster_info *cluster_info;
 	unsigned long *frontswap_map;
-	struct file *swap_file, *victim;
+	struct file *swap_file;
 	struct address_space *mapping;
 	struct inode *inode;
+	unsigned int old_block_size;
+
+	mutex_lock(&swapon_mutex);
+	spin_lock(&swap_lock);
+	spin_lock(&p->lock);
+	drain_mmlist();
+
+	/* wait for anyone still in scan_swap_map */
+	p->highest_bit = 0;		/* cuts scans short */
+	while (p->flags >= SWP_SCANNING) {
+		spin_unlock(&p->lock);
+		spin_unlock(&swap_lock);
+		schedule_timeout_uninterruptible(1);
+		spin_lock(&swap_lock);
+		spin_lock(&p->lock);
+	}
+
+	swap_file = p->swap_file;
+	mapping = p->swap_file->f_mapping;
+	old_block_size = p->old_block_size;
+	p->swap_file = NULL;
+	p->max = 0;
+	swap_map = p->swap_map;
+	p->swap_map = NULL;
+	cluster_info = p->cluster_info;
+	p->cluster_info = NULL;
+	frontswap_map = frontswap_map_get(p);
+	spin_unlock(&p->lock);
+	spin_unlock(&swap_lock);
+	arch_swap_invalidate_area(p->type);
+	frontswap_invalidate_area(p->type);
+	frontswap_map_set(p, NULL);
+	mutex_unlock(&swapon_mutex);
+	free_percpu(p->percpu_cluster);
+	p->percpu_cluster = NULL;
+	free_percpu(p->cluster_next_cpu);
+	p->cluster_next_cpu = NULL;
+	vfree(swap_map);
+	kvfree(cluster_info);
+	kvfree(frontswap_map);
+	/* Destroy swap account information */
+	swap_cgroup_swapoff(p->type);
+	exit_swap_address_space(p->type);
+
+	inode = mapping->host;
+	if (S_ISBLK(inode->i_mode)) {
+		struct block_device *bdev = I_BDEV(inode);
+
+		set_blocksize(bdev, old_block_size);
+		blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
+	}
+
+	inode_lock(inode);
+	inode->i_flags &= ~S_SWAPFILE;
+	inode_unlock(inode);
+	filp_close(swap_file, NULL);
+}
+
+static void del_useless_swap_info(struct swap_info_struct *p)
+{
+	del_from_avail_list(p);
+	spin_lock(&p->lock);
+	if (p->prio < 0) {
+		struct swap_info_struct *si = p;
+		int nid;
+
+		plist_for_each_entry_continue(si, &swap_active_head, list) {
+			si->prio++;
+			si->list.prio--;
+			for_each_node(nid) {
+				if (si->avail_lists[nid].prio != 1)
+					si->avail_lists[nid].prio--;
+			}
+		}
+		least_priority++;
+	}
+	plist_del(&p->list, &swap_active_head);
+	atomic_long_sub(p->pages, &nr_swap_pages);
+	p->flags &= ~SWP_WRITEOK;
+	spin_unlock(&p->lock);
+}
+
+SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
+{
+	struct swap_info_struct *p = NULL;
+	struct file *victim;
+	struct address_space *mapping;
 	struct filename *pathname;
 	int err, found = 0;
-	unsigned int old_block_size;
 	unsigned int inuse_pages;
 
 	if (!capable(CAP_SYS_ADMIN))
@@ -2440,26 +2525,8 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
 		spin_unlock(&swap_lock);
 		goto out_dput;
 	}
-	del_from_avail_list(p);
-	spin_lock(&p->lock);
-	if (p->prio < 0) {
-		struct swap_info_struct *si = p;
-		int nid;
 
-		plist_for_each_entry_continue(si, &swap_active_head, list) {
-			si->prio++;
-			si->list.prio--;
-			for_each_node(nid) {
-				if (si->avail_lists[nid].prio != 1)
-					si->avail_lists[nid].prio--;
-			}
-		}
-		least_priority++;
-	}
-	plist_del(&p->list, &swap_active_head);
-	atomic_long_sub(p->pages, &nr_swap_pages);
-	p->flags &= ~SWP_WRITEOK;
-	spin_unlock(&p->lock);
+	del_useless_swap_info(p);
 	spin_unlock(&swap_lock);
 
 	disable_swap_slots_cache_lock();
@@ -2497,60 +2564,7 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
 	if (!p->bdev || !bdev_nonrot(p->bdev))
 		atomic_dec(&nr_rotate_swap);
 
-	mutex_lock(&swapon_mutex);
-	spin_lock(&swap_lock);
-	spin_lock(&p->lock);
-	drain_mmlist();
-
-	/* wait for anyone still in scan_swap_map_slots */
-	p->highest_bit = 0;		/* cuts scans short */
-	while (p->flags >= SWP_SCANNING) {
-		spin_unlock(&p->lock);
-		spin_unlock(&swap_lock);
-		schedule_timeout_uninterruptible(1);
-		spin_lock(&swap_lock);
-		spin_lock(&p->lock);
-	}
-
-	swap_file = p->swap_file;
-	old_block_size = p->old_block_size;
-	p->swap_file = NULL;
-	p->max = 0;
-	swap_map = p->swap_map;
-	p->swap_map = NULL;
-	cluster_info = p->cluster_info;
-	p->cluster_info = NULL;
-	frontswap_map = frontswap_map_get(p);
-	spin_unlock(&p->lock);
-	spin_unlock(&swap_lock);
-	arch_swap_invalidate_area(p->type);
-	frontswap_invalidate_area(p->type);
-	frontswap_map_set(p, NULL);
-	mutex_unlock(&swapon_mutex);
-	free_percpu(p->percpu_cluster);
-	p->percpu_cluster = NULL;
-	free_percpu(p->cluster_next_cpu);
-	p->cluster_next_cpu = NULL;
-	vfree(swap_map);
-	kvfree(cluster_info);
-	kvfree(frontswap_map);
-	/* Destroy swap account information */
-	swap_cgroup_swapoff(p->type);
-	exit_swap_address_space(p->type);
-
-	inode = mapping->host;
-	if (S_ISBLK(inode->i_mode)) {
-		struct block_device *bdev = I_BDEV(inode);
-
-		set_blocksize(bdev, old_block_size);
-		blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
-	}
-
-	inode_lock(inode);
-	inode->i_flags &= ~S_SWAPFILE;
-	inode_unlock(inode);
-	filp_close(swap_file, NULL);
-
+	release_swap_info_memory(p);
 	/*
 	 * Clear the SWP_USED flag after all resources are freed so that swapon
 	 * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
-- 
2.27.0

[PATCH v2 2/2] mm/swapfile: release swap info when swap device is unpluged

[liubo]

When the swap partition is mounted through the swapon command,
the kernel will create the swap_info_struct data structure
and initialize it, and save it in the swap_info global array.
When the swap partition is no longer in use, the disk is
unloaded through the swapoff command.

However, if the disk is pulled out after swapon, an error will
occur when swapoff the disk, causing the swap_info_struct
data structure to remain in the kernel and cannot be cleared.

This patch identifies which disks are no longer available
by adding a traversal operation for swap_active_head available
swap partitions in the swapon and swapoff processes,
so as to clear the above data structures and
release the corresponding resources.

Example:
[root@localhost ~]# swapon -s
[root@localhost ~]# lsblk
NAME             MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sda                8:0    0  1.1T  0 disk
├─sda1             8:1    0  600M  0 part /boot/efi
├─sda2             8:2    0    1G  0 part /boot
└─sda3             8:3    0  1.1T  0 part
  ├─root 253:0    0   70G  0 lvm  /
  ├─swap 253:1    0    4G  0 lvm
  └─home 253:2    0    1T  0 lvm  /home
nvme0n1          259:0    0  3.6T  0 disk
└─nvme0n1p1      259:5    0   60G  0 part
[root@localhost ~]# swapon /dev/nvme0n1p1
[root@localhost ~]# swapon -s
Filename                   Type            Size    Used    Priority
/dev/nvme0n1p1             partition       62914556        0       -2
[root@localhost ~]# echo 1 > /sys/bus/pci/devices/0000:d8:00.0/remove
[root@localhost ~]# lsblk
NAME             MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sda                8:0    0  1.1T  0 disk
├─sda1             8:1    0  600M  0 part /boot/efi
├─sda2             8:2    0    1G  0 part /boot
└─sda3             8:3    0  1.1T  0 part
  ├─root 253:0    0   70G  0 lvm  /
  ├─swap 253:1    0    4G  0 lvm
  └─home 253:2    0    1T  0 lvm  /home
[root@localhost ~]# swapon -s
Filename                   Type            Size    Used    Priority
/dev/nvme0n1p1             partition       62914556        0       -2
[root@localhost ~]# swapoff /dev/nvme0n1p1
swapoff: /dev/nvme0n1p1: swapoff failed: No such file or directory
[root@localhost ~]# swapoff -a
[root@localhost ~]# swapon -s
Filename                   Type            Size    Used    Priority
/dev/nvme0n1p1             partition       62914556        0       -2

In the swapoff command, the device is acquired in the following ways,
but the device has been unplugged at this time, causing the "victim"
acquisition to fail, thus returning an error directly.
And the invalid swap_info_struct cannot be effectively released.

pathname = getname(specialfile);
if (IS_ERR(pathname))
	return PTR_ERR(pathname);

victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
err = PTR_ERR(victim);
if (IS_ERR(victim))
	goto out;

In order to solve the above problems, by adding traversal of
swap_avail_heads (available swap partitions) in the swapoff
and swapon processes,
find the swap_info_struct whose disk partition has been
unplugged, and release resources.

The reason why the judgment of unavailable swap information is also
added to the swapon process is that the swapoff is executed by the
user, and the timing is uncontrollable.
The system supports swapon multiple disks, and the unavailable swap
can be deleted at the same time as swapon is mounted.

Signed-off-by: liubo <liubo254@huawei.com>
---
v2:
The possible solution to the problem that the swap resources
are not released when the swap partition is hot-pluged.
The actual fix of the former patch.

Former:https://lore.kernel.org/linux-mm/20220528084941.28391-1-liubo254@huawei.com/
---
 mm/swapfile.c | 93 ++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 92 insertions(+), 1 deletion(-)

diff --git a/mm/swapfile.c b/mm/swapfile.c
index 2ef5e7b4918e..91208de9f689 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -68,6 +68,7 @@ static const char Bad_file[] = "Bad swap file entry ";
 static const char Unused_file[] = "Unused swap file entry ";
 static const char Bad_offset[] = "Bad swap offset entry ";
 static const char Unused_offset[] = "Unused swap offset entry ";
+static const char invalid_info[] = "deleted";
 
 /*
  * all active swap_info_structs
@@ -2476,6 +2477,88 @@ static void del_useless_swap_info(struct swap_info_struct *p)
 	spin_unlock(&p->lock);
 }
 
+static int swapoff_invalid_swapinfo(void)
+{
+	struct swap_info_struct *p = NULL;
+	struct file *swap_file;
+	int err, found = 0;
+
+	char *tmp = NULL;
+	char *swap_name = NULL;
+
+	tmp = kvzalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!tmp)
+		return -ENOMEM;
+rescan:
+	memset(tmp, 0, PAGE_SIZE);
+	spin_lock(&swap_lock);
+	plist_for_each_entry(p, &swap_active_head, list) {
+		if (p->flags & SWP_WRITEOK) {
+			swap_file = p->swap_file;
+			swap_name = d_path(&swap_file->f_path, tmp, PAGE_SIZE);
+
+			if (strstr(swap_name, invalid_info)) {
+				found = 1;
+				break;
+			}
+		}
+	}
+
+	if (!found) {
+		err = 0;
+		spin_unlock(&swap_lock);
+		goto out;
+	}
+
+	total_swap_pages -= p->pages;
+
+	del_useless_swap_info(p);
+	spin_unlock(&swap_lock);
+
+	disable_swap_slots_cache_lock();
+	set_current_oom_origin();
+	try_to_unuse(p->type);
+	clear_current_oom_origin();
+
+	reenable_swap_slots_cache_unlock();
+
+	/*
+	 * wait for swap operations protected by get/put_swap_device()
+	 * to complete
+	 */
+	synchronize_rcu();
+
+	flush_work(&p->discard_work);
+
+	destroy_swap_extents(p);
+	if (p->flags & SWP_CONTINUED)
+		free_swap_count_continuations(p);
+
+	if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
+		atomic_dec(&nr_rotate_swap);
+
+	release_swap_info_memory(p);
+
+	/*
+	 * Clear the SWP_USED flag after all resources are freed so that swapon
+	 * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
+	 * not hold p->lock after we cleared its SWP_WRITEOK.
+	 */
+	spin_lock(&swap_lock);
+	p->flags = 0;
+	spin_unlock(&swap_lock);
+
+	err = 0;
+	atomic_inc(&proc_poll_event);
+	wake_up_interruptible(&proc_poll_wait);
+
+	found = 0;
+	goto rescan;
+out:
+	kfree(tmp);
+	return err;
+}
+
 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
 {
 	struct swap_info_struct *p = NULL;
@@ -2496,8 +2579,12 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
 
 	victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
 	err = PTR_ERR(victim);
-	if (IS_ERR(victim))
+	if (IS_ERR(victim)) {
+		/* check if the pathname is a device that has been unpluged */
+		err = swapoff_invalid_swapinfo();
+		err = err < 0 ? err : PTR_ERR(victim);
 		goto out;
+	}
 
 	mapping = victim->f_mapping;
 	spin_lock(&swap_lock);
@@ -3028,6 +3115,10 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
 	if (!swap_avail_heads)
 		return -ENOMEM;
 
+	error = swapoff_invalid_swapinfo();
+	if (error < 0)
+		return error;
+
 	p = alloc_swap_info();
 	if (IS_ERR(p))
 		return PTR_ERR(p);
-- 
2.27.0

Re: [PATCH v2 2/2] mm/swapfile: release swap info when swap device is unpluged

[Christoph Hellwig]

Sorry, but this is simply broken.  We can't just do d_path magic
here.  That being said the problem is real, and I think the right
answer is to offer a swapoff variant that doesn't need to open the
underlying device or file, but useѕ some kind of handle.  I'm just
not sure what would be a good handle here.

Re: [PATCH v2 2/2] mm/swapfile: release swap info when swap device is unpluged

[Miaohe Lin]

On 2022/6/15 19:22, Christoph Hellwig wrote:
> Sorry, but this is simply broken.  We can't just do d_path magic
> here.  That being said the problem is real, and I think the right
> answer is to offer a swapoff variant that doesn't need to open the
> underlying device or file, but useѕ some kind of handle.  I'm just

When swap device is unplugged, the swap_file->f_path.dentry will be in the
d_unlinked() status. Could we use it to detect the problem instead? IMHO, we
can found the unplugged swap devices this way and further get the right swap
device to swapoff through comparing the swap device name with @specialfile.
Or am I miss something?

> not sure what would be a good handle here.

Using handle sounds like a good idea too. :) But can the above solution be acceptable?

> .
> 

Thanks!