[PATCH v4 0/4] zonefs: introduce helper for zone management

[PATCH v4 0/4] zonefs: introduce helper for zone management

[Johannes Thumshirn]

Introduce a mount option for explicitly opening a device's zones when opening
the seq zone file for writing. This way we prevent resource exhaustion on
devices that export a maximum open zones limit. 

Changes to v3:
- Introduce unlocked zonefs_io_error
- Rework failure of zone close on file close

Changes to v2:
- Clear ZONEFS_ZONE_OPEN flag on error
- Fix missing newline

Johannes Thumshirn (4):
  zonefs: introduce helper for zone management
  zonefs: provide zonefs_io_error variant that can be called with
    i_truncate_mutex held
  zonefs: open/close zone on file open/close
  zonefs: document the explicit-open mount option

 Documentation/filesystems/zonefs.rst |  15 ++
 fs/zonefs/super.c                    | 239 ++++++++++++++++++++++++---
 fs/zonefs/zonefs.h                   |  10 ++
 3 files changed, 244 insertions(+), 20 deletions(-)

-- 
2.26.2

[PATCH v4 1/4] zonefs: introduce helper for zone management

[Johannes Thumshirn]

Introduce a helper function for sending zone management commands to the
block device.

As zone management commands can change a zone write pointer position
reflected in the size of the zone file, this function expects the truncate
mutex to be held.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
---
Changes to v2:
- add missing '\n'

Changes to v1:
- centralize failure logging
---
 fs/zonefs/super.c | 29 ++++++++++++++++++++++-------
 1 file changed, 22 insertions(+), 7 deletions(-)

diff --git a/fs/zonefs/super.c b/fs/zonefs/super.c
index 8ec7c8f109d7..6e13a5127b01 100644
--- a/fs/zonefs/super.c
+++ b/fs/zonefs/super.c
@@ -24,6 +24,26 @@
 
 #include "zonefs.h"
 
+static inline int zonefs_zone_mgmt(struct inode *inode,
+				   enum req_opf op)
+{
+	struct zonefs_inode_info *zi = ZONEFS_I(inode);
+	int ret;
+
+	lockdep_assert_held(&zi->i_truncate_mutex);
+
+	ret = blkdev_zone_mgmt(inode->i_sb->s_bdev, op, zi->i_zsector,
+			       zi->i_zone_size >> SECTOR_SHIFT, GFP_NOFS);
+	if (ret) {
+		zonefs_err(inode->i_sb,
+			   "Zone management operation %s at %llu failed %d\n",
+			   blk_op_str(op), zi->i_zsector, ret);
+		return ret;
+	}
+
+	return 0;
+}
+
 static int zonefs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 			      unsigned int flags, struct iomap *iomap,
 			      struct iomap *srcmap)
@@ -397,14 +417,9 @@ static int zonefs_file_truncate(struct inode *inode, loff_t isize)
 	if (isize == old_isize)
 		goto unlock;
 
-	ret = blkdev_zone_mgmt(inode->i_sb->s_bdev, op, zi->i_zsector,
-			       zi->i_zone_size >> SECTOR_SHIFT, GFP_NOFS);
-	if (ret) {
-		zonefs_err(inode->i_sb,
-			   "Zone management operation at %llu failed %d",
-			   zi->i_zsector, ret);
+	ret = zonefs_zone_mgmt(inode, op);
+	if (ret)
 		goto unlock;
-	}
 
 	zonefs_update_stats(inode, isize);
 	truncate_setsize(inode, isize);
-- 
2.26.2

[PATCH v4 2/4] zonefs: provide zonefs_io_error variant that can be called with i_truncate_mutex held

[Johannes Thumshirn]

Subsequent patches need to call zonefs_io_error() with the i_truncate_mutex
already held, so factor out the body of zonefs_io_error() into
__zonefs_io_error() which can be called from with the i_truncate_mutex
held.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---
 fs/zonefs/super.c | 25 +++++++++++++++----------
 1 file changed, 15 insertions(+), 10 deletions(-)

diff --git a/fs/zonefs/super.c b/fs/zonefs/super.c
index 6e13a5127b01..3db28a06e1a2 100644
--- a/fs/zonefs/super.c
+++ b/fs/zonefs/super.c
@@ -341,14 +341,7 @@ static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
 	return 0;
 }
 
-/*
- * When an file IO error occurs, check the file zone to see if there is a change
- * in the zone condition (e.g. offline or read-only). For a failed write to a
- * sequential zone, the zone write pointer position must also be checked to
- * eventually correct the file size and zonefs inode write pointer offset
- * (which can be out of sync with the drive due to partial write failures).
- */
-static void zonefs_io_error(struct inode *inode, bool write)
+static void __zonefs_io_error(struct inode *inode, bool write)
 {
 	struct zonefs_inode_info *zi = ZONEFS_I(inode);
 	struct super_block *sb = inode->i_sb;
@@ -362,8 +355,6 @@ static void zonefs_io_error(struct inode *inode, bool write)
 	};
 	int ret;
 
-	mutex_lock(&zi->i_truncate_mutex);
-
 	/*
 	 * Memory allocations in blkdev_report_zones() can trigger a memory
 	 * reclaim which may in turn cause a recursion into zonefs as well as
@@ -379,7 +370,21 @@ static void zonefs_io_error(struct inode *inode, bool write)
 		zonefs_err(sb, "Get inode %lu zone information failed %d\n",
 			   inode->i_ino, ret);
 	memalloc_noio_restore(noio_flag);
+}
 
+/*
+ * When an file IO error occurs, check the file zone to see if there is a change
+ * in the zone condition (e.g. offline or read-only). For a failed write to a
+ * sequential zone, the zone write pointer position must also be checked to
+ * eventually correct the file size and zonefs inode write pointer offset
+ * (which can be out of sync with the drive due to partial write failures).
+ */
+static void zonefs_io_error(struct inode *inode, bool write)
+{
+	struct zonefs_inode_info *zi = ZONEFS_I(inode);
+
+	mutex_lock(&zi->i_truncate_mutex);
+	__zonefs_io_error(inode, write);
 	mutex_unlock(&zi->i_truncate_mutex);
 }
 
-- 
2.26.2

Re: [PATCH v4 2/4] zonefs: provide zonefs_io_error variant that can be called with i_truncate_mutex held

[Damien Le Moal]

On 2020/09/10 23:37, Johannes Thumshirn wrote:
> Subsequent patches need to call zonefs_io_error() with the i_truncate_mutex
> already held, so factor out the body of zonefs_io_error() into
> __zonefs_io_error() which can be called from with the i_truncate_mutex
> held.
> 
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> ---
>  fs/zonefs/super.c | 25 +++++++++++++++----------
>  1 file changed, 15 insertions(+), 10 deletions(-)
> 
> diff --git a/fs/zonefs/super.c b/fs/zonefs/super.c
> index 6e13a5127b01..3db28a06e1a2 100644
> --- a/fs/zonefs/super.c
> +++ b/fs/zonefs/super.c
> @@ -341,14 +341,7 @@ static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
>  	return 0;
>  }
>  
> -/*
> - * When an file IO error occurs, check the file zone to see if there is a change
> - * in the zone condition (e.g. offline or read-only). For a failed write to a
> - * sequential zone, the zone write pointer position must also be checked to
> - * eventually correct the file size and zonefs inode write pointer offset
> - * (which can be out of sync with the drive due to partial write failures).
> - */
> -static void zonefs_io_error(struct inode *inode, bool write)
> +static void __zonefs_io_error(struct inode *inode, bool write)
>  {
>  	struct zonefs_inode_info *zi = ZONEFS_I(inode);
>  	struct super_block *sb = inode->i_sb;
> @@ -362,8 +355,6 @@ static void zonefs_io_error(struct inode *inode, bool write)
>  	};
>  	int ret;
>  
> -	mutex_lock(&zi->i_truncate_mutex);
> -
>  	/*
>  	 * Memory allocations in blkdev_report_zones() can trigger a memory
>  	 * reclaim which may in turn cause a recursion into zonefs as well as
> @@ -379,7 +370,21 @@ static void zonefs_io_error(struct inode *inode, bool write)
>  		zonefs_err(sb, "Get inode %lu zone information failed %d\n",
>  			   inode->i_ino, ret);
>  	memalloc_noio_restore(noio_flag);
> +}
>  
> +/*
> + * When an file IO error occurs, check the file zone to see if there is a change
> + * in the zone condition (e.g. offline or read-only). For a failed write to a
> + * sequential zone, the zone write pointer position must also be checked to
> + * eventually correct the file size and zonefs inode write pointer offset
> + * (which can be out of sync with the drive due to partial write failures).
> + */

I would prefer to leave this comment attached to the function body, so
__zonefs_io_error() now.

> +static void zonefs_io_error(struct inode *inode, bool write)
> +{
> +	struct zonefs_inode_info *zi = ZONEFS_I(inode);
> +
> +	mutex_lock(&zi->i_truncate_mutex);
> +	__zonefs_io_error(inode, write);
>  	mutex_unlock(&zi->i_truncate_mutex);
>  }
>  
> 

Apart from this nit, looks good.


-- 
Damien Le Moal
Western Digital Research

Re: [PATCH v4 2/4] zonefs: provide zonefs_io_error variant that can be called with i_truncate_mutex held

[Johannes Thumshirn]

On 11/09/2020 07:05, Damien Le Moal wrote:
> I would prefer to leave this comment attached to the function body, so
> __zonefs_io_error() now.

Also fixed.

Last test run is running and then I'm sending out the new version.

[PATCH v4 3/4] zonefs: open/close zone on file open/close

[Johannes Thumshirn]

NVMe Zoned Namespace introduced the concept of active zones, which are
zones in the implicit open, explicit open or closed condition. Drives may
have a limit on the number of zones that can be simultaneously active.
This potential limitation translate into a risk for applications to see
write IO errors due to this limit if the zone of a file being written to is
not already active when a write request is issued.

To avoid these potential errors, the zone of a file can explicitly be made
active using an open zone command when the file is open for the first
time. If the zone open command succeeds, the application is then
guaranteed that write requests can be processed. This indirect management
of active zones relies on the maximum number of open zones of a drive,
which is always lower or equal to the maximum number of active zones.

On the first open of a sequential zone file, send a REQ_OP_ZONE_OPEN
command to the block device. Conversely, on the last release of a zone
file and send a REQ_OP_ZONE_CLOSE to the device if the zone is not full or
empty.

As truncating a zone file to 0 or max can deactivate a zone as well, we
need to serialize against truncates and also be careful not to close a
zone as the file may still be open for writing, e.g. the user called
ftruncate(). If the zone file is not open and a process does a truncate(),
then no close operation is needed.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---
Changes to v2:
- Clear open flag on error (Damien)
---
 fs/zonefs/super.c  | 187 ++++++++++++++++++++++++++++++++++++++++++++-
 fs/zonefs/zonefs.h |  10 +++
 2 files changed, 193 insertions(+), 4 deletions(-)

diff --git a/fs/zonefs/super.c b/fs/zonefs/super.c
index 3db28a06e1a2..0b266b4212df 100644
--- a/fs/zonefs/super.c
+++ b/fs/zonefs/super.c
@@ -44,6 +44,19 @@ static inline int zonefs_zone_mgmt(struct inode *inode,
 	return 0;
 }
 
+static inline void zonefs_i_size_write(struct inode *inode, loff_t isize)
+{
+	struct zonefs_inode_info *zi = ZONEFS_I(inode);
+
+	i_size_write(inode, isize);
+	/*
+	 * A full zone is no longer open/active and does not need
+	 * explicit closing.
+	 */
+	if (isize >= zi->i_max_size)
+		zi->i_flags &= ~ZONEFS_ZONE_OPEN;
+}
+
 static int zonefs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 			      unsigned int flags, struct iomap *iomap,
 			      struct iomap *srcmap)
@@ -321,6 +334,17 @@ static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
 		}
 	}
 
+	/*
+	 * If the filesystem is mounted with the explicit-open mount option, we
+	 * need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to
+	 * the read-only or offline condition, to avoid attempting an explicit
+	 * close of the zone when the inode file is closed.
+	 */
+	if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) &&
+	    (zone->cond == BLK_ZONE_COND_OFFLINE ||
+	     zone->cond == BLK_ZONE_COND_READONLY))
+		zi->i_flags &= ~ZONEFS_ZONE_OPEN;
+
 	/*
 	 * If error=remount-ro was specified, any error result in remounting
 	 * the volume as read-only.
@@ -335,7 +359,7 @@ static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
 	 * invalid data.
 	 */
 	zonefs_update_stats(inode, data_size);
-	i_size_write(inode, data_size);
+	zonefs_i_size_write(inode, data_size);
 	zi->i_wpoffset = data_size;
 
 	return 0;
@@ -426,6 +450,25 @@ static int zonefs_file_truncate(struct inode *inode, loff_t isize)
 	if (ret)
 		goto unlock;
 
+	/*
+	 * If the mount option ZONEFS_MNTOPT_EXPLICIT_OPEN is set,
+	 * take care of open zones.
+	 */
+	if (zi->i_flags & ZONEFS_ZONE_OPEN) {
+		/*
+		 * Truncating a zone to EMPTY or FULL is the equivalent of
+		 * closing the zone. For a truncation to 0, we need to
+		 * re-open the zone to ensure new writes can be processed.
+		 * For a truncation to the maximum file size, the zone is
+		 * closed and writes cannot be accepted anymore, so clear
+		 * the open flag.
+		 */
+		if (!isize)
+			ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_OPEN);
+		else
+			zi->i_flags &= ~ZONEFS_ZONE_OPEN;
+	}
+
 	zonefs_update_stats(inode, isize);
 	truncate_setsize(inode, isize);
 	zi->i_wpoffset = isize;
@@ -604,7 +647,7 @@ static int zonefs_file_write_dio_end_io(struct kiocb *iocb, ssize_t size,
 		mutex_lock(&zi->i_truncate_mutex);
 		if (i_size_read(inode) < iocb->ki_pos + size) {
 			zonefs_update_stats(inode, iocb->ki_pos + size);
-			i_size_write(inode, iocb->ki_pos + size);
+			zonefs_i_size_write(inode, iocb->ki_pos + size);
 		}
 		mutex_unlock(&zi->i_truncate_mutex);
 	}
@@ -885,8 +928,132 @@ static ssize_t zonefs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 	return ret;
 }
 
+static inline bool zonefs_file_use_exp_open(struct inode *inode, struct file *file)
+{
+	struct zonefs_inode_info *zi = ZONEFS_I(inode);
+	struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
+
+	if (!(sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN))
+		return false;
+
+	if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
+		return false;
+
+	if (!(file->f_mode & FMODE_WRITE))
+		return false;
+
+	return true;
+}
+
+static int zonefs_open_zone(struct inode *inode)
+{
+	struct zonefs_inode_info *zi = ZONEFS_I(inode);
+	struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
+	int ret = 0;
+
+	mutex_lock(&zi->i_truncate_mutex);
+
+	zi->i_wr_refcnt++;
+	if (zi->i_wr_refcnt == 1) {
+
+		if (atomic_inc_return(&sbi->s_open_zones) > sbi->s_max_open_zones) {
+			atomic_dec(&sbi->s_open_zones);
+			ret = -EBUSY;
+			goto unlock;
+		}
+
+		if (i_size_read(inode) < zi->i_max_size) {
+			ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_OPEN);
+			if (ret) {
+				zi->i_wr_refcnt--;
+				atomic_dec(&sbi->s_open_zones);
+				goto unlock;
+			}
+			zi->i_flags |= ZONEFS_ZONE_OPEN;
+		}
+	}
+
+unlock:
+	mutex_unlock(&zi->i_truncate_mutex);
+
+	return ret;
+}
+
+static int zonefs_file_open(struct inode *inode, struct file *file)
+{
+	int ret;
+
+	ret = generic_file_open(inode, file);
+	if (ret)
+		return ret;
+
+	if (zonefs_file_use_exp_open(inode, file))
+		return zonefs_open_zone(inode);
+
+	return 0;
+}
+
+static void zonefs_close_zone(struct inode *inode)
+{
+	struct zonefs_inode_info *zi = ZONEFS_I(inode);
+	int ret = 0;
+
+	mutex_lock(&zi->i_truncate_mutex);
+	zi->i_wr_refcnt--;
+	if (!zi->i_wr_refcnt) {
+		struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
+		struct super_block *sb = inode->i_sb;
+
+		/*
+		 * If the file zone is full, it is not open anymore and we only
+		 * need to decrement the open count.
+		 */
+		if (!(zi->i_flags & ZONEFS_ZONE_OPEN))
+			goto dec;
+
+		ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_CLOSE);
+		if (ret) {
+			__zonefs_io_error(inode, false);
+			/*
+			 * If the zone open flag is still set, it means that
+			 * the zone is neither read-only nor offline now. This
+			 * can only happen if sending the REQ_OP_ZONE_CLOSE
+			 * request to the drive failed, for whatever reasons.
+			 *
+			 * It's better to mark the FS read-only (if
+			 * __zonefs_io_error hasn't done that already) than to
+			 * loose any data.
+			 */
+			if (zi->i_flags & ZONEFS_ZONE_OPEN &&
+			    !(sb->s_flags & SB_RDONLY)) {
+				zonefs_warn(sb, "closing zone failed, remounting filesystem read-only\n");
+				sb->s_flags |= SB_RDONLY;
+			}
+		}
+		zi->i_flags &= ~ZONEFS_ZONE_OPEN;
+dec:
+		atomic_dec(&sbi->s_open_zones);
+	}
+	mutex_unlock(&zi->i_truncate_mutex);
+}
+
+static int zonefs_file_release(struct inode *inode, struct file *file)
+{
+	/*
+	 * If we explicitly open a zone we must close it again as well, but the
+	 * zone management operation can fail (either due to an IO error or as
+	 * the zone has gone offline or read-only). Make sure we don't fail the
+	 * close(2) for user-space.
+	 */
+	if (zonefs_file_use_exp_open(inode, file))
+		zonefs_close_zone(inode);
+
+	return 0;
+}
+
 static const struct file_operations zonefs_file_operations = {
-	.open		= generic_file_open,
+	.open		= zonefs_file_open,
+	.release	= zonefs_file_release,
 	.fsync		= zonefs_file_fsync,
 	.mmap		= zonefs_file_mmap,
 	.llseek		= zonefs_file_llseek,
@@ -910,6 +1077,7 @@ static struct inode *zonefs_alloc_inode(struct super_block *sb)
 	inode_init_once(&zi->i_vnode);
 	mutex_init(&zi->i_truncate_mutex);
 	init_rwsem(&zi->i_mmap_sem);
+	zi->i_wr_refcnt = 0;
 
 	return &zi->i_vnode;
 }
@@ -960,7 +1128,7 @@ static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
 
 enum {
 	Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
-	Opt_err,
+	Opt_explicit_open, Opt_err,
 };
 
 static const match_table_t tokens = {
@@ -968,6 +1136,7 @@ static const match_table_t tokens = {
 	{ Opt_errors_zro,	"errors=zone-ro"},
 	{ Opt_errors_zol,	"errors=zone-offline"},
 	{ Opt_errors_repair,	"errors=repair"},
+	{ Opt_explicit_open,	"explicit-open" },
 	{ Opt_err,		NULL}
 };
 
@@ -1004,6 +1173,9 @@ static int zonefs_parse_options(struct super_block *sb, char *options)
 			sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
 			sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
 			break;
+		case Opt_explicit_open:
+			sbi->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN;
+			break;
 		default:
 			return -EINVAL;
 		}
@@ -1423,6 +1595,13 @@ static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
 	sbi->s_gid = GLOBAL_ROOT_GID;
 	sbi->s_perm = 0640;
 	sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
+	sbi->s_max_open_zones = bdev_max_open_zones(sb->s_bdev);
+	atomic_set(&sbi->s_open_zones, 0);
+	if (!sbi->s_max_open_zones &&
+	    sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
+		zonefs_info(sb, "No open zones limit. Ignoring explicit_open mount option\n");
+		sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN;
+	}
 
 	ret = zonefs_read_super(sb);
 	if (ret)
diff --git a/fs/zonefs/zonefs.h b/fs/zonefs/zonefs.h
index 55b39970acb2..51141907097c 100644
--- a/fs/zonefs/zonefs.h
+++ b/fs/zonefs/zonefs.h
@@ -38,6 +38,8 @@ static inline enum zonefs_ztype zonefs_zone_type(struct blk_zone *zone)
 	return ZONEFS_ZTYPE_SEQ;
 }
 
+#define ZONEFS_ZONE_OPEN	(1 << 0)
+
 /*
  * In-memory inode data.
  */
@@ -74,6 +76,10 @@ struct zonefs_inode_info {
 	 */
 	struct mutex		i_truncate_mutex;
 	struct rw_semaphore	i_mmap_sem;
+
+	/* guarded by i_truncate_mutex */
+	unsigned int		i_wr_refcnt;
+	unsigned int		i_flags;
 };
 
 static inline struct zonefs_inode_info *ZONEFS_I(struct inode *inode)
@@ -154,6 +160,7 @@ enum zonefs_features {
 #define ZONEFS_MNTOPT_ERRORS_MASK	\
 	(ZONEFS_MNTOPT_ERRORS_RO | ZONEFS_MNTOPT_ERRORS_ZRO | \
 	 ZONEFS_MNTOPT_ERRORS_ZOL | ZONEFS_MNTOPT_ERRORS_REPAIR)
+#define ZONEFS_MNTOPT_EXPLICIT_OPEN	(1 << 4) /* Explicit open/close of zones on open/close */
 
 /*
  * In-memory Super block information.
@@ -175,6 +182,9 @@ struct zonefs_sb_info {
 
 	loff_t			s_blocks;
 	loff_t			s_used_blocks;
+
+	unsigned int		s_max_open_zones;
+	atomic_t		s_open_zones;
 };
 
 static inline struct zonefs_sb_info *ZONEFS_SB(struct super_block *sb)
-- 
2.26.2

Re: [PATCH v4 3/4] zonefs: open/close zone on file open/close

[Damien Le Moal]

On 2020/09/10 23:37, Johannes Thumshirn wrote:
> NVMe Zoned Namespace introduced the concept of active zones, which are
> zones in the implicit open, explicit open or closed condition. Drives may
> have a limit on the number of zones that can be simultaneously active.
> This potential limitation translate into a risk for applications to see
> write IO errors due to this limit if the zone of a file being written to is
> not already active when a write request is issued.
> 
> To avoid these potential errors, the zone of a file can explicitly be made
> active using an open zone command when the file is open for the first
> time. If the zone open command succeeds, the application is then
> guaranteed that write requests can be processed. This indirect management
> of active zones relies on the maximum number of open zones of a drive,
> which is always lower or equal to the maximum number of active zones.
> 
> On the first open of a sequential zone file, send a REQ_OP_ZONE_OPEN
> command to the block device. Conversely, on the last release of a zone
> file and send a REQ_OP_ZONE_CLOSE to the device if the zone is not full or
> empty.
> 
> As truncating a zone file to 0 or max can deactivate a zone as well, we
> need to serialize against truncates and also be careful not to close a
> zone as the file may still be open for writing, e.g. the user called
> ftruncate(). If the zone file is not open and a process does a truncate(),
> then no close operation is needed.
> 
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> ---
> Changes to v2:
> - Clear open flag on error (Damien)
> ---
>  fs/zonefs/super.c  | 187 ++++++++++++++++++++++++++++++++++++++++++++-
>  fs/zonefs/zonefs.h |  10 +++
>  2 files changed, 193 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/zonefs/super.c b/fs/zonefs/super.c
> index 3db28a06e1a2..0b266b4212df 100644
> --- a/fs/zonefs/super.c
> +++ b/fs/zonefs/super.c
> @@ -44,6 +44,19 @@ static inline int zonefs_zone_mgmt(struct inode *inode,
>  	return 0;
>  }
>  
> +static inline void zonefs_i_size_write(struct inode *inode, loff_t isize)
> +{
> +	struct zonefs_inode_info *zi = ZONEFS_I(inode);
> +
> +	i_size_write(inode, isize);
> +	/*
> +	 * A full zone is no longer open/active and does not need
> +	 * explicit closing.
> +	 */
> +	if (isize >= zi->i_max_size)
> +		zi->i_flags &= ~ZONEFS_ZONE_OPEN;
> +}
> +
>  static int zonefs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
>  			      unsigned int flags, struct iomap *iomap,
>  			      struct iomap *srcmap)
> @@ -321,6 +334,17 @@ static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
>  		}
>  	}
>  
> +	/*
> +	 * If the filesystem is mounted with the explicit-open mount option, we
> +	 * need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to
> +	 * the read-only or offline condition, to avoid attempting an explicit
> +	 * close of the zone when the inode file is closed.
> +	 */
> +	if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) &&
> +	    (zone->cond == BLK_ZONE_COND_OFFLINE ||
> +	     zone->cond == BLK_ZONE_COND_READONLY))
> +		zi->i_flags &= ~ZONEFS_ZONE_OPEN;
> +
>  	/*
>  	 * If error=remount-ro was specified, any error result in remounting
>  	 * the volume as read-only.
> @@ -335,7 +359,7 @@ static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
>  	 * invalid data.
>  	 */
>  	zonefs_update_stats(inode, data_size);
> -	i_size_write(inode, data_size);
> +	zonefs_i_size_write(inode, data_size);
>  	zi->i_wpoffset = data_size;
>  
>  	return 0;
> @@ -426,6 +450,25 @@ static int zonefs_file_truncate(struct inode *inode, loff_t isize)
>  	if (ret)
>  		goto unlock;
>  
> +	/*
> +	 * If the mount option ZONEFS_MNTOPT_EXPLICIT_OPEN is set,
> +	 * take care of open zones.
> +	 */
> +	if (zi->i_flags & ZONEFS_ZONE_OPEN) {
> +		/*
> +		 * Truncating a zone to EMPTY or FULL is the equivalent of
> +		 * closing the zone. For a truncation to 0, we need to
> +		 * re-open the zone to ensure new writes can be processed.
> +		 * For a truncation to the maximum file size, the zone is
> +		 * closed and writes cannot be accepted anymore, so clear
> +		 * the open flag.
> +		 */
> +		if (!isize)
> +			ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_OPEN);
> +		else
> +			zi->i_flags &= ~ZONEFS_ZONE_OPEN;
> +	}
> +
>  	zonefs_update_stats(inode, isize);
>  	truncate_setsize(inode, isize);
>  	zi->i_wpoffset = isize;
> @@ -604,7 +647,7 @@ static int zonefs_file_write_dio_end_io(struct kiocb *iocb, ssize_t size,
>  		mutex_lock(&zi->i_truncate_mutex);
>  		if (i_size_read(inode) < iocb->ki_pos + size) {
>  			zonefs_update_stats(inode, iocb->ki_pos + size);
> -			i_size_write(inode, iocb->ki_pos + size);
> +			zonefs_i_size_write(inode, iocb->ki_pos + size);
>  		}
>  		mutex_unlock(&zi->i_truncate_mutex);
>  	}
> @@ -885,8 +928,132 @@ static ssize_t zonefs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
>  	return ret;
>  }
>  
> +static inline bool zonefs_file_use_exp_open(struct inode *inode, struct file *file)
> +{
> +	struct zonefs_inode_info *zi = ZONEFS_I(inode);
> +	struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
> +
> +	if (!(sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN))
> +		return false;
> +
> +	if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
> +		return false;
> +
> +	if (!(file->f_mode & FMODE_WRITE))
> +		return false;
> +
> +	return true;
> +}
> +
> +static int zonefs_open_zone(struct inode *inode)
> +{
> +	struct zonefs_inode_info *zi = ZONEFS_I(inode);
> +	struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
> +	int ret = 0;
> +
> +	mutex_lock(&zi->i_truncate_mutex);
> +
> +	zi->i_wr_refcnt++;
> +	if (zi->i_wr_refcnt == 1) {
> +
> +		if (atomic_inc_return(&sbi->s_open_zones) > sbi->s_max_open_zones) {
> +			atomic_dec(&sbi->s_open_zones);
> +			ret = -EBUSY;
> +			goto unlock;
> +		}
> +
> +		if (i_size_read(inode) < zi->i_max_size) {
> +			ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_OPEN);
> +			if (ret) {
> +				zi->i_wr_refcnt--;
> +				atomic_dec(&sbi->s_open_zones);
> +				goto unlock;
> +			}
> +			zi->i_flags |= ZONEFS_ZONE_OPEN;
> +		}
> +	}
> +
> +unlock:
> +	mutex_unlock(&zi->i_truncate_mutex);
> +
> +	return ret;
> +}
> +
> +static int zonefs_file_open(struct inode *inode, struct file *file)
> +{
> +	int ret;
> +
> +	ret = generic_file_open(inode, file);
> +	if (ret)
> +		return ret;
> +
> +	if (zonefs_file_use_exp_open(inode, file))
> +		return zonefs_open_zone(inode);
> +
> +	return 0;
> +}
> +
> +static void zonefs_close_zone(struct inode *inode)
> +{
> +	struct zonefs_inode_info *zi = ZONEFS_I(inode);
> +	int ret = 0;
> +
> +	mutex_lock(&zi->i_truncate_mutex);
> +	zi->i_wr_refcnt--;
> +	if (!zi->i_wr_refcnt) {
> +		struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
> +		struct super_block *sb = inode->i_sb;
> +
> +		/*
> +		 * If the file zone is full, it is not open anymore and we only
> +		 * need to decrement the open count.
> +		 */
> +		if (!(zi->i_flags & ZONEFS_ZONE_OPEN))
> +			goto dec;
> +
> +		ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_CLOSE);
> +		if (ret) {
> +			__zonefs_io_error(inode, false);
> +			/*
> +			 * If the zone open flag is still set, it means that
> +			 * the zone is neither read-only nor offline now. This
> +			 * can only happen if sending the REQ_OP_ZONE_CLOSE
> +			 * request to the drive failed, for whatever reasons.
> +			 *
> +			 * It's better to mark the FS read-only (if
> +			 * __zonefs_io_error hasn't done that already) than to
> +			 * loose any data.

I do not think it is about loosing data. If this type of error is repeated with
the open zones left as is, you can end up with only these zones being writable
and everything else read-only. So better take action and use the remount
read-only big hammer right away rather than ending up in a weird FS state. So
may be something like:

			/*
			 * Leaving zones explicitly open may lead to a state
			 * where most zones cannot be written (zone resources
			 * exhausted). So take preventive action by remounting
			 * read-only.
			 */

> +			 */
> +			if (zi->i_flags & ZONEFS_ZONE_OPEN &&
> +			    !(sb->s_flags & SB_RDONLY)) {
> +				zonefs_warn(sb, "closing zone failed, remounting filesystem read-only\n");
> +				sb->s_flags |= SB_RDONLY;
> +			}
> +		}
> +		zi->i_flags &= ~ZONEFS_ZONE_OPEN;
> +dec:
> +		atomic_dec(&sbi->s_open_zones);
> +	}
> +	mutex_unlock(&zi->i_truncate_mutex);
> +}
> +
> +static int zonefs_file_release(struct inode *inode, struct file *file)
> +{
> +	/*
> +	 * If we explicitly open a zone we must close it again as well, but the
> +	 * zone management operation can fail (either due to an IO error or as
> +	 * the zone has gone offline or read-only). Make sure we don't fail the
> +	 * close(2) for user-space.
> +	 */
> +	if (zonefs_file_use_exp_open(inode, file))
> +		zonefs_close_zone(inode);
> +
> +	return 0;
> +}
> +
>  static const struct file_operations zonefs_file_operations = {
> -	.open		= generic_file_open,
> +	.open		= zonefs_file_open,
> +	.release	= zonefs_file_release,
>  	.fsync		= zonefs_file_fsync,
>  	.mmap		= zonefs_file_mmap,
>  	.llseek		= zonefs_file_llseek,
> @@ -910,6 +1077,7 @@ static struct inode *zonefs_alloc_inode(struct super_block *sb)
>  	inode_init_once(&zi->i_vnode);
>  	mutex_init(&zi->i_truncate_mutex);
>  	init_rwsem(&zi->i_mmap_sem);
> +	zi->i_wr_refcnt = 0;
>  
>  	return &zi->i_vnode;
>  }
> @@ -960,7 +1128,7 @@ static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
>  
>  enum {
>  	Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
> -	Opt_err,
> +	Opt_explicit_open, Opt_err,
>  };
>  
>  static const match_table_t tokens = {
> @@ -968,6 +1136,7 @@ static const match_table_t tokens = {
>  	{ Opt_errors_zro,	"errors=zone-ro"},
>  	{ Opt_errors_zol,	"errors=zone-offline"},
>  	{ Opt_errors_repair,	"errors=repair"},
> +	{ Opt_explicit_open,	"explicit-open" },
>  	{ Opt_err,		NULL}
>  };
>  
> @@ -1004,6 +1173,9 @@ static int zonefs_parse_options(struct super_block *sb, char *options)
>  			sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
>  			sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
>  			break;
> +		case Opt_explicit_open:
> +			sbi->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN;
> +			break;
>  		default:
>  			return -EINVAL;
>  		}
> @@ -1423,6 +1595,13 @@ static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
>  	sbi->s_gid = GLOBAL_ROOT_GID;
>  	sbi->s_perm = 0640;
>  	sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
> +	sbi->s_max_open_zones = bdev_max_open_zones(sb->s_bdev);
> +	atomic_set(&sbi->s_open_zones, 0);
> +	if (!sbi->s_max_open_zones &&
> +	    sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
> +		zonefs_info(sb, "No open zones limit. Ignoring explicit_open mount option\n");
> +		sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN;
> +	}
>  
>  	ret = zonefs_read_super(sb);
>  	if (ret)
> diff --git a/fs/zonefs/zonefs.h b/fs/zonefs/zonefs.h
> index 55b39970acb2..51141907097c 100644
> --- a/fs/zonefs/zonefs.h
> +++ b/fs/zonefs/zonefs.h
> @@ -38,6 +38,8 @@ static inline enum zonefs_ztype zonefs_zone_type(struct blk_zone *zone)
>  	return ZONEFS_ZTYPE_SEQ;
>  }
>  
> +#define ZONEFS_ZONE_OPEN	(1 << 0)
> +
>  /*
>   * In-memory inode data.
>   */
> @@ -74,6 +76,10 @@ struct zonefs_inode_info {
>  	 */
>  	struct mutex		i_truncate_mutex;
>  	struct rw_semaphore	i_mmap_sem;
> +
> +	/* guarded by i_truncate_mutex */
> +	unsigned int		i_wr_refcnt;
> +	unsigned int		i_flags;
>  };
>  
>  static inline struct zonefs_inode_info *ZONEFS_I(struct inode *inode)
> @@ -154,6 +160,7 @@ enum zonefs_features {
>  #define ZONEFS_MNTOPT_ERRORS_MASK	\
>  	(ZONEFS_MNTOPT_ERRORS_RO | ZONEFS_MNTOPT_ERRORS_ZRO | \
>  	 ZONEFS_MNTOPT_ERRORS_ZOL | ZONEFS_MNTOPT_ERRORS_REPAIR)
> +#define ZONEFS_MNTOPT_EXPLICIT_OPEN	(1 << 4) /* Explicit open/close of zones on open/close */
>  
>  /*
>   * In-memory Super block information.
> @@ -175,6 +182,9 @@ struct zonefs_sb_info {
>  
>  	loff_t			s_blocks;
>  	loff_t			s_used_blocks;
> +
> +	unsigned int		s_max_open_zones;
> +	atomic_t		s_open_zones;
>  };
>  
>  static inline struct zonefs_sb_info *ZONEFS_SB(struct super_block *sb)
> 

With the comment fixed, I think this is good.

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH v4 3/4] zonefs: open/close zone on file open/close

[Johannes Thumshirn]

On 11/09/2020 07:11, Damien Le Moal wrote:
> I do not think it is about loosing data. If this type of error is repeated with
> the open zones left as is, you can end up with only these zones being writable
> and everything else read-only. So better take action and use the remount
> read-only big hammer right away rather than ending up in a weird FS state. So
> may be something like:
> 
> 			/*
> 			 * Leaving zones explicitly open may lead to a state
> 			 * where most zones cannot be written (zone resources
> 			 * exhausted). So take preventive action by remounting
> 			 * read-only.
> 			 */


Fixed

[PATCH v4 4/4] zonefs: document the explicit-open mount option

[Johannes Thumshirn]

Document the newly introduced explicit-open mount option.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
---
Changes to v1:
- Address Randy's comments
---
 Documentation/filesystems/zonefs.rst | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/Documentation/filesystems/zonefs.rst b/Documentation/filesystems/zonefs.rst
index 6c18bc8ce332..6b213fe9a33e 100644
--- a/Documentation/filesystems/zonefs.rst
+++ b/Documentation/filesystems/zonefs.rst
@@ -326,6 +326,21 @@ discover the amount of data that has been written to the zone. In the case of a
 read-only zone discovered at run-time, as indicated in the previous section.
 The size of the zone file is left unchanged from its last updated value.
 
+A zoned block device (e.g. an NVMe Zoned Namespace device) may have limits on
+the number of zones that can be active, that is, zones that are in the
+implicit open, explicit open or closed conditions.  This potential limitation
+translates into a risk for applications to see write IO errors due to this
+limit being exceeded if the zone of a file is not already active when a write
+request is issued by the user.
+
+To avoid these potential errors, the "explicit-open" mount option forces zones
+to be made active using an open zone command when a file is opened for writing
+for the first time. If the zone open command succeeds, the application is then
+guaranteed that write requests can be processed. Conversely, the
+"explicit-open" mount option will result in a zone close command being issued
+to the device on the last close() of a zone file if the zone is not full nor
+empty.
+
 Zonefs User Space Tools
 =======================
 
-- 
2.26.2