Merge patch series "file: remove f_version"

Christian Brauner <brauner@kernel.org> says:

The f_version member in struct file isn't particularly well-defined. It
is mainly used as a cookie to detect concurrent seeks when iterating
directories. But it is also abused by some subsystems for completely
unrelated things.

It is mostly a directory specific thing that doesn't really need to live
in struct file and with its wonky semantics it really lacks a specific
function.

For pipes, f_version is (ab)used to defer poll notifications until a
write has happened. And struct pipe_inode_info is used by multiple
struct files in their ->private_data so there's no chance of pushing
that down into file->private_data without introducing another pointer
indirection.

But this should be a solvable problem. Only regular files with
FMODE_ATOMIC_POS and directories require f_pos_lock. Pipes and other
files don't. So this adds a union into struct file encompassing
f_pos_lock and a pipe specific f_pipe member that pipes can use. This
union of course can be extended to other file types and is similar to
what we do in struct inode already.

* patches from https://lore.kernel.org/r/20240830-vfs-file-f_version-v1-0-6d3e4816aa7b@kernel.org:
  fs: remove f_version
  pipe: use f_pipe
  fs: add f_pipe
  ubifs: store cookie in private data
  ufs: store cookie in private data
  udf: store cookie in private data
  proc: store cookie in private data
  ocfs2: store cookie in private data
  input: remove f_version abuse
  ext4: store cookie in private data
  ext2: store cookie in private data
  affs: store cookie in private data
  fs: add generic_llseek_cookie()
  fs: use must_set_pos()
  fs: add must_set_pos()
  fs: add vfs_setpos_cookie()
  s390: remove unused f_version
  ceph: remove unused f_version
  adi: remove unused f_version
  file: remove pointless comment

Link: https://lore.kernel.org/r/20240830-vfs-file-f_version-v1-0-6d3e4816aa7b@kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>

drivers/char/adi.c

@@ -196,7 +196,6 @@ static loff_t adi_llseek(struct file *file, loff_t offset, int whence)
 
 	if (offset != file->f_pos) {
 		file->f_pos = offset;
-		file->f_version = 0;
 		ret = offset;
 	}
 

drivers/input/input.c

@@ -1079,33 +1079,31 @@ static inline void input_wakeup_procfs_readers(void)
 	wake_up(&input_devices_poll_wait);
 }
 
+struct input_seq_state {
+	unsigned short pos;
+	bool mutex_acquired;
+	int input_devices_state;
+};
+
 static __poll_t input_proc_devices_poll(struct file *file, poll_table *wait)
 {
+	struct seq_file *seq = file->private_data;
+	struct input_seq_state *state = seq->private;
+
 	poll_wait(file, &input_devices_poll_wait, wait);
-	if (file->f_version != input_devices_state) {
+	if (state->input_devices_state != input_devices_state) {
-		file->f_version = input_devices_state;
+		state->input_devices_state = input_devices_state;
 		return EPOLLIN | EPOLLRDNORM;
 	}
 
 	return 0;
 }
 
-union input_seq_state {
-	struct {
-		unsigned short pos;
-		bool mutex_acquired;
-	};
-	void *p;
-};
-
 static void *input_devices_seq_start(struct seq_file *seq, loff_t *pos)
 {
-	union input_seq_state *state = (union input_seq_state *)&seq->private;
+	struct input_seq_state *state = seq->private;
 	int error;
 
-	/* We need to fit into seq->private pointer */
-	BUILD_BUG_ON(sizeof(union input_seq_state) != sizeof(seq->private));
-
 	error = mutex_lock_interruptible(&input_mutex);
 	if (error) {
 		state->mutex_acquired = false;
@@ -1124,7 +1122,7 @@ static void *input_devices_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 
 static void input_seq_stop(struct seq_file *seq, void *v)
 {
-	union input_seq_state *state = (union input_seq_state *)&seq->private;
+	struct input_seq_state *state = seq->private;
 
 	if (state->mutex_acquired)
 		mutex_unlock(&input_mutex);
@@ -1210,7 +1208,8 @@ static const struct seq_operations input_devices_seq_ops = {
 
 static int input_proc_devices_open(struct inode *inode, struct file *file)
 {
-	return seq_open(file, &input_devices_seq_ops);
+	return seq_open_private(file, &input_devices_seq_ops,
+				sizeof(struct input_seq_state));
 }
 
 static const struct proc_ops input_devices_proc_ops = {
@@ -1218,17 +1217,14 @@ static const struct proc_ops input_devices_proc_ops = {
 	.proc_poll	= input_proc_devices_poll,
 	.proc_read	= seq_read,
 	.proc_lseek	= seq_lseek,
-	.proc_release	= seq_release,
+	.proc_release	= seq_release_private,
 };
 
 static void *input_handlers_seq_start(struct seq_file *seq, loff_t *pos)
 {
-	union input_seq_state *state = (union input_seq_state *)&seq->private;
+	struct input_seq_state *state = seq->private;
 	int error;
 
-	/* We need to fit into seq->private pointer */
-	BUILD_BUG_ON(sizeof(union input_seq_state) != sizeof(seq->private));
-
 	error = mutex_lock_interruptible(&input_mutex);
 	if (error) {
 		state->mutex_acquired = false;
@@ -1243,7 +1239,7 @@ static void *input_handlers_seq_start(struct seq_file *seq, loff_t *pos)
 
 static void *input_handlers_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 {
-	union input_seq_state *state = (union input_seq_state *)&seq->private;
+	struct input_seq_state *state = seq->private;
 
 	state->pos = *pos + 1;
 	return seq_list_next(v, &input_handler_list, pos);
@@ -1252,7 +1248,7 @@ static void *input_handlers_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 static int input_handlers_seq_show(struct seq_file *seq, void *v)
 {
 	struct input_handler *handler = container_of(v, struct input_handler, node);
-	union input_seq_state *state = (union input_seq_state *)&seq->private;
+	struct input_seq_state *state = seq->private;
 
 	seq_printf(seq, "N: Number=%u Name=%s", state->pos, handler->name);
 	if (handler->filter)
@@ -1273,14 +1269,15 @@ static const struct seq_operations input_handlers_seq_ops = {
 
 static int input_proc_handlers_open(struct inode *inode, struct file *file)
 {
-	return seq_open(file, &input_handlers_seq_ops);
+	return seq_open_private(file, &input_handlers_seq_ops,
+				sizeof(struct input_seq_state));
 }
 
 static const struct proc_ops input_handlers_proc_ops = {
 	.proc_open	= input_proc_handlers_open,
 	.proc_read	= seq_read,
 	.proc_lseek	= seq_lseek,
-	.proc_release	= seq_release,
+	.proc_release	= seq_release_private,
 };
 
 static int __init input_proc_init(void)

drivers/s390/char/hmcdrv_dev.c

@@ -186,9 +186,6 @@ static loff_t hmcdrv_dev_seek(struct file *fp, loff_t pos, int whence)
 	if (pos < 0)
 		return -EINVAL;
 
-	if (fp->f_pos != pos)
-		++fp->f_version;
-
 	fp->f_pos = pos;
 	return pos;
 }

fs/affs/dir.c

@@ -17,13 +17,44 @@
 #include <linux/iversion.h>
 #include "affs.h"
 
+struct affs_dir_data {
+	unsigned long ino;
+	u64 cookie;
+};
+
 static int affs_readdir(struct file *, struct dir_context *);
 
+static loff_t affs_dir_llseek(struct file *file, loff_t offset, int whence)
+{
+	struct affs_dir_data *data = file->private_data;
+
+	return generic_llseek_cookie(file, offset, whence, &data->cookie);
+}
+
+static int affs_dir_open(struct inode *inode, struct file *file)
+{
+	struct affs_dir_data	*data;
+
+	data = kzalloc(sizeof(struct affs_dir_data), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+	file->private_data = data;
+	return 0;
+}
+
+static int affs_dir_release(struct inode *inode, struct file *file)
+{
+	kfree(file->private_data);
+	return 0;
+}
+
 const struct file_operations affs_dir_operations = {
+	.open		= affs_dir_open,
 	.read		= generic_read_dir,
-	.llseek		= generic_file_llseek,
+	.llseek		= affs_dir_llseek,
 	.iterate_shared	= affs_readdir,
 	.fsync		= affs_file_fsync,
+	.release	= affs_dir_release,
 };
 
 /*
@@ -45,6 +76,7 @@ static int
 affs_readdir(struct file *file, struct dir_context *ctx)
 {
 	struct inode		*inode = file_inode(file);
+	struct affs_dir_data	*data = file->private_data;
 	struct super_block	*sb = inode->i_sb;
 	struct buffer_head	*dir_bh = NULL;
 	struct buffer_head	*fh_bh = NULL;
@@ -59,7 +91,7 @@ affs_readdir(struct file *file, struct dir_context *ctx)
 	pr_debug("%s(ino=%lu,f_pos=%llx)\n", __func__, inode->i_ino, ctx->pos);
 
 	if (ctx->pos < 2) {
-		file->private_data = (void *)0;
+		data->ino = 0;
 		if (!dir_emit_dots(file, ctx))
 			return 0;
 	}
@@ -80,8 +112,8 @@ affs_readdir(struct file *file, struct dir_context *ctx)
 	/* If the directory hasn't changed since the last call to readdir(),
 	 * we can jump directly to where we left off.
 	 */
-	ino = (u32)(long)file->private_data;
+	ino = data->ino;
-	if (ino && inode_eq_iversion(inode, file->f_version)) {
+	if (ino && inode_eq_iversion(inode, data->cookie)) {
 		pr_debug("readdir() left off=%d\n", ino);
 		goto inside;
 	}
@@ -131,8 +163,8 @@ inside:
 		} while (ino);
 	}
 done:
-	file->f_version = inode_query_iversion(inode);
+	data->cookie = inode_query_iversion(inode);
-	file->private_data = (void *)(long)ino;
+	data->ino = ino;
 	affs_brelse(fh_bh);
 
 out_brelse_dir:

fs/ceph/dir.c

@@ -707,7 +707,6 @@ static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
 
 		if (offset != file->f_pos) {
 			file->f_pos = offset;
-			file->f_version = 0;
 			dfi->file_info.flags &= ~CEPH_F_ATEND;
 		}
 		retval = offset;

fs/ext2/dir.c

@@ -263,7 +263,7 @@ ext2_readdir(struct file *file, struct dir_context *ctx)
 	unsigned long n = pos >> PAGE_SHIFT;
 	unsigned long npages = dir_pages(inode);
 	unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
-	bool need_revalidate = !inode_eq_iversion(inode, file->f_version);
+	bool need_revalidate = !inode_eq_iversion(inode, *(u64 *)file->private_data);
 	bool has_filetype;
 
 	if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
@@ -290,7 +290,7 @@ ext2_readdir(struct file *file, struct dir_context *ctx)
 				offset = ext2_validate_entry(kaddr, offset, chunk_mask);
 				ctx->pos = (n<<PAGE_SHIFT) + offset;
 			}
-			file->f_version = inode_query_iversion(inode);
+			*(u64 *)file->private_data = inode_query_iversion(inode);
 			need_revalidate = false;
 		}
 		de = (ext2_dirent *)(kaddr+offset);
@@ -703,8 +703,30 @@ not_empty:
 	return 0;
 }
 
+static int ext2_dir_open(struct inode *inode, struct file *file)
+{
+	file->private_data = kzalloc(sizeof(u64), GFP_KERNEL);
+	if (!file->private_data)
+		return -ENOMEM;
+	return 0;
+}
+
+static int ext2_dir_release(struct inode *inode, struct file *file)
+{
+	kfree(file->private_data);
+	return 0;
+}
+
+static loff_t ext2_dir_llseek(struct file *file, loff_t offset, int whence)
+{
+	return generic_llseek_cookie(file, offset, whence,
+				     (u64 *)file->private_data);
+}
+
 const struct file_operations ext2_dir_operations = {
-	.llseek		= generic_file_llseek,
+	.open		= ext2_dir_open,
+	.release	= ext2_dir_release,
+	.llseek		= ext2_dir_llseek,
 	.read		= generic_read_dir,
 	.iterate_shared	= ext2_readdir,
 	.unlocked_ioctl = ext2_ioctl,

fs/ext4/dir.c

@@ -133,6 +133,7 @@ static int ext4_readdir(struct file *file, struct dir_context *ctx)
 	struct super_block *sb = inode->i_sb;
 	struct buffer_head *bh = NULL;
 	struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
+	struct dir_private_info *info = file->private_data;
 
 	err = fscrypt_prepare_readdir(inode);
 	if (err)
@@ -229,7 +230,7 @@ static int ext4_readdir(struct file *file, struct dir_context *ctx)
 		 * readdir(2), then we might be pointing to an invalid
 		 * dirent right now.  Scan from the start of the block
 		 * to make sure. */
-		if (!inode_eq_iversion(inode, file->f_version)) {
+		if (!inode_eq_iversion(inode, info->cookie)) {
 			for (i = 0; i < sb->s_blocksize && i < offset; ) {
 				de = (struct ext4_dir_entry_2 *)
 					(bh->b_data + i);
@@ -249,7 +250,7 @@ static int ext4_readdir(struct file *file, struct dir_context *ctx)
 			offset = i;
 			ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
 				| offset;
-			file->f_version = inode_query_iversion(inode);
+			info->cookie = inode_query_iversion(inode);
 		}
 
 		while (ctx->pos < inode->i_size
@@ -384,6 +385,7 @@ static inline loff_t ext4_get_htree_eof(struct file *filp)
 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
 {
 	struct inode *inode = file->f_mapping->host;
+	struct dir_private_info *info = file->private_data;
 	int dx_dir = is_dx_dir(inode);
 	loff_t ret, htree_max = ext4_get_htree_eof(file);
 
@@ -392,7 +394,7 @@ static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
 						    htree_max, htree_max);
 	else
 		ret = ext4_llseek(file, offset, whence);
-	file->f_version = inode_peek_iversion(inode) - 1;
+	info->cookie = inode_peek_iversion(inode) - 1;
 	return ret;
 }
 
@@ -429,18 +431,15 @@ static void free_rb_tree_fname(struct rb_root *root)
 	*root = RB_ROOT;
 }
 
-
+static void ext4_htree_init_dir_info(struct file *filp, loff_t pos)
-static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
-							   loff_t pos)
 {
-	struct dir_private_info *p;
+	struct dir_private_info *p = filp->private_data;
 
-	p = kzalloc(sizeof(*p), GFP_KERNEL);
+	if (is_dx_dir(file_inode(filp)) && !p->initialized) {
-	if (!p)
+		p->curr_hash = pos2maj_hash(filp, pos);
-		return NULL;
+		p->curr_minor_hash = pos2min_hash(filp, pos);
-	p->curr_hash = pos2maj_hash(filp, pos);
+		p->initialized = true;
-	p->curr_minor_hash = pos2min_hash(filp, pos);
+	}
-	return p;
 }
 
 void ext4_htree_free_dir_info(struct dir_private_info *p)
@@ -552,12 +551,7 @@ static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
 	struct fname *fname;
 	int ret = 0;
 
-	if (!info) {
+	ext4_htree_init_dir_info(file, ctx->pos);
-		info = ext4_htree_create_dir_info(file, ctx->pos);
-		if (!info)
-			return -ENOMEM;
-		file->private_data = info;
-	}
 
 	if (ctx->pos == ext4_get_htree_eof(file))
 		return 0;	/* EOF */
@@ -590,10 +584,10 @@ static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
 		 * cached entries.
 		 */
 		if ((!info->curr_node) ||
-		    !inode_eq_iversion(inode, file->f_version)) {
+		    !inode_eq_iversion(inode, info->cookie)) {
 			info->curr_node = NULL;
 			free_rb_tree_fname(&info->root);
-			file->f_version = inode_query_iversion(inode);
+			info->cookie = inode_query_iversion(inode);
 			ret = ext4_htree_fill_tree(file, info->curr_hash,
 						   info->curr_minor_hash,
 						   &info->next_hash);
@@ -664,7 +658,19 @@ int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
 	return 0;
 }
 
+static int ext4_dir_open(struct inode *inode, struct file *file)
+{
+	struct dir_private_info *info;
+
+	info = kzalloc(sizeof(*info), GFP_KERNEL);
+	if (!info)
+		return -ENOMEM;
+	file->private_data = info;
+	return 0;
+}
+
 const struct file_operations ext4_dir_operations = {
+	.open		= ext4_dir_open,
 	.llseek		= ext4_dir_llseek,
 	.read		= generic_read_dir,
 	.iterate_shared	= ext4_readdir,

fs/ext4/ext4.h

@@ -2553,6 +2553,8 @@ struct dir_private_info {
 	__u32		curr_hash;
 	__u32		curr_minor_hash;
 	__u32		next_hash;
+	u64		cookie;
+	bool		initialized;
 };
 
 /* calculate the first block number of the group */

fs/ext4/inline.c

@@ -1460,6 +1460,7 @@ int ext4_read_inline_dir(struct file *file,
 	struct ext4_iloc iloc;
 	void *dir_buf = NULL;
 	int dotdot_offset, dotdot_size, extra_offset, extra_size;
+	struct dir_private_info *info = file->private_data;
 
 	ret = ext4_get_inode_loc(inode, &iloc);
 	if (ret)
@@ -1503,12 +1504,12 @@ int ext4_read_inline_dir(struct file *file,
 	extra_size = extra_offset + inline_size;
 
 	/*
-	 * If the version has changed since the last call to
+	 * If the cookie has changed since the last call to
 	 * readdir(2), then we might be pointing to an invalid
 	 * dirent right now.  Scan from the start of the inline
 	 * dir to make sure.
 	 */
-	if (!inode_eq_iversion(inode, file->f_version)) {
+	if (!inode_eq_iversion(inode, info->cookie)) {
 		for (i = 0; i < extra_size && i < offset;) {
 			/*
 			 * "." is with offset 0 and
@@ -1540,7 +1541,7 @@ int ext4_read_inline_dir(struct file *file,
 		}
 		offset = i;
 		ctx->pos = offset;
-		file->f_version = inode_query_iversion(inode);
+		info->cookie = inode_query_iversion(inode);
 	}
 
 	while (ctx->pos < extra_size) {

fs/file_table.c

@@ -156,6 +156,13 @@ static int init_file(struct file *f, int flags, const struct cred *cred)
 	}
 
 	spin_lock_init(&f->f_lock);
+	/*
+	 * Note that f_pos_lock is only used for files raising
+	 * FMODE_ATOMIC_POS and directories. Other files such as pipes
+	 * don't need it and since f_pos_lock is in a union may reuse
+	 * the space for other purposes. They are expected to initialize
+	 * the respective member when opening the file.
+	 */
 	mutex_init(&f->f_pos_lock);
 	f->f_flags = flags;
 	f->f_mode = OPEN_FMODE(flags);

fs/ocfs2/dir.c

@@ -1932,6 +1932,7 @@ int ocfs2_readdir(struct file *file, struct dir_context *ctx)
 {
 	int error = 0;
 	struct inode *inode = file_inode(file);
+	struct ocfs2_file_private *fp = file->private_data;
 	int lock_level = 0;
 
 	trace_ocfs2_readdir((unsigned long long)OCFS2_I(inode)->ip_blkno);
@@ -1952,7 +1953,7 @@ int ocfs2_readdir(struct file *file, struct dir_context *ctx)
 		goto bail_nolock;
 	}
 
-	error = ocfs2_dir_foreach_blk(inode, &file->f_version, ctx, false);
+	error = ocfs2_dir_foreach_blk(inode, &fp->cookie, ctx, false);
 
 	ocfs2_inode_unlock(inode, lock_level);
 	if (error)

fs/ocfs2/file.c

@@ -2750,6 +2750,13 @@ out_unlock:
 	return remapped > 0 ? remapped : ret;
 }
 
+static loff_t ocfs2_dir_llseek(struct file *file, loff_t offset, int whence)
+{
+	struct ocfs2_file_private *fp = file->private_data;
+
+	return generic_llseek_cookie(file, offset, whence, &fp->cookie);
+}
+
 const struct inode_operations ocfs2_file_iops = {
 	.setattr	= ocfs2_setattr,
 	.getattr	= ocfs2_getattr,
@@ -2797,7 +2804,7 @@ const struct file_operations ocfs2_fops = {
 
 WRAP_DIR_ITER(ocfs2_readdir) // FIXME!
 const struct file_operations ocfs2_dops = {
-	.llseek		= generic_file_llseek,
+	.llseek		= ocfs2_dir_llseek,
 	.read		= generic_read_dir,
 	.iterate_shared	= shared_ocfs2_readdir,
 	.fsync		= ocfs2_sync_file,
@@ -2843,7 +2850,7 @@ const struct file_operations ocfs2_fops_no_plocks = {
 };
 
 const struct file_operations ocfs2_dops_no_plocks = {
-	.llseek		= generic_file_llseek,
+	.llseek		= ocfs2_dir_llseek,
 	.read		= generic_read_dir,
 	.iterate_shared	= shared_ocfs2_readdir,
 	.fsync		= ocfs2_sync_file,

fs/ocfs2/file.h

@@ -20,6 +20,7 @@ struct ocfs2_alloc_context;
 enum ocfs2_alloc_restarted;
 
 struct ocfs2_file_private {
+	u64			cookie;
 	struct file		*fp_file;
 	struct mutex		fp_mutex;
 	struct ocfs2_lock_res	fp_flock;

fs/pipe.c

@@ -686,7 +686,7 @@ pipe_poll(struct file *filp, poll_table *wait)
 	if (filp->f_mode & FMODE_READ) {
 		if (!pipe_empty(head, tail))
 			mask |= EPOLLIN | EPOLLRDNORM;
-		if (!pipe->writers && filp->f_version != pipe->w_counter)
+		if (!pipe->writers && filp->f_pipe != pipe->w_counter)
 			mask |= EPOLLHUP;
 	}
 
@@ -945,6 +945,7 @@ int create_pipe_files(struct file **res, int flags)
 	}
 
 	f->private_data = inode->i_pipe;
+	f->f_pipe = 0;
 
 	res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK),
 				  &pipefifo_fops);
@@ -954,6 +955,7 @@ int create_pipe_files(struct file **res, int flags)
 		return PTR_ERR(res[0]);
 	}
 	res[0]->private_data = inode->i_pipe;
+	res[0]->f_pipe = 0;
 	res[1] = f;
 	stream_open(inode, res[0]);
 	stream_open(inode, res[1]);
@@ -1108,7 +1110,7 @@ static int fifo_open(struct inode *inode, struct file *filp)
 	bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
 	int ret;
 
-	filp->f_version = 0;
+	filp->f_pipe = 0;
 
 	spin_lock(&inode->i_lock);
 	if (inode->i_pipe) {
@@ -1155,7 +1157,7 @@ static int fifo_open(struct inode *inode, struct file *filp)
 			if ((filp->f_flags & O_NONBLOCK)) {
 				/* suppress EPOLLHUP until we have
 				 * seen a writer */
-				filp->f_version = pipe->w_counter;
+				filp->f_pipe = pipe->w_counter;
 			} else {
 				if (wait_for_partner(pipe, &pipe->w_counter))
 					goto err_rd;

fs/proc/base.c

@@ -3870,12 +3870,12 @@ static int proc_task_readdir(struct file *file, struct dir_context *ctx)
 	if (!dir_emit_dots(file, ctx))
 		return 0;
 
-	/* f_version caches the tgid value that the last readdir call couldn't
+	/* We cache the tgid value that the last readdir call couldn't
-	 * return. lseek aka telldir automagically resets f_version to 0.
+	 * return and lseek resets it to 0.
 	 */
 	ns = proc_pid_ns(inode->i_sb);
-	tid = (int)file->f_version;
+	tid = (int)(intptr_t)file->private_data;
-	file->f_version = 0;
+	file->private_data = NULL;
 	for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
 	     task;
 	     task = next_tid(task), ctx->pos++) {
@@ -3890,7 +3890,7 @@ static int proc_task_readdir(struct file *file, struct dir_context *ctx)
 				proc_task_instantiate, task, NULL)) {
 			/* returning this tgid failed, save it as the first
 			 * pid for the next readir call */
-			file->f_version = (u64)tid;
+			file->private_data = (void *)(intptr_t)tid;
 			put_task_struct(task);
 			break;
 		}
@@ -3915,6 +3915,24 @@ static int proc_task_getattr(struct mnt_idmap *idmap,
 	return 0;
 }
 
+/*
+ * proc_task_readdir() set @file->private_data to a positive integer
+ * value, so casting that to u64 is safe. generic_llseek_cookie() will
+ * set @cookie to 0, so casting to an int is safe. The WARN_ON_ONCE() is
+ * here to catch any unexpected change in behavior either in
+ * proc_task_readdir() or generic_llseek_cookie().
+ */
+static loff_t proc_dir_llseek(struct file *file, loff_t offset, int whence)
+{
+	u64 cookie = (u64)(intptr_t)file->private_data;
+	loff_t off;
+
+	off = generic_llseek_cookie(file, offset, whence, &cookie);
+	WARN_ON_ONCE(cookie > INT_MAX);
+	file->private_data = (void *)(intptr_t)cookie; /* serialized by f_pos_lock */
+	return off;
+}
+
 static const struct inode_operations proc_task_inode_operations = {
 	.lookup		= proc_task_lookup,
 	.getattr	= proc_task_getattr,
@@ -3925,7 +3943,7 @@ static const struct inode_operations proc_task_inode_operations = {
 static const struct file_operations proc_task_operations = {
 	.read		= generic_read_dir,
 	.iterate_shared	= proc_task_readdir,
-	.llseek		= generic_file_llseek,
+	.llseek		= proc_dir_llseek,
 };
 
 void __init set_proc_pid_nlink(void)

fs/read_write.c

@@ -39,6 +39,35 @@ static inline bool unsigned_offsets(struct file *file)
 	return file->f_mode & FMODE_UNSIGNED_OFFSET;
 }
 
+/**
+ * vfs_setpos_cookie - update the file offset for lseek and reset cookie
+ * @file:	file structure in question
+ * @offset:	file offset to seek to
+ * @maxsize:	maximum file size
+ * @cookie:	cookie to reset
+ *
+ * Update the file offset to the value specified by @offset if the given
+ * offset is valid and it is not equal to the current file offset and
+ * reset the specified cookie to indicate that a seek happened.
+ *
+ * Return the specified offset on success and -EINVAL on invalid offset.
+ */
+static loff_t vfs_setpos_cookie(struct file *file, loff_t offset,
+				loff_t maxsize, u64 *cookie)
+{
+	if (offset < 0 && !unsigned_offsets(file))
+		return -EINVAL;
+	if (offset > maxsize)
+		return -EINVAL;
+
+	if (offset != file->f_pos) {
+		file->f_pos = offset;
+		if (cookie)
+			*cookie = 0;
+	}
+	return offset;
+}
+
 /**
  * vfs_setpos - update the file offset for lseek
  * @file:	file structure in question
@@ -53,19 +82,63 @@ static inline bool unsigned_offsets(struct file *file)
  */
 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
 {
-	if (offset < 0 && !unsigned_offsets(file))
+	return vfs_setpos_cookie(file, offset, maxsize, NULL);
-		return -EINVAL;
-	if (offset > maxsize)
-		return -EINVAL;
-
-	if (offset != file->f_pos) {
-		file->f_pos = offset;
-		file->f_version = 0;
-	}
-	return offset;
 }
 EXPORT_SYMBOL(vfs_setpos);
 
+/**
+ * must_set_pos - check whether f_pos has to be updated
+ * @file: file to seek on
+ * @offset: offset to use
+ * @whence: type of seek operation
+ * @eof: end of file
+ *
+ * Check whether f_pos needs to be updated and update @offset according
+ * to @whence.
+ *
+ * Return: 0 if f_pos doesn't need to be updated, 1 if f_pos has to be
+ * updated, and negative error code on failure.
+ */
+static int must_set_pos(struct file *file, loff_t *offset, int whence, loff_t eof)
+{
+	switch (whence) {
+	case SEEK_END:
+		*offset += eof;
+		break;
+	case SEEK_CUR:
+		/*
+		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
+		 * position-querying operation.  Avoid rewriting the "same"
+		 * f_pos value back to the file because a concurrent read(),
+		 * write() or lseek() might have altered it
+		 */
+		if (*offset == 0) {
+			*offset = file->f_pos;
+			return 0;
+		}
+		break;
+	case SEEK_DATA:
+		/*
+		 * In the generic case the entire file is data, so as long as
+		 * offset isn't at the end of the file then the offset is data.
+		 */
+		if ((unsigned long long)*offset >= eof)
+			return -ENXIO;
+		break;
+	case SEEK_HOLE:
+		/*
+		 * There is a virtual hole at the end of the file, so as long as
+		 * offset isn't i_size or larger, return i_size.
+		 */
+		if ((unsigned long long)*offset >= eof)
+			return -ENXIO;
+		*offset = eof;
+		break;
+	}
+
+	return 1;
+}
+
 /**
  * generic_file_llseek_size - generic llseek implementation for regular files
  * @file:	file structure to seek on
@@ -86,51 +159,73 @@ loff_t
 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
 		loff_t maxsize, loff_t eof)
 {
-	switch (whence) {
+	int ret;
-	case SEEK_END:
+
-		offset += eof;
+	ret = must_set_pos(file, &offset, whence, eof);
-		break;
+	if (ret < 0)
-	case SEEK_CUR:
+		return ret;
-		/*
+	if (ret == 0)
-		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
-		 * position-querying operation.  Avoid rewriting the "same"
-		 * f_pos value back to the file because a concurrent read(),
-		 * write() or lseek() might have altered it
-		 */
-		if (offset == 0)
-			return file->f_pos;
-		/*
-		 * f_lock protects against read/modify/write race with other
-		 * SEEK_CURs. Note that parallel writes and reads behave
-		 * like SEEK_SET.
-		 */
-		spin_lock(&file->f_lock);
-		offset = vfs_setpos(file, file->f_pos + offset, maxsize);
-		spin_unlock(&file->f_lock);
 		return offset;
-	case SEEK_DATA:
+
+	if (whence == SEEK_CUR) {
 		/*
-		 * In the generic case the entire file is data, so as long as
+		 * f_lock protects against read/modify/write race with
-		 * offset isn't at the end of the file then the offset is data.
+		 * other SEEK_CURs. Note that parallel writes and reads
+		 * behave like SEEK_SET.
 		 */
-		if ((unsigned long long)offset >= eof)
+		guard(spinlock)(&file->f_lock);
-			return -ENXIO;
+		return vfs_setpos(file, file->f_pos + offset, maxsize);
-		break;
-	case SEEK_HOLE:
-		/*
-		 * There is a virtual hole at the end of the file, so as long as
-		 * offset isn't i_size or larger, return i_size.
-		 */
-		if ((unsigned long long)offset >= eof)
-			return -ENXIO;
-		offset = eof;
-		break;
 	}
 
 	return vfs_setpos(file, offset, maxsize);
 }
 EXPORT_SYMBOL(generic_file_llseek_size);
 
+/**
+ * generic_llseek_cookie - versioned llseek implementation
+ * @file:	file structure to seek on
+ * @offset:	file offset to seek to
+ * @whence:	type of seek
+ * @cookie:	cookie to update
+ *
+ * See generic_file_llseek for a general description and locking assumptions.
+ *
+ * In contrast to generic_file_llseek, this function also resets a
+ * specified cookie to indicate a seek took place.
+ */
+loff_t generic_llseek_cookie(struct file *file, loff_t offset, int whence,
+			     u64 *cookie)
+{
+	struct inode *inode = file->f_mapping->host;
+	loff_t maxsize = inode->i_sb->s_maxbytes;
+	loff_t eof = i_size_read(inode);
+	int ret;
+
+	if (WARN_ON_ONCE(!cookie))
+		return -EINVAL;
+
+	/*
+	 * Require that this is only used for directories that guarantee
+	 * synchronization between readdir and seek so that an update to
+	 * @cookie is correctly synchronized with concurrent readdir.
+	 */
+	if (WARN_ON_ONCE(!(file->f_mode & FMODE_ATOMIC_POS)))
+		return -EINVAL;
+
+	ret = must_set_pos(file, &offset, whence, eof);
+	if (ret < 0)
+		return ret;
+	if (ret == 0)
+		return offset;
+
+	/* No need to hold f_lock because we know that f_pos_lock is held. */
+	if (whence == SEEK_CUR)
+		return vfs_setpos_cookie(file, file->f_pos + offset, maxsize, cookie);
+
+	return vfs_setpos_cookie(file, offset, maxsize, cookie);
+}
+EXPORT_SYMBOL(generic_llseek_cookie);
+
 /**
  * generic_file_llseek - generic llseek implementation for regular files
  * @file:	file structure to seek on
@@ -270,10 +365,8 @@ loff_t default_llseek(struct file *file, loff_t offset, int whence)
 	}
 	retval = -EINVAL;
 	if (offset >= 0 || unsigned_offsets(file)) {
-		if (offset != file->f_pos) {
+		if (offset != file->f_pos)
 			file->f_pos = offset;
-			file->f_version = 0;
-		}
 		retval = offset;
 	}
 out:

fs/ubifs/dir.c

@@ -555,6 +555,11 @@ static unsigned int vfs_dent_type(uint8_t type)
 	return 0;
 }
 
+struct ubifs_dir_data {
+	struct ubifs_dent_node *dent;
+	u64 cookie;
+};
+
 /*
  * The classical Unix view for directory is that it is a linear array of
  * (name, inode number) entries. Linux/VFS assumes this model as well.
@@ -582,6 +587,7 @@ static int ubifs_readdir(struct file *file, struct dir_context *ctx)
 	struct inode *dir = file_inode(file);
 	struct ubifs_info *c = dir->i_sb->s_fs_info;
 	bool encrypted = IS_ENCRYPTED(dir);
+	struct ubifs_dir_data *data = file->private_data;
 
 	dbg_gen("dir ino %lu, f_pos %#llx", dir->i_ino, ctx->pos);
 
@@ -604,27 +610,27 @@ static int ubifs_readdir(struct file *file, struct dir_context *ctx)
 		fstr_real_len = fstr.len;
 	}
 
-	if (file->f_version == 0) {
+	if (data->cookie == 0) {
 		/*
-		 * The file was seek'ed, which means that @file->private_data
+		 * The file was seek'ed, which means that @data->dent
 		 * is now invalid. This may also be just the first
 		 * 'ubifs_readdir()' invocation, in which case
-		 * @file->private_data is NULL, and the below code is
+		 * @data->dent is NULL, and the below code is
 		 * basically a no-op.
 		 */
-		kfree(file->private_data);
+		kfree(data->dent);
-		file->private_data = NULL;
+		data->dent = NULL;
 	}
 
 	/*
-	 * 'generic_file_llseek()' unconditionally sets @file->f_version to
+	 * 'ubifs_dir_llseek()' sets @data->cookie to zero, and we use this
-	 * zero, and we use this for detecting whether the file was seek'ed.
+	 * for detecting whether the file was seek'ed.
 	 */
-	file->f_version = 1;
+	data->cookie = 1;
 
 	/* File positions 0 and 1 correspond to "." and ".." */
 	if (ctx->pos < 2) {
-		ubifs_assert(c, !file->private_data);
+		ubifs_assert(c, !data->dent);
 		if (!dir_emit_dots(file, ctx)) {
 			if (encrypted)
 				fscrypt_fname_free_buffer(&fstr);
@@ -641,10 +647,10 @@ static int ubifs_readdir(struct file *file, struct dir_context *ctx)
 		}
 
 		ctx->pos = key_hash_flash(c, &dent->key);
-		file->private_data = dent;
+		data->dent = dent;
 	}
 
-	dent = file->private_data;
+	dent = data->dent;
 	if (!dent) {
 		/*
 		 * The directory was seek'ed to and is now readdir'ed.
@@ -658,7 +664,7 @@ static int ubifs_readdir(struct file *file, struct dir_context *ctx)
 			goto out;
 		}
 		ctx->pos = key_hash_flash(c, &dent->key);
-		file->private_data = dent;
+		data->dent = dent;
 	}
 
 	while (1) {
@@ -701,15 +707,15 @@ static int ubifs_readdir(struct file *file, struct dir_context *ctx)
 			goto out;
 		}
 
-		kfree(file->private_data);
+		kfree(data->dent);
 		ctx->pos = key_hash_flash(c, &dent->key);
-		file->private_data = dent;
+		data->dent = dent;
 		cond_resched();
 	}
 
 out:
-	kfree(file->private_data);
+	kfree(data->dent);
-	file->private_data = NULL;
+	data->dent = NULL;
 
 	if (encrypted)
 		fscrypt_fname_free_buffer(&fstr);
@@ -733,7 +739,10 @@ out:
 /* Free saved readdir() state when the directory is closed */
 static int ubifs_dir_release(struct inode *dir, struct file *file)
 {
-	kfree(file->private_data);
+	struct ubifs_dir_data *data = file->private_data;
+
+	kfree(data->dent);
+	kfree(data);
 	file->private_data = NULL;
 	return 0;
 }
@@ -1712,6 +1721,24 @@ int ubifs_getattr(struct mnt_idmap *idmap, const struct path *path,
 	return 0;
 }
 
+static int ubifs_dir_open(struct inode *inode, struct file *file)
+{
+	struct ubifs_dir_data *data;
+
+	data = kzalloc(sizeof(struct ubifs_dir_data), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+	file->private_data = data;
+	return 0;
+}
+
+static loff_t ubifs_dir_llseek(struct file *file, loff_t offset, int whence)
+{
+	struct ubifs_dir_data *data = file->private_data;
+
+	return generic_llseek_cookie(file, offset, whence, &data->cookie);
+}
+
 const struct inode_operations ubifs_dir_inode_operations = {
 	.lookup      = ubifs_lookup,
 	.create      = ubifs_create,
@@ -1732,7 +1759,8 @@ const struct inode_operations ubifs_dir_inode_operations = {
 };
 
 const struct file_operations ubifs_dir_operations = {
-	.llseek         = generic_file_llseek,
+	.open		= ubifs_dir_open,
+	.llseek         = ubifs_dir_llseek,
 	.release        = ubifs_dir_release,
 	.read           = generic_read_dir,
 	.iterate_shared = ubifs_readdir,

fs/udf/dir.c

@@ -60,7 +60,7 @@ static int udf_readdir(struct file *file, struct dir_context *ctx)
 	 * identifying beginning of dir entry (names are under user control),
 	 * we need to scan the directory from the beginning.
 	 */
-	if (!inode_eq_iversion(dir, file->f_version)) {
+	if (!inode_eq_iversion(dir, *(u64 *)file->private_data)) {
 		emit_pos = nf_pos;
 		nf_pos = 0;
 	} else {
@@ -122,15 +122,37 @@ out_iter:
 	udf_fiiter_release(&iter);
 out:
 	if (pos_valid)
-		file->f_version = inode_query_iversion(dir);
+		*(u64 *)file->private_data = inode_query_iversion(dir);
 	kfree(fname);
 
 	return ret;
 }
 
+static int udf_dir_open(struct inode *inode, struct file *file)
+{
+	file->private_data = kzalloc(sizeof(u64), GFP_KERNEL);
+	if (!file->private_data)
+		return -ENOMEM;
+	return 0;
+}
+
+static int udf_dir_release(struct inode *inode, struct file *file)
+{
+	kfree(file->private_data);
+	return 0;
+}
+
+static loff_t udf_dir_llseek(struct file *file, loff_t offset, int whence)
+{
+	return generic_llseek_cookie(file, offset, whence,
+				     (u64 *)file->private_data);
+}
+
 /* readdir and lookup functions */
 const struct file_operations udf_dir_operations = {
-	.llseek			= generic_file_llseek,
+	.open			= udf_dir_open,
+	.release		= udf_dir_release,
+	.llseek			= udf_dir_llseek,
 	.read			= generic_read_dir,
 	.iterate_shared		= udf_readdir,
 	.unlocked_ioctl		= udf_ioctl,

fs/ufs/dir.c

@@ -435,7 +435,7 @@ ufs_readdir(struct file *file, struct dir_context *ctx)
 	unsigned long n = pos >> PAGE_SHIFT;
 	unsigned long npages = dir_pages(inode);
 	unsigned chunk_mask = ~(UFS_SB(sb)->s_uspi->s_dirblksize - 1);
-	bool need_revalidate = !inode_eq_iversion(inode, file->f_version);
+	bool need_revalidate = !inode_eq_iversion(inode, *(u64 *)file->private_data);
 	unsigned flags = UFS_SB(sb)->s_flags;
 
 	UFSD("BEGIN\n");
@@ -462,7 +462,7 @@ ufs_readdir(struct file *file, struct dir_context *ctx)
 				offset = ufs_validate_entry(sb, kaddr, offset, chunk_mask);
 				ctx->pos = (n<<PAGE_SHIFT) + offset;
 			}
-			file->f_version = inode_query_iversion(inode);
+			*(u64 *)file->private_data = inode_query_iversion(inode);
 			need_revalidate = false;
 		}
 		de = (struct ufs_dir_entry *)(kaddr+offset);
@@ -646,9 +646,31 @@ not_empty:
 	return 0;
 }
 
+static int ufs_dir_open(struct inode *inode, struct file *file)
+{
+	file->private_data = kzalloc(sizeof(u64), GFP_KERNEL);
+	if (!file->private_data)
+		return -ENOMEM;
+	return 0;
+}
+
+static int ufs_dir_release(struct inode *inode, struct file *file)
+{
+	kfree(file->private_data);
+	return 0;
+}
+
+static loff_t ufs_dir_llseek(struct file *file, loff_t offset, int whence)
+{
+	return generic_llseek_cookie(file, offset, whence,
+				     (u64 *)file->private_data);
+}
+
 const struct file_operations ufs_dir_operations = {
+	.open		= ufs_dir_open,
+	.release	= ufs_dir_release,
 	.read		= generic_read_dir,
 	.iterate_shared	= ufs_readdir,
 	.fsync		= generic_file_fsync,
-	.llseek		= generic_file_llseek,
+	.llseek		= ufs_dir_llseek,
 };

include/linux/fs.h

@@ -1001,8 +1001,8 @@ static inline int ra_has_index(struct file_ra_state *ra, pgoff_t index)
  * @f_cred: stashed credentials of creator/opener
  * @f_path: path of the file
  * @f_pos_lock: lock protecting file position
+ * @f_pipe: specific to pipes
  * @f_pos: file position
- * @f_version: file version
  * @f_security: LSM security context of this file
  * @f_owner: file owner
  * @f_wb_err: writeback error
@@ -1026,13 +1026,17 @@ struct file {
 	const struct cred		*f_cred;
 	/* --- cacheline 1 boundary (64 bytes) --- */
 	struct path			f_path;
-	struct mutex			f_pos_lock;
+	union {
+		/* regular files (with FMODE_ATOMIC_POS) and directories */
+		struct mutex		f_pos_lock;
+		/* pipes */
+		u64			f_pipe;
+	};
 	loff_t				f_pos;
-	u64				f_version;
-	/* --- cacheline 2 boundary (128 bytes) --- */
 #ifdef CONFIG_SECURITY
 	void				*f_security;
 #endif
+	/* --- cacheline 2 boundary (128 bytes) --- */
 	struct fown_struct		*f_owner;
 	errseq_t			f_wb_err;
 	errseq_t			f_sb_err;
@@ -3202,6 +3206,8 @@ extern loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize);
 extern loff_t generic_file_llseek(struct file *file, loff_t offset, int whence);
 extern loff_t generic_file_llseek_size(struct file *file, loff_t offset,
 		int whence, loff_t maxsize, loff_t eof);
+loff_t generic_llseek_cookie(struct file *file, loff_t offset, int whence,
+			     u64 *cookie);
 extern loff_t fixed_size_llseek(struct file *file, loff_t offset,
 		int whence, loff_t size);
 extern loff_t no_seek_end_llseek_size(struct file *, loff_t, int, loff_t);