[v5,3/3] shmem: stable directory offsets

Commit Message

Chuck Lever June 27, 2023, 8:53 p.m. UTC

From: Chuck Lever <[e-mail]>

The current cursor-based directory offset mechanism doesn't work
when a tmpfs filesystem is exported via NFS. This is because NFS
clients do not open directories. Each server-side READDIR operation
has to open the directory, read it, then close it. The cursor state
for that directory, being associated strictly with the opened
struct file, is thus discarded after each NFS READDIR operation.

Directory offsets are cached not only by NFS clients, but also by
user space libraries on those clients. Essentially there is no way
to invalidate those caches when directory offsets have changed on
an NFS server after the offset-to-dentry mapping changes. Thus the
whole application stack depends on unchanging directory offsets.

The solution we've come up with is to make the directory offset for
each file in a tmpfs filesystem stable for the life of the directory
entry it represents.

shmem_readdir() and shmem_dir_llseek() now use an xarray to map each
directory offset (an loff_t integer) to the memory address of a
struct dentry.

Signed-off-by: Chuck Lever <[e-mail]>
---
 include/linux/shmem_fs.h |    1 +
 mm/shmem.c               |   47 +++++++++++++++++++++++++++++++++++++++-------
 2 files changed, 41 insertions(+), 7 deletions(-)

Patch

diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
index 9029abd29b1c..d2c67333028c 100644
--- a/include/linux/shmem_fs.h
+++ b/include/linux/shmem_fs.h
@@ -27,6 +27,7 @@  struct shmem_inode_info {
 	atomic_t		stop_eviction;	/* hold when working on inode */
 	struct timespec64	i_crtime;	/* file creation time */
 	unsigned int		fsflags;	/* flags for FS_IOC_[SG]ETFLAGS */
+	struct stable_offset_ctx dir_offsets;
 	struct inode		vfs_inode;
 };
 
diff --git a/mm/shmem.c b/mm/shmem.c
index 721f9fd064aa..5782fe1edc75 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -2355,6 +2355,11 @@  static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
 #define shmem_initxattrs NULL
 #endif
 
+static struct stable_offset_ctx *shmem_so_ctx(struct inode *inode)
+{
+	return &SHMEM_I(inode)->dir_offsets;
+}
+
 static struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb,
 				     struct inode *dir, umode_t mode, dev_t dev,
 				     unsigned long flags)
@@ -2410,7 +2415,8 @@  static struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block
 			/* Some things misbehave if size == 0 on a directory */
 			inode->i_size = 2 * BOGO_DIRENT_SIZE;
 			inode->i_op = &shmem_dir_inode_operations;
-			inode->i_fop = &simple_dir_operations;
+			inode->i_fop = &stable_dir_operations;
+			stable_offset_init(shmem_so_ctx(inode));
 			break;
 		case S_IFLNK:
 			/*
@@ -2950,7 +2956,10 @@  shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
 		if (error && error != -EOPNOTSUPP)
 			goto out_iput;
 
-		error = 0;
+		error = stable_offset_add(shmem_so_ctx(dir), dentry);
+		if (error)
+			goto out_iput;
+
 		dir->i_size += BOGO_DIRENT_SIZE;
 		dir->i_ctime = dir->i_mtime = current_time(dir);
 		inode_inc_iversion(dir);
@@ -3027,6 +3036,13 @@  static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentr
 			goto out;
 	}
 
+	ret = stable_offset_add(shmem_so_ctx(dir), dentry);
+	if (ret) {
+		if (inode->i_nlink)
+			shmem_free_inode(inode->i_sb);
+		goto out;
+	}
+
 	dir->i_size += BOGO_DIRENT_SIZE;
 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
 	inode_inc_iversion(dir);
@@ -3045,6 +3061,8 @@  static int shmem_unlink(struct inode *dir, struct dentry *dentry)
 	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
 		shmem_free_inode(inode->i_sb);
 
+	stable_offset_remove(shmem_so_ctx(dir), dentry);
+
 	dir->i_size -= BOGO_DIRENT_SIZE;
 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
 	inode_inc_iversion(dir);
@@ -3103,24 +3121,29 @@  static int shmem_rename2(struct mnt_idmap *idmap,
 {
 	struct inode *inode = d_inode(old_dentry);
 	int they_are_dirs = S_ISDIR(inode->i_mode);
+	int error;
 
 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
 		return -EINVAL;
 
 	if (flags & RENAME_EXCHANGE)
-		return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
+		return stable_offset_rename_exchange(old_dir, old_dentry,
+						     new_dir, new_dentry);
 
 	if (!simple_empty(new_dentry))
 		return -ENOTEMPTY;
 
 	if (flags & RENAME_WHITEOUT) {
-		int error;
-
 		error = shmem_whiteout(idmap, old_dir, old_dentry);
 		if (error)
 			return error;
 	}
 
+	stable_offset_remove(shmem_so_ctx(old_dir), old_dentry);
+	error = stable_offset_add(shmem_so_ctx(new_dir), old_dentry);
+	if (error)
+		return error;
+
 	if (d_really_is_positive(new_dentry)) {
 		(void) shmem_unlink(new_dir, new_dentry);
 		if (they_are_dirs) {
@@ -3164,19 +3187,23 @@  static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
 	if (error && error != -EOPNOTSUPP)
 		goto out_iput;
 
+	error = stable_offset_add(shmem_so_ctx(dir), dentry);
+	if (error)
+		goto out_iput;
+
 	inode->i_size = len-1;
 	if (len <= SHORT_SYMLINK_LEN) {
 		inode->i_link = kmemdup(symname, len, GFP_KERNEL);
 		if (!inode->i_link) {
 			error = -ENOMEM;
-			goto out_iput;
+			goto out_remove_offset;
 		}
 		inode->i_op = &shmem_short_symlink_operations;
 	} else {
 		inode_nohighmem(inode);
 		error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
 		if (error)
-			goto out_iput;
+			goto out_remove_offset;
 		inode->i_mapping->a_ops = &shmem_aops;
 		inode->i_op = &shmem_symlink_inode_operations;
 		memcpy(folio_address(folio), symname, len);
@@ -3191,6 +3218,9 @@  static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
 	d_instantiate(dentry, inode);
 	dget(dentry);
 	return 0;
+
+out_remove_offset:
+	stable_offset_remove(shmem_so_ctx(dir), dentry);
 out_iput:
 	iput(inode);
 	return error;
@@ -3920,6 +3950,8 @@  static void shmem_destroy_inode(struct inode *inode)
 {
 	if (S_ISREG(inode->i_mode))
 		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
+	if (S_ISDIR(inode->i_mode))
+		stable_offset_destroy(shmem_so_ctx(inode));
 }
 
 static void shmem_init_inode(void *foo)
@@ -4000,6 +4032,7 @@  static const struct inode_operations shmem_dir_inode_operations = {
 	.mknod		= shmem_mknod,
 	.rename		= shmem_rename2,
 	.tmpfile	= shmem_tmpfile,
+	.get_so_ctx	= shmem_so_ctx,
 #endif
 #ifdef CONFIG_TMPFS_XATTR
 	.listxattr	= shmem_listxattr,