shmem: stable directory offsets

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 <chuck.lever@oracle.com>
Message-Id: <168814734331.530310.3911190551060453102.stgit@manet.1015granger.net>
Signed-off-by: Christian Brauner <brauner@kernel.org>

1 files changed, 40 insertions, 7 deletions

diff --git a/mm/shmem.c b/mm/shmem.c
index f5006ea3b8e1..81973914b294 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -2427,6 +2427,11 @@ static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
 #define shmem_initxattrs NULL
 #endif
 
+static struct offset_ctx *shmem_get_offset_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,
@@ -2492,7 +2497,8 @@ static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
 		/* 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 = &simple_offset_dir_operations;
+		simple_offset_init(shmem_get_offset_ctx(inode));
 		break;
 	case S_IFLNK:
 		/*
@@ -3204,7 +3210,10 @@ shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
 	if (error && error != -EOPNOTSUPP)
 		goto out_iput;
 
-	error = 0;
+	error = simple_offset_add(shmem_get_offset_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);
@@ -3287,6 +3296,13 @@ static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentr
 			goto out;
 	}
 
+	ret = simple_offset_add(shmem_get_offset_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);
@@ -3305,6 +3321,8 @@ 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);
 
+	simple_offset_remove(shmem_get_offset_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);
@@ -3363,24 +3381,29 @@ 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 simple_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;
 	}
 
+	simple_offset_remove(shmem_get_offset_ctx(old_dir), old_dentry);
+	error = simple_offset_add(shmem_get_offset_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) {
@@ -3425,19 +3448,23 @@ static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
 	if (error && error != -EOPNOTSUPP)
 		goto out_iput;
 
+	error = simple_offset_add(shmem_get_offset_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);
@@ -3452,6 +3479,9 @@ static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
 	d_instantiate(dentry, inode);
 	dget(dentry);
 	return 0;
+
+out_remove_offset:
+	simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
 out_iput:
 	iput(inode);
 	return error;
@@ -4295,6 +4325,8 @@ 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))
+		simple_offset_destroy(shmem_get_offset_ctx(inode));
 }
 
 static void shmem_init_inode(void *foo)
@@ -4375,6 +4407,7 @@ static const struct inode_operations shmem_dir_inode_operations = {
 	.mknod		= shmem_mknod,
 	.rename		= shmem_rename2,
 	.tmpfile	= shmem_tmpfile,
+	.get_offset_ctx	= shmem_get_offset_ctx,
 #endif
 #ifdef CONFIG_TMPFS_XATTR
 	.listxattr	= shmem_listxattr,