Merge patch series "symlink length caching"

Mateusz Guzik <[email protected]> says:

quote:
    When utilized it dodges strlen() in vfs_readlink(), giving about 1.5%
    speed up when issuing readlink on /initrd.img on ext4.

The size is stored in a union with i_devices, which is never looked at
unless the inode is for a device.

ext4 and tmpfs are patched, other filesystems can also get there with
some more work.

benchmark:
plug into will-it-scale into tests/readlink1.c:

char *testcase_description = "readlink /initrd.img";

void testcase(unsigned long long *iterations, unsigned long nr)
{
        char *tmplink = "/initrd.img";
        char buf[1024];

        while (1) {
                int error = readlink(tmplink, buf, sizeof(buf));
                assert(error > 0);

                (*iterations)++;
        }
}

* patches from https://lore.kernel.org/r/[email protected]:
  tmpfs: use inode_set_cached_link()
  ext4: use inode_set_cached_link()
  vfs: support caching symlink lengths in inodes

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Christian Brauner <[email protected]>

Author: brauner

fs/ext4/inode.c

@@ -5006,10 +5006,11 @@ struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
 		if (IS_ENCRYPTED(inode)) {
 			inode->i_op = &ext4_encrypted_symlink_inode_operations;
 		} else if (ext4_inode_is_fast_symlink(inode)) {
-			inode->i_link = (char *)ei->i_data;
 			inode->i_op = &ext4_fast_symlink_inode_operations;
 			nd_terminate_link(ei->i_data, inode->i_size,
 				sizeof(ei->i_data) - 1);
+			inode_set_cached_link(inode, (char *)ei->i_data,
+					      inode->i_size);
 		} else {
 			inode->i_op = &ext4_symlink_inode_operations;
 		}

fs/ext4/namei.c

@@ -3418,7 +3418,6 @@ static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
 			inode->i_op = &ext4_symlink_inode_operations;
 		} else {
 			inode->i_op = &ext4_fast_symlink_inode_operations;
-			inode->i_link = (char *)&EXT4_I(inode)->i_data;
 		}
 	}
 
@@ -3434,6 +3433,9 @@ static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
 		       disk_link.len);
 		inode->i_size = disk_link.len - 1;
 		EXT4_I(inode)->i_disksize = inode->i_size;
+		if (!IS_ENCRYPTED(inode))
+			inode_set_cached_link(inode, (char *)&EXT4_I(inode)->i_data,
+					      inode->i_size);
 	}
 	err = ext4_add_nondir(handle, dentry, &inode);
 	if (handle)

fs/namei.c

@@ -5272,19 +5272,16 @@ SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newna
 				getname(newname), 0);
 }
 
-int readlink_copy(char __user *buffer, int buflen, const char *link)
+int readlink_copy(char __user *buffer, int buflen, const char *link, int linklen)
 {
-	int len = PTR_ERR(link);
-	if (IS_ERR(link))
-		goto out;
+	int copylen;
 
-	len = strlen(link);
-	if (len > (unsigned) buflen)
-		len = buflen;
-	if (copy_to_user(buffer, link, len))
-		len = -EFAULT;
-out:
-	return len;
+	copylen = linklen;
+	if (unlikely(copylen > (unsigned) buflen))
+		copylen = buflen;
+	if (copy_to_user(buffer, link, copylen))
+		copylen = -EFAULT;
+	return copylen;
 }
 
 /**
@@ -5304,6 +5301,9 @@ int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
 	const char *link;
 	int res;
 
+	if (inode->i_opflags & IOP_CACHED_LINK)
+		return readlink_copy(buffer, buflen, inode->i_link, inode->i_linklen);
+
 	if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
 		if (unlikely(inode->i_op->readlink))
 			return inode->i_op->readlink(dentry, buffer, buflen);
@@ -5322,7 +5322,7 @@ int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
 		if (IS_ERR(link))
 			return PTR_ERR(link);
 	}
-	res = readlink_copy(buffer, buflen, link);
+	res = readlink_copy(buffer, buflen, link, strlen(link));
 	do_delayed_call(&done);
 	return res;
 }
@@ -5391,10 +5391,14 @@ EXPORT_SYMBOL(page_put_link);
 
 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
 {
+	const char *link;
+	int res;
+
 	DEFINE_DELAYED_CALL(done);
-	int res = readlink_copy(buffer, buflen,
-				page_get_link(dentry, d_inode(dentry),
-					      &done));
+	link = page_get_link(dentry, d_inode(dentry), &done);
+	res = PTR_ERR(link);
+	if (!IS_ERR(link))
+		res = readlink_copy(buffer, buflen, link, strlen(link));
 	do_delayed_call(&done);
 	return res;
 }

fs/proc/namespaces.c

@@ -83,7 +83,7 @@ static int proc_ns_readlink(struct dentry *dentry, char __user *buffer, int bufl
 	if (ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
 		res = ns_get_name(name, sizeof(name), task, ns_ops);
 		if (res >= 0)
-			res = readlink_copy(buffer, buflen, name);
+			res = readlink_copy(buffer, buflen, name, strlen(name));
 	}
 	put_task_struct(task);
 	return res;

include/linux/fs.h

@@ -626,6 +626,7 @@ is_uncached_acl(struct posix_acl *acl)
 #define IOP_XATTR	0x0008
 #define IOP_DEFAULT_READLINK	0x0010
 #define IOP_MGTIME	0x0020
+#define IOP_CACHED_LINK	0x0040
 
 /*
  * Keep mostly read-only and often accessed (especially for
@@ -723,7 +724,10 @@ struct inode {
 	};
 	struct file_lock_context	*i_flctx;
 	struct address_space	i_data;
-	struct list_head	i_devices;
+	union {
+		struct list_head	i_devices;
+		int			i_linklen;
+	};
 	union {
 		struct pipe_inode_info	*i_pipe;
 		struct cdev		*i_cdev;
@@ -749,6 +753,13 @@ struct inode {
 	void			*i_private; /* fs or device private pointer */
 } __randomize_layout;
 
+static inline void inode_set_cached_link(struct inode *inode, char *link, int linklen)
+{
+	inode->i_link = link;
+	inode->i_linklen = linklen;
+	inode->i_opflags |= IOP_CACHED_LINK;
+}
+
 /*
  * Get bit address from inode->i_state to use with wait_var_event()
  * infrastructre.
@@ -3351,7 +3362,7 @@ extern const struct file_operations generic_ro_fops;
 
 #define special_file(m) (S_ISCHR(m)||S_ISBLK(m)||S_ISFIFO(m)||S_ISSOCK(m))
 
-extern int readlink_copy(char __user *, int, const char *);
+extern int readlink_copy(char __user *, int, const char *, int);
 extern int page_readlink(struct dentry *, char __user *, int);
 extern const char *page_get_link(struct dentry *, struct inode *,
 				 struct delayed_call *);

mm/shmem.c

@@ -3914,6 +3914,7 @@ static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
 	int len;
 	struct inode *inode;
 	struct folio *folio;
+	char *link;
 
 	len = strlen(symname) + 1;
 	if (len > PAGE_SIZE)
@@ -3935,12 +3936,13 @@ static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
 
 	inode->i_size = len-1;
 	if (len <= SHORT_SYMLINK_LEN) {
-		inode->i_link = kmemdup(symname, len, GFP_KERNEL);
-		if (!inode->i_link) {
+		link = kmemdup(symname, len, GFP_KERNEL);
+		if (!link) {
 			error = -ENOMEM;
 			goto out_remove_offset;
 		}
 		inode->i_op = &shmem_short_symlink_operations;
+		inode_set_cached_link(inode, link, len - 1);
 	} else {
 		inode_nohighmem(inode);
 		inode->i_mapping->a_ops = &shmem_aops;

security/apparmor/apparmorfs.c

@@ -2612,7 +2612,7 @@ static int policy_readlink(struct dentry *dentry, char __user *buffer,
 	res = snprintf(name, sizeof(name), "%s:[%lu]", AAFS_NAME,
 		       d_inode(dentry)->i_ino);
 	if (res > 0 && res < sizeof(name))
-		res = readlink_copy(buffer, buflen, name);
+		res = readlink_copy(buffer, buflen, name, strlen(name));
 	else
 		res = -ENOENT;