John, thanks for the patch. If no problems are found I'd like to add
it to the 2.6.31 queue.
Csaba, Anand, does this meet your requirements?
Other comments?
Thanks,
Miklos
---
From: John Muir <[email protected]>
Add notification messages that allow the filesystem to invalidate VFS
caches. This is a much requested feature in FUSE.
Two notifications are added:
1) inode invalidation
- invalidate cached attributes
- invalidate a range of pages in the page cache (this is optional)
2) dentry invalidation
- try to invalidate a subtree in the dentry cache
Care must be taken while accessing the 'struct super_block' for the
mount, as it can go away while an invalidation is in progress. To
prevent this, introduce a rw-semaphore, that is taken for read during
the invalidation and taken for write in the ->kill_sb callback.
Cc: Csaba Henk <[email protected]>
Cc: Anand Avati <[email protected]>
Signed-off-by: Miklos Szeredi <[email protected]>
---
fs/fuse/dev.c | 81 +++++++++++++++++++++++++++++++++++++++++++++++++++
fs/fuse/dir.c | 37 +++++++++++++++++++++++
fs/fuse/fuse_i.h | 24 +++++++++++++++
fs/fuse/inode.c | 59 +++++++++++++++++++++++++++++++++++--
include/linux/fuse.h | 16 ++++++++++
5 files changed, 214 insertions(+), 3 deletions(-)
Index: linux-2.6/fs/fuse/dev.c
===================================================================
--- linux-2.6.orig/fs/fuse/dev.c 2009-06-02 16:20:59.000000000 +0200
+++ linux-2.6/fs/fuse/dev.c 2009-06-02 16:21:23.000000000 +0200
@@ -849,6 +849,81 @@ err:
return err;
}
+static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
+ struct fuse_copy_state *cs)
+{
+ struct fuse_notify_inval_inode_out outarg;
+ int err = -EINVAL;
+
+ if (size != sizeof(outarg))
+ goto err;
+
+ err = fuse_copy_one(cs, &outarg, sizeof(outarg));
+ if (err)
+ goto err;
+ fuse_copy_finish(cs);
+
+ down_read(&fc->killsb);
+ err = -ENOENT;
+ if (!fc->sb)
+ goto err_unlock;
+
+ err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
+ outarg.off, outarg.len);
+
+err_unlock:
+ up_read(&fc->killsb);
+ return err;
+
+err:
+ fuse_copy_finish(cs);
+ return err;
+}
+
+static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
+ struct fuse_copy_state *cs)
+{
+ struct fuse_notify_inval_entry_out outarg;
+ int err = -EINVAL;
+ char buf[FUSE_NAME_MAX+1];
+ struct qstr name;
+
+ if (size < sizeof(outarg))
+ goto err;
+
+ err = fuse_copy_one(cs, &outarg, sizeof(outarg));
+ if (err)
+ goto err;
+
+ err = -ENAMETOOLONG;
+ if (outarg.namelen > FUSE_NAME_MAX)
+ goto err;
+
+ name.name = buf;
+ name.len = outarg.namelen;
+ err = fuse_copy_one(cs, buf, outarg.namelen + 1);
+ if (err)
+ goto err;
+ fuse_copy_finish(cs);
+ buf[outarg.namelen] = 0;
+ name.hash = full_name_hash(name.name, name.len);
+
+ down_read(&fc->killsb);
+ err = -ENOENT;
+ if (!fc->sb)
+ goto err_unlock;
+
+ err = fuse_reverse_inval_entry(fc->sb, outarg.parent, &name);
+
+err_unlock:
+ up_read(&fc->killsb);
+ return err;
+
+err:
+ fuse_copy_finish(cs);
+ return err;
+}
+
static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
unsigned int size, struct fuse_copy_state *cs)
{
@@ -856,6 +931,12 @@ static int fuse_notify(struct fuse_conn
case FUSE_NOTIFY_POLL:
return fuse_notify_poll(fc, size, cs);
+ case FUSE_NOTIFY_INVAL_INODE:
+ return fuse_notify_inval_inode(fc, size, cs);
+
+ case FUSE_NOTIFY_INVAL_ENTRY:
+ return fuse_notify_inval_entry(fc, size, cs);
+
default:
fuse_copy_finish(cs);
return -EINVAL;
Index: linux-2.6/fs/fuse/dir.c
===================================================================
--- linux-2.6.orig/fs/fuse/dir.c 2009-06-02 16:20:59.000000000 +0200
+++ linux-2.6/fs/fuse/dir.c 2009-06-02 16:21:23.000000000 +0200
@@ -845,6 +845,43 @@ int fuse_update_attributes(struct inode
return err;
}
+int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
+ struct qstr *name)
+{
+ int err = -ENOTDIR;
+ struct inode *parent;
+ struct dentry *dir;
+ struct dentry *entry;
+
+ parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
+ if (!parent)
+ return -ENOENT;
+
+ mutex_lock(&parent->i_mutex);
+ if (!S_ISDIR(parent->i_mode))
+ goto unlock;
+
+ err = -ENOENT;
+ dir = d_find_alias(parent);
+ if (!dir)
+ goto unlock;
+
+ entry = d_lookup(dir, name);
+ dput(dir);
+ if (!entry)
+ goto unlock;
+
+ fuse_invalidate_attr(parent);
+ fuse_invalidate_entry(entry);
+ dput(entry);
+ err = 0;
+
+ unlock:
+ mutex_unlock(&parent->i_mutex);
+ iput(parent);
+ return err;
+}
+
/*
* Calling into a user-controlled filesystem gives the filesystem
* daemon ptrace-like capabilities over the requester process. This
Index: linux-2.6/fs/fuse/fuse_i.h
===================================================================
--- linux-2.6.orig/fs/fuse/fuse_i.h 2009-06-02 16:20:59.000000000 +0200
+++ linux-2.6/fs/fuse/fuse_i.h 2009-06-02 16:21:23.000000000 +0200
@@ -481,6 +481,12 @@ struct fuse_conn {
/** Called on final put */
void (*release)(struct fuse_conn *);
+
+ /** Super block for this connection. */
+ struct super_block *sb;
+
+ /** Read/write semaphore to hold when accessing sb. */
+ struct rw_semaphore killsb;
};
static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
@@ -509,6 +515,11 @@ extern const struct file_operations fuse
extern const struct dentry_operations fuse_dentry_operations;
/**
+ * Inode to nodeid comparison.
+ */
+int fuse_inode_eq(struct inode *inode, void *_nodeidp);
+
+/**
* Get a filled in inode
*/
struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
@@ -708,6 +719,19 @@ void fuse_release_nowrite(struct inode *
u64 fuse_get_attr_version(struct fuse_conn *fc);
+/**
+ * File-system tells the kernel to invalidate cache for the given node id.
+ */
+int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
+ loff_t offset, loff_t len);
+
+/**
+ * File-system tells the kernel to invalidate parent attributes and
+ * the dentry matching parent/name.
+ */
+int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
+ struct qstr *name);
+
int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
bool isdir);
ssize_t fuse_direct_io(struct file *file, const char __user *buf,
Index: linux-2.6/fs/fuse/inode.c
===================================================================
--- linux-2.6.orig/fs/fuse/inode.c 2009-06-02 16:20:59.000000000 +0200
+++ linux-2.6/fs/fuse/inode.c 2009-06-02 16:21:23.000000000 +0200
@@ -207,7 +207,7 @@ static void fuse_init_inode(struct inode
BUG();
}
-static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
+int fuse_inode_eq(struct inode *inode, void *_nodeidp)
{
u64 nodeid = *(u64 *) _nodeidp;
if (get_node_id(inode) == nodeid)
@@ -258,6 +258,31 @@ struct inode *fuse_iget(struct super_blo
return inode;
}
+int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
+ loff_t offset, loff_t len)
+{
+ struct inode *inode;
+ pgoff_t pg_start;
+ pgoff_t pg_end;
+
+ inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
+ if (!inode)
+ return -ENOENT;
+
+ fuse_invalidate_attr(inode);
+ if (offset >= 0) {
+ pg_start = offset >> PAGE_CACHE_SHIFT;
+ if (len <= 0)
+ pg_end = -1;
+ else
+ pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
+ invalidate_inode_pages2_range(inode->i_mapping,
+ pg_start, pg_end);
+ }
+ iput(inode);
+ return 0;
+}
+
static void fuse_umount_begin(struct super_block *sb)
{
lock_kernel();
@@ -483,6 +508,7 @@ void fuse_conn_init(struct fuse_conn *fc
memset(fc, 0, sizeof(*fc));
spin_lock_init(&fc->lock);
mutex_init(&fc->inst_mutex);
+ init_rwsem(&fc->killsb);
atomic_set(&fc->count, 1);
init_waitqueue_head(&fc->waitq);
init_waitqueue_head(&fc->blocked_waitq);
@@ -863,6 +889,7 @@ static int fuse_fill_super(struct super_
fuse_conn_init(fc);
fc->dev = sb->s_dev;
+ fc->sb = sb;
err = fuse_bdi_init(fc, sb);
if (err)
goto err_put_conn;
@@ -944,12 +971,25 @@ static int fuse_get_sb(struct file_syste
return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
}
+static void fuse_kill_sb_anon(struct super_block *sb)
+{
+ struct fuse_conn *fc = get_fuse_conn_super(sb);
+
+ if (fc) {
+ down_write(&fc->killsb);
+ fc->sb = NULL;
+ up_write(&fc->killsb);
+ }
+
+ kill_anon_super(sb);
+}
+
static struct file_system_type fuse_fs_type = {
.owner = THIS_MODULE,
.name = "fuse",
.fs_flags = FS_HAS_SUBTYPE,
.get_sb = fuse_get_sb,
- .kill_sb = kill_anon_super,
+ .kill_sb = fuse_kill_sb_anon,
};
#ifdef CONFIG_BLOCK
@@ -961,11 +1001,24 @@ static int fuse_get_sb_blk(struct file_s
mnt);
}
+static void fuse_kill_sb_blk(struct super_block *sb)
+{
+ struct fuse_conn *fc = get_fuse_conn_super(sb);
+
+ if (fc) {
+ down_write(&fc->killsb);
+ fc->sb = NULL;
+ up_write(&fc->killsb);
+ }
+
+ kill_block_super(sb);
+}
+
static struct file_system_type fuseblk_fs_type = {
.owner = THIS_MODULE,
.name = "fuseblk",
.get_sb = fuse_get_sb_blk,
- .kill_sb = kill_block_super,
+ .kill_sb = fuse_kill_sb_blk,
.fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
};
Index: linux-2.6/include/linux/fuse.h
===================================================================
--- linux-2.6.orig/include/linux/fuse.h 2009-06-02 16:20:59.000000000 +0200
+++ linux-2.6/include/linux/fuse.h 2009-06-02 16:21:23.000000000 +0200
@@ -25,6 +25,8 @@
* - add IOCTL message
* - add unsolicited notification support
* - add POLL message and NOTIFY_POLL notification
+ * - add notification messages for invalidation of inodes and
+ * directory entries
*/
#ifndef _LINUX_FUSE_H
@@ -224,6 +226,8 @@ enum fuse_opcode {
enum fuse_notify_code {
FUSE_NOTIFY_POLL = 1,
+ FUSE_NOTIFY_INVAL_INODE = 2,
+ FUSE_NOTIFY_INVAL_ENTRY = 3,
FUSE_NOTIFY_CODE_MAX,
};
@@ -508,4 +512,16 @@ struct fuse_dirent {
#define FUSE_DIRENT_SIZE(d) \
FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + (d)->namelen)
+struct fuse_notify_inval_inode_out {
+ __u64 ino;
+ __s64 off;
+ __s64 len;
+};
+
+struct fuse_notify_inval_entry_out {
+ __u64 parent;
+ __u32 namelen;
+ __u32 padding;
+};
+
#endif /* _LINUX_FUSE_H */
----- "Miklos Szeredi" <[email protected]> wrote:
> John, thanks for the patch. If no problems are found I'd like to add
> it to the 2.6.31 queue.
>
> Csaba, Anand, does this meet your requirements?
>
> Other comments?
I was wondering how the userland libfuse would be made aware of the presence of these two notification features?
Avati
On Wed, 3 Jun 2009, Anand Avati wrote:
> I was wondering how the userland libfuse would be made aware of the
> presence of these two notification features?
Here's a patch against fuse CVS. It's also from John Muir.
I'm still thinking a bit about the API, how it would work in the
presence of stacked filesystems or multiple channels.
Thanks,
Miklos
Index: fuse/include/fuse_kernel.h
===================================================================
--- fuse.orig/include/fuse_kernel.h 2009-06-04 11:09:56.000000000 +0200
+++ fuse/include/fuse_kernel.h 2009-06-04 11:10:11.000000000 +0200
@@ -51,6 +51,8 @@
* - add IOCTL message
* - add unsolicited notification support
* - add POLL message and NOTIFY_POLL notification
+ * - add notification messages for invalidation of inodes and
+ * directory entries
*/
#ifndef _LINUX_FUSE_H
@@ -58,6 +60,7 @@
#include <sys/types.h>
#define __u64 uint64_t
+#define __s64 int64_t
#define __u32 uint32_t
#define __s32 int32_t
@@ -243,6 +246,8 @@ enum fuse_opcode {
enum fuse_notify_code {
FUSE_NOTIFY_POLL = 1,
+ FUSE_NOTIFY_INVAL_INODE = 2,
+ FUSE_NOTIFY_INVAL_ENTRY = 3,
FUSE_NOTIFY_CODE_MAX,
};
@@ -506,4 +511,16 @@ struct fuse_dirent {
#define FUSE_DIRENT_SIZE(d) \
FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + (d)->namelen)
+struct fuse_notify_inval_inode_out {
+ __u64 ino;
+ __s64 off;
+ __s64 len;
+};
+
+struct fuse_notify_inval_entry_out {
+ __u64 parent;
+ __u32 namelen;
+ __u32 padding;
+};
+
#endif /* _LINUX_FUSE_H */
Index: fuse/include/fuse_lowlevel.h
===================================================================
--- fuse.orig/include/fuse_lowlevel.h 2009-06-04 11:09:56.000000000 +0200
+++ fuse/include/fuse_lowlevel.h 2009-06-04 11:10:11.000000000 +0200
@@ -1136,6 +1136,32 @@ int fuse_reply_poll(fuse_req_t req, unsi
*/
int fuse_lowlevel_notify_poll(struct fuse_pollhandle *ph);
+/**
+ * Notify to invalidate cache for an inode
+ *
+ * @param ch the channel through which to send the invalidation
+ * @param ino the inode number
+ * @param off the offset in the inode where to start invalidating
+ * or negative to invalidate attributes only
+ * @param len the amount of cache to invalidate or 0 for all
+ * @return zero for success, -errno for failure
+ */
+int fuse_lowlevel_notify_inval_inode(struct fuse_chan *ch, fuse_ino_t ino,
+ off_t off, off_t len);
+
+/**
+ * Notify to invalidate parent attributes and the dentry matching
+ * parent/name
+ *
+ * @param ch the channel through which to send the invalidation
+ * @param parent inode number
+ * @param name file name
+ * @param namelen strlen() of file name
+ * @return zero for success, -errno for failure
+ */
+int fuse_lowlevel_notify_inval_entry(struct fuse_chan *ch, fuse_ino_t parent,
+ const char *name, size_t namelen);
+
/* ----------------------------------------------------------- *
* Utility functions *
* ----------------------------------------------------------- */
@@ -1340,6 +1366,14 @@ void fuse_session_reset(struct fuse_sess
int fuse_session_exited(struct fuse_session *se);
/**
+ * Get the user data provided to the session
+ *
+ * @param se the session
+ * @return the user data
+ */
+void *fuse_session_data(struct fuse_session *se);
+
+/**
* Enter a single threaded event loop
*
* @param se the session
Index: fuse/lib/fuse_lowlevel.c
===================================================================
--- fuse.orig/lib/fuse_lowlevel.c 2009-06-04 11:09:56.000000000 +0200
+++ fuse/lib/fuse_lowlevel.c 2009-06-04 11:10:11.000000000 +0200
@@ -1280,6 +1280,56 @@ int fuse_lowlevel_notify_poll(struct fus
}
}
+int fuse_lowlevel_notify_inval_inode(struct fuse_chan *ch, fuse_ino_t ino,
+ off_t off, off_t len)
+{
+ struct fuse_notify_inval_inode_out outarg;
+ struct fuse_ll *f;
+ struct iovec iov[2];
+
+ if (!ch)
+ return -EINVAL;
+
+ f = (struct fuse_ll *)fuse_session_data(fuse_chan_session(ch));
+ if (!f)
+ return -ENODEV;
+
+ outarg.ino = ino;
+ outarg.off = off;
+ outarg.len = len;
+
+ iov[1].iov_base = &outarg;
+ iov[1].iov_len = sizeof(outarg);
+
+ return send_notify_iov(f, ch, FUSE_NOTIFY_INVAL_INODE, iov, 2);
+}
+
+int fuse_lowlevel_notify_inval_entry(struct fuse_chan *ch, fuse_ino_t parent,
+ const char *name, size_t namelen)
+{
+ struct fuse_notify_inval_entry_out outarg;
+ struct fuse_ll *f;
+ struct iovec iov[3];
+
+ if (!ch)
+ return -EINVAL;
+
+ f = (struct fuse_ll *)fuse_session_data(fuse_chan_session(ch));
+ if (!f)
+ return -ENODEV;
+
+ outarg.parent = parent;
+ outarg.namelen = namelen;
+ outarg.padding = 0;
+
+ iov[1].iov_base = &outarg;
+ iov[1].iov_len = sizeof(outarg);
+ iov[2].iov_base = (void *)name;
+ iov[2].iov_len = namelen + 1;
+
+ return send_notify_iov(f, ch, FUSE_NOTIFY_INVAL_ENTRY, iov, 3);
+}
+
void *fuse_req_userdata(fuse_req_t req)
{
return req->f->userdata;
Index: fuse/lib/fuse_session.c
===================================================================
--- fuse.orig/lib/fuse_session.c 2009-06-04 11:09:56.000000000 +0200
+++ fuse/lib/fuse_session.c 2009-06-04 11:10:11.000000000 +0200
@@ -121,6 +121,11 @@ int fuse_session_exited(struct fuse_sess
return se->exited;
}
+void *fuse_session_data(struct fuse_session *se)
+{
+ return se->data;
+}
+
static struct fuse_chan *fuse_chan_new_common(struct fuse_chan_ops *op, int fd,
size_t bufsize, void *data,
int compat)
Index: fuse/lib/fuse_versionscript
===================================================================
--- fuse.orig/lib/fuse_versionscript 2009-06-04 11:09:56.000000000 +0200
+++ fuse/lib/fuse_versionscript 2009-06-04 11:10:11.000000000 +0200
@@ -165,6 +165,8 @@ FUSE_2.8 {
fuse_reply_ioctl;
fuse_reply_ioctl_retry;
fuse_reply_poll;
+ fuse_lowlevel_notify_inval_inode;
+ fuse_lowlevel_notify_inval_entry;
local:
*;
On Thu, Jun 4, 2009 at 11:22 AM, Miklos Szeredi <[email protected]> wrote:
> On Wed, 3 Jun 2009, Anand Avati wrote:
>> I was wondering how the userland libfuse would be made aware of the
>> presence of these two notification features?
>
> Here's a patch against fuse CVS. ?It's also from John Muir.
In fact, what we were meditating about is the lack of the
FUSE_KERNEL_MINOR_VERSION bump. Without that, how do the two parties
negotiate if the rev inval messages are understood? One of John's
patches in his patchset had it bumped though...
Csaba
On Thu, 4 Jun 2009, Csaba Henk wrote:
> On Thu, Jun 4, 2009 at 11:22 AM, Miklos Szeredi <[email protected]> wrote:
> > On Wed, 3 Jun 2009, Anand Avati wrote:
> >> I was wondering how the userland libfuse would be made aware of the
> >> presence of these two notification features?
> >
> > Here's a patch against fuse CVS. It's also from John Muir.
>
> In fact, what we were meditating about is the lack of the
> FUSE_KERNEL_MINOR_VERSION bump. Without that, how do the two parties
> negotiate if the rev inval messages are understood? One of John's
> patches in his patchset had it bumped though...
Yes, I removed that intentionally as hopefully this will go in 2.6.31
with the other 10 -> 11 API changes.
Thanks,
Miklos