Skip to contentSkip to conversation

Sort out fsnotify_nameremove() mess

PATCH In review Amir Goldstein 5 participants 29 messages May 26, 2019 – Sep 2, 2019

Subsystems (10)
lore
Share & tools
lore

Amir Goldstein

author reviewer
#0 lore raw

v3 cover Sort out fsnotify_nameremove() mess

Jan,

For v3 I went with a straight forward approach. Filesystems that have fsnotify_{create,mkdir} hooks also get explicit fsnotify_{unlink,rmdir} hooks.

Hopefully, this approach is orthogonal to whatever changes Al is planning for recursive tree remove code, because in many of the cases, the hooks are added at the entry point for the recursive tree remove.

After looking closer at all the filesystems that were converted to simple_remove in v2, I decided to exempt another 3 filesystems from the fsnotify delete hooks: hypfs,qibfs and aafs.
hypfs is pure cleanup (*). qibfs and aafs can remove dentry on user configuration change, but they do not generate create events, so it is less likely that users depend on the delete events.

That leaves configfs the only filesystem that gets the new delete hooks even though it does not have create hooks.

The following d_delete() call sites have been audited and will no longer generate fsnotify event after this series:

arch/s390/hypfs/inode.c:hypfs_remove() - cleanup (*) .../usb/gadget/function/f_fs.c:ffs_epfiles_destroy() - no create events .../infiniband/hw/qib/qib_fs.c:remove_device_files() - no create events fs/ceph/dir.c:ceph_unlink() - from vfs_unlink() fs/ceph/inode.c:ceph_fill_trace() - invalidate (**) fs/ceph/inode.c:ceph_readdir_prepopulate() - invalidate (**) fs/configfs/dir.c:detach_groups() - hooks added, from vfs or cleanup (*) fs/configfs/dir.c:configfs_attach_item() - cleanup (*) fs/configfs/dir.c:configfs_attach_group() - cleanup (*) fs/efivarfs/file.c:efivarfs_file_write() - invalidate (**) fs/fuse/dir.c:fuse_reverse_inval_entry() - invalidate (**) fs/nfs/dir.c:nfs_dentry_handle_enoent() - invalidate (**) fs/nsfs.c:__ns_get_path() - cleanup (*) fs/ocfs2/dlmglue.c:ocfs2_dentry_convert_worker() - invalidate (**) fs/reiserfs/xattr.c:xattr_{unlink,rmdir}() - hidden xattr inode security/apparmor/apparmorfs.c:aafs_remove() - no create events

(*) There are 2 "cleanup" use cases:
- Create;init;delete if init failed
- Batch delete of files within dir before removing dir Both those cases are not interesting for users that wish to observe configuration changes on pseudo filesystems. Often, there is already an fsnotify event generated on the directory removal which is what users should find interesting, for example:
configfs_unregister_{group,subsystem}().

(**) The different "invalidate" use cases differ, but they all share
one thing in common - user is not guarantied to get an event with current kernel. For example, when a file is deleted remotely on nfs server, nfs client is not guarantied to get an fsnotify delete event. On current kernel, nfs client could generate an fsnotify delete event if the local entry happens to be in cache and client finds out that entry is deleted on server during another user operation. Incidentally, this group of use cases is where most of the call sites are with "unstable" d_name, which is the reason for this patch series to begin with.

Thanks,
Amir.

Changes since v2:
- Drop simple_rename() conversions (add explicit hooks instead)
- Drop hooks from hypfs/qibfs/aafs
- Split out debugfs re-factoring patch

Changes since v1:
- Split up per filesystem patches
- New hook names fsnotify_{unlink,rmdir}()
- Simplify fsnotify code in separate final patch

Amir Goldstein (10):
fsnotify: add empty fsnotify_{unlink,rmdir}() hooks
btrfs: call fsnotify_rmdir() hook
rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks
tracefs: call fsnotify_{unlink,rmdir}() hooks
devpts: call fsnotify_unlink() hook
debugfs: simplify __debugfs_remove_file()
debugfs: call fsnotify_{unlink,rmdir}() hooks
configfs: call fsnotify_rmdir() hook
fsnotify: move fsnotify_nameremove() hook out of d_delete()
fsnotify: get rid of fsnotify_nameremove()

 fs/afs/dir_silly.c               |  5 ----
 fs/btrfs/ioctl.c                 |  4 +++-
 fs/configfs/dir.c                |  3 +++
 fs/dcache.c                      |  2 --
 fs/debugfs/inode.c               | 21 ++++++++--------
 fs/devpts/inode.c                |  1 +
 fs/namei.c                       |  2 ++
 fs/nfs/unlink.c                  |  6 -----
 fs/notify/fsnotify.c             | 41 --------------------------------
 fs/tracefs/inode.c               |  3 +++
 include/linux/fsnotify.h         | 26 ++++++++++++++++++++
 include/linux/fsnotify_backend.h |  4 ----
 net/sunrpc/rpc_pipe.c            |  4 ++++
 13 files changed, 52 insertions(+), 70 deletions(-)
Signature · 1 line
2.17.1
author reviewer
#1 lore raw

PATCH v3 1/10 fsnotify: add empty fsnotify_{unlink,rmdir}() hooks

We would like to move fsnotify_nameremove() calls from d_delete() into a higher layer where the hook makes more sense and so we can consider every d_delete() call site individually.

Start by creating empty hook fsnotify_{unlink,rmdir}() and place them in the proper VFS call sites. After all d_delete() call sites will be converted to use the new hook, the new hook will generate the delete events and fsnotify_nameremove() hook will be removed.

---
 fs/namei.c               |  2 ++
 include/linux/fsnotify.h | 26 ++++++++++++++++++++++++++
 2 files changed, 28 insertions(+)
fs/namei.c +2 -0
index 20831c2fbb34..209c51a5226c 100644
@@ -3883,6 +3883,7 @@ int vfs_rmdir(struct inode *dir, struct dentry *dentry)
 	dentry->d_inode->i_flags |= S_DEAD;
 	dont_mount(dentry);
 	detach_mounts(dentry);
+	fsnotify_rmdir(dir, dentry);
 
 out:
 	inode_unlock(dentry->d_inode);
@@ -3999,6 +4000,7 @@ int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegate
 			if (!error) {
 				dont_mount(dentry);
 				detach_mounts(dentry);
+				fsnotify_unlink(dir, dentry);
 			}
 		}
 	}
include/linux/fsnotify.h +26 -0
index 94972e8eb6d1..7f23eddefcd0 100644
@@ -188,6 +188,19 @@ static inline void fsnotify_link(struct inode *dir, struct inode *inode, struct
 	fsnotify(dir, FS_CREATE, inode, FSNOTIFY_EVENT_INODE, &new_dentry->d_name, 0);
 }
 
+/*
+ * fsnotify_unlink - 'name' was unlinked
+ *
+ * Caller must make sure that dentry->d_name is stable.
+ */
+static inline void fsnotify_unlink(struct inode *dir, struct dentry *dentry)
+{
+	/* Expected to be called before d_delete() */
+	WARN_ON_ONCE(d_is_negative(dentry));
+
+	/* TODO: call fsnotify_dirent() */
+}
+
 /*
  * fsnotify_mkdir - directory 'name' was created
  */
@@ -198,6 +211,19 @@ static inline void fsnotify_mkdir(struct inode *inode, struct dentry *dentry)
 	fsnotify_dirent(inode, dentry, FS_CREATE | FS_ISDIR);
 }
 
+/*
+ * fsnotify_rmdir - directory 'name' was removed
+ *
+ * Caller must make sure that dentry->d_name is stable.
+ */
+static inline void fsnotify_rmdir(struct inode *dir, struct dentry *dentry)
+{
+	/* Expected to be called before d_delete() */
+	WARN_ON_ONCE(d_is_negative(dentry));
+
+	/* TODO: call fsnotify_dirent() */
+}
+
 /*
  * fsnotify_access - file was read
  */
Signature · 1 line
2.17.1
author reviewer
#2 lore raw

PATCH v3 2/10 btrfs: call fsnotify_rmdir() hook

In reply to Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

---
 fs/btrfs/ioctl.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
fs/btrfs/ioctl.c +3 -1
index 6dafa857bbb9..2cfd1bfb3871 100644
@@ -2930,8 +2930,10 @@ static noinline int btrfs_ioctl_snap_destroy(struct file *file,
 	inode_lock(inode);
 	err = btrfs_delete_subvolume(dir, dentry);
 	inode_unlock(inode);
-	if (!err)
+	if (!err) {
+		fsnotify_rmdir(dir, dentry);
 		d_delete(dentry);
+	}
 
 out_dput:
 	dput(dentry);
Signature · 1 line
2.17.1
author reviewer
#3 lore raw

PATCH v3 3/10 rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks

In reply to Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Trond Myklebust <[email protected]>
Cc Anna Schumaker <[email protected]>
---
 net/sunrpc/rpc_pipe.c | 4 ++++
 1 file changed, 4 insertions(+)
net/sunrpc/rpc_pipe.c +4 -0
index 979d23646e33..917c85f15a0b 100644
@@ -597,6 +597,8 @@ static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
 
 	dget(dentry);
 	ret = simple_rmdir(dir, dentry);
+	if (!ret)
+		fsnotify_rmdir(dir, dentry);
 	d_delete(dentry);
 	dput(dentry);
 	return ret;
@@ -608,6 +610,8 @@ static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
 
 	dget(dentry);
 	ret = simple_unlink(dir, dentry);
+	if (!ret)
+		fsnotify_unlink(dir, dentry);
 	d_delete(dentry);
 	dput(dentry);
 	return ret;
Signature · 1 line
2.17.1
maintainer
#4 lore raw

Re: [PATCH v3 03/10] rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks

Quoted 9 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Trond Myklebust <[email protected]>
Cc Anna Schumaker <[email protected]>
---
 net/sunrpc/rpc_pipe.c | 4 ++++
 1 file changed, 4 insertions(+)

Hum, I don't think all __rpc_depopulate() calls are guarded by i_rwsem (e.g. those in rpc_gssd_dummy_populate()). Why aren't we using rpc_depopulate() in those cases? Trond, Anna?

Honza

Quoted 26 lines
net/sunrpc/rpc_pipe.c +4 -0
index 979d23646e33..917c85f15a0b 100644
@@ -597,6 +597,8 @@ static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
 
 	dget(dentry);
 	ret = simple_rmdir(dir, dentry);
+	if (!ret)
+		fsnotify_rmdir(dir, dentry);
 	d_delete(dentry);
 	dput(dentry);
 	return ret;
@@ -608,6 +610,8 @@ static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
 
 	dget(dentry);
 	ret = simple_unlink(dir, dentry);
+	if (!ret)
+		fsnotify_unlink(dir, dentry);
 	d_delete(dentry);
 	dput(dentry);
 	return ret;
Signature · 1 line
2.17.1
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author reviewer
#5 lore raw

Re: [PATCH v3 03/10] rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks

Quoted 12 lines · Jan Kara
Quoted 9 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Trond Myklebust <[email protected]>
Cc Anna Schumaker <[email protected]>
---
 net/sunrpc/rpc_pipe.c | 4 ++++
 1 file changed, 4 insertions(+)
Quoted 4 lines

Hum, I don't think all __rpc_depopulate() calls are guarded by i_rwsem (e.g. those in rpc_gssd_dummy_populate()). Why aren't we using rpc_depopulate() in those cases? Trond, Anna?

Do we care? For fsnotify hook, we should only care that d_parent/d_name are stable.
They are stable because rpc_pipefs has no rename.

Thanks,
Amir.

maintainer
#6 lore raw

Re: [PATCH v3 03/10] rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks

Quoted 13 lines · Amir Goldstein
Quoted 12 lines · Jan Kara
Quoted 9 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Trond Myklebust <[email protected]>
Cc Anna Schumaker <[email protected]>
---
 net/sunrpc/rpc_pipe.c | 4 ++++
 1 file changed, 4 insertions(+)
Quoted 8 lines
Quoted 4 lines

Hum, I don't think all __rpc_depopulate() calls are guarded by i_rwsem (e.g. those in rpc_gssd_dummy_populate()). Why aren't we using rpc_depopulate() in those cases? Trond, Anna?

Do we care? For fsnotify hook, we should only care that d_parent/d_name are stable.
They are stable because rpc_pipefs has no rename.

Yeah, good point. Probably we don't. I was just wondering...

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author reviewer
#7 lore raw

Re: [PATCH v3 03/10] rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks

In reply to Amir Goldstein

Hi Trond,Anna,Bruce,

Seems that rpc_pipefs is co-maintained by client and server trees, so not sure who is the best to ask for an ack.
We need to add those explicit fsnotify hooks to match the existing fsnotify_create/mkdir hooks, because
the hook embedded inside d_delete() is going away [1].

Thanks,
Amir.

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/

Quoted 36 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Trond Myklebust <[email protected]>
Cc Anna Schumaker <[email protected]>
---
 net/sunrpc/rpc_pipe.c | 4 ++++
 1 file changed, 4 insertions(+)
net/sunrpc/rpc_pipe.c +4 -1
index 979d23646e33..917c85f15a0b 100644
@@ -597,6 +597,8 @@ static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
 
        dget(dentry);
        ret = simple_rmdir(dir, dentry);
+       if (!ret)
+               fsnotify_rmdir(dir, dentry);
        d_delete(dentry);
        dput(dentry);
        return ret;
@@ -608,6 +610,8 @@ static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
 
        dget(dentry);
        ret = simple_unlink(dir, dentry);
+       if (!ret)
+               fsnotify_unlink(dir, dentry);
        d_delete(dentry);
        dput(dentry);
        return ret;
--

2.17.1

#8 lore raw

Re: [PATCH v3 03/10] rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks

Quoted 43 lines · Amir Goldstein

Hi Trond,Anna,Bruce,

Seems that rpc_pipefs is co-maintained by client and server trees, so not sure who is the best to ask for an ack.
We need to add those explicit fsnotify hooks to match the existing fsnotify_create/mkdir hooks, because
the hook embedded inside d_delete() is going away [1].

Thanks,
Amir.

[1]
https://lore.kernel.org/linux-fsdevel/[email protected]/

Quoted 27 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Trond Myklebust <[email protected]>
Cc Anna Schumaker <[email protected]>
---
 net/sunrpc/rpc_pipe.c | 4 ++++
 1 file changed, 4 insertions(+)
net/sunrpc/rpc_pipe.c +2 -0
index 979d23646e33..917c85f15a0b 100644
@@ -597,6 +597,8 @@ static int __rpc_rmdir(struct inode *dir,
 struct dentry *dentry)
 
        dget(dentry);
        ret = simple_rmdir(dir, dentry);
+       if (!ret)
+               fsnotify_rmdir(dir, dentry);
        d_delete(dentry);
        dput(dentry);
        return ret;
net/sunrpc/rpc_pipe.c +0 -0
@@ -608,6 +610,8 @@ static int __rpc_unlink(struct inode *dir,
 struct dentry *dentry)
 
Quoted 10 lines
Quoted 10 lines
net/sunrpc/rpc_pipe.c +2 -0
        dget(dentry);
        ret = simple_unlink(dir, dentry);
+       if (!ret)
+               fsnotify_unlink(dir, dentry);
        d_delete(dentry);
        dput(dentry);
        return ret;

--
2.17.1

This looks just fine to me.

Signature · 3 lines
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
author reviewer
#9 lore raw

PATCH v3 4/10 tracefs: call fsnotify_{unlink,rmdir}() hooks

In reply to Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

---
 fs/tracefs/inode.c | 3 +++
 1 file changed, 3 insertions(+)
fs/tracefs/inode.c +3 -0
index 7098c49f3693..497a8682b5b9 100644
@@ -509,9 +509,12 @@ static int __tracefs_remove(struct dentry *dentry, struct dentry *parent)
 			switch (dentry->d_inode->i_mode & S_IFMT) {
 			case S_IFDIR:
 				ret = simple_rmdir(parent->d_inode, dentry);
+				if (!ret)
+					fsnotify_rmdir(parent->d_inode, dentry);
 				break;
 			default:
 				simple_unlink(parent->d_inode, dentry);
+				fsnotify_unlink(parent->d_inode, dentry);
 				break;
 			}
 			if (!ret)
Signature · 1 line
2.17.1
author reviewer
#10 lore raw

Re: [PATCH v3 04/10] tracefs: call fsnotify_{unlink,rmdir}() hooks

Quoted 6 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Hi Steven,

Would you be able to provide an ACK on this patch?
We need to add those explicit fsnotify hooks to match the existing fsnotify_create/mkdir hooks in tracefs, because
the hook embedded inside d_delete() is going away [1].

Thanks,
Amir.

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/

Quoted 24 lines
---
 fs/tracefs/inode.c | 3 +++
 1 file changed, 3 insertions(+)
fs/tracefs/inode.c +3 -1
index 7098c49f3693..497a8682b5b9 100644
@@ -509,9 +509,12 @@ static int __tracefs_remove(struct dentry *dentry, struct dentry *parent)
                        switch (dentry->d_inode->i_mode & S_IFMT) {
                        case S_IFDIR:
                                ret = simple_rmdir(parent->d_inode, dentry);
+                               if (!ret)
+                                       fsnotify_rmdir(parent->d_inode, dentry);
                                break;
                        default:
                                simple_unlink(parent->d_inode, dentry);
+                               fsnotify_unlink(parent->d_inode, dentry);
                                break;
                        }
                        if (!ret)
--

2.17.1

maintainer
#11 lore raw

Re: [PATCH v3 04/10] tracefs: call fsnotify_{unlink,rmdir}() hooks

Quoted 15 lines · On Thu, 13 Jun 2019 19:53:25 +0300 Amir Goldstein <[email protected]> wrote:
Quoted 6 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Hi Steven,

Would you be able to provide an ACK on this patch?
We need to add those explicit fsnotify hooks to match the existing fsnotify_create/mkdir hooks in tracefs, because
the hook embedded inside d_delete() is going away [1].

Quoted 5 lines

Thanks,
Amir.

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/

Sorry, this got lost in my INBOX. I see it was already merged, but I would have acked it ;-)

-- Steve

maintainer
#12 lore raw

Re: [PATCH v3 04/10] tracefs: call fsnotify_{unlink,rmdir}() hooks

Quoted 18 lines · Steven Rostedt
Quoted 15 lines · On Thu, 13 Jun 2019 19:53:25 +0300 Amir Goldstein <[email protected]> wrote:
Quoted 6 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Hi Steven,

Would you be able to provide an ACK on this patch?
We need to add those explicit fsnotify hooks to match the existing fsnotify_create/mkdir hooks in tracefs, because
the hook embedded inside d_delete() is going away [1].

Quoted 8 lines
Quoted 5 lines

Thanks,
Amir.

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/

Sorry, this got lost in my INBOX. I see it was already merged, but I would have acked it ;-)

Thanks for the dealayed ack. I've used my best judgement and decided that you probably would not object when pushing the patch :).
Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author reviewer
#13 lore raw

PATCH v3 5/10 devpts: call fsnotify_unlink() hook

In reply to Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

---
 fs/devpts/inode.c | 1 +
 1 file changed, 1 insertion(+)
fs/devpts/inode.c +1 -0
index 553a3f3300ae..1d7844809f01 100644
@@ -624,6 +624,7 @@ void devpts_pty_kill(struct dentry *dentry)
 
 	dentry->d_fsdata = NULL;
 	drop_nlink(dentry->d_inode);
+	fsnotify_unlink(d_inode(dentry->d_parent), dentry);
 	d_delete(dentry);
 	dput(dentry);	/* d_alloc_name() in devpts_pty_new() */
 }
Signature · 1 line
2.17.1
author reviewer
#14 lore raw

PATCH v3 6/10 debugfs: simplify __debugfs_remove_file()

In reply to Amir Goldstein

Move simple_unlink()+d_delete() from __debugfs_remove_file() into caller __debugfs_remove() and rename helper for post remove file to __debugfs_file_removed().

This will simplify adding fsnotify_unlink() hook.

Cc Greg Kroah-Hartman <[email protected]>
---
 fs/debugfs/inode.c | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)
fs/debugfs/inode.c +8 -12
index acef14ad53db..d89874da9791 100644
@@ -617,13 +617,10 @@ struct dentry *debugfs_create_symlink(const char *name, struct dentry *parent,
 }
 EXPORT_SYMBOL_GPL(debugfs_create_symlink);
 
-static void __debugfs_remove_file(struct dentry *dentry, struct dentry *parent)
+static void __debugfs_file_removed(struct dentry *dentry)
 {
 	struct debugfs_fsdata *fsd;
 
-	simple_unlink(d_inode(parent), dentry);
-	d_delete(dentry);
-
 	/*
 	 * Paired with the closing smp_mb() implied by a successful
 	 * cmpxchg() in debugfs_file_get(): either
@@ -644,16 +641,15 @@ static int __debugfs_remove(struct dentry *dentry, struct dentry *parent)
 
 	if (simple_positive(dentry)) {
 		dget(dentry);
-		if (!d_is_reg(dentry)) {
-			if (d_is_dir(dentry))
-				ret = simple_rmdir(d_inode(parent), dentry);
-			else
-				simple_unlink(d_inode(parent), dentry);
-			if (!ret)
-				d_delete(dentry);
+		if (d_is_dir(dentry)) {
+			ret = simple_rmdir(d_inode(parent), dentry);
 		} else {
-			__debugfs_remove_file(dentry, parent);
+			simple_unlink(d_inode(parent), dentry);
 		}
+		if (!ret)
+			d_delete(dentry);
+		if (d_is_reg(dentry))
+			__debugfs_file_removed(dentry);
 		dput(dentry);
 	}
 	return ret;
Signature · 1 line
2.17.1
author reviewer
#15 lore raw

Re: [PATCH v3 06/10] debugfs: simplify __debugfs_remove_file()

Quoted 8 lines · Amir Goldstein

Move simple_unlink()+d_delete() from __debugfs_remove_file() into caller __debugfs_remove() and rename helper for post remove file to __debugfs_file_removed().

This will simplify adding fsnotify_unlink() hook.

Cc Greg Kroah-Hartman <[email protected]>

Hi Greg,

Will be you be able to provide an ACK on this debugfs patch and the next one.

Thanks,
Amir.

Quoted 51 lines
---
 fs/debugfs/inode.c | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)
fs/debugfs/inode.c +8 -13
index acef14ad53db..d89874da9791 100644
@@ -617,13 +617,10 @@ struct dentry *debugfs_create_symlink(const char *name, struct dentry *parent,
 }
 EXPORT_SYMBOL_GPL(debugfs_create_symlink);
 
-static void __debugfs_remove_file(struct dentry *dentry, struct dentry *parent)
+static void __debugfs_file_removed(struct dentry *dentry)
 {
        struct debugfs_fsdata *fsd;
 
-       simple_unlink(d_inode(parent), dentry);
-       d_delete(dentry);
-
        /*
         * Paired with the closing smp_mb() implied by a successful
         * cmpxchg() in debugfs_file_get(): either
@@ -644,16 +641,15 @@ static int __debugfs_remove(struct dentry *dentry, struct dentry *parent)
 
        if (simple_positive(dentry)) {
                dget(dentry);
-               if (!d_is_reg(dentry)) {
-                       if (d_is_dir(dentry))
-                               ret = simple_rmdir(d_inode(parent), dentry);
-                       else
-                               simple_unlink(d_inode(parent), dentry);
-                       if (!ret)
-                               d_delete(dentry);
+               if (d_is_dir(dentry)) {
+                       ret = simple_rmdir(d_inode(parent), dentry);
                } else {
-                       __debugfs_remove_file(dentry, parent);
+                       simple_unlink(d_inode(parent), dentry);
                }
+               if (!ret)
+                       d_delete(dentry);
+               if (d_is_reg(dentry))
+                       __debugfs_file_removed(dentry);
                dput(dentry);
        }
        return ret;
--

2.17.1

maintainer
#16 lore raw

Re: [PATCH v3 06/10] debugfs: simplify __debugfs_remove_file()

Quoted 6 lines · Amir Goldstein
Quoted 5 lines · Amir Goldstein

Move simple_unlink()+d_delete() from __debugfs_remove_file() into caller __debugfs_remove() and rename helper for post remove file to __debugfs_file_removed().

Quoted 7 lines
Quoted 3 lines

This will simplify adding fsnotify_unlink() hook.

Cc Greg Kroah-Hartman <[email protected]>

Hi Greg,

Will be you be able to provide an ACK on this debugfs patch and the next one.

It's only been 4 days, and I'm traveling this week, give me a chance...

maintainer Reviewed-by on v3 6/10
#17 lore raw

Re: [PATCH v3 06/10] debugfs: simplify __debugfs_remove_file()

In reply to Amir Goldstein
Quoted 45 lines · Amir Goldstein

Move simple_unlink()+d_delete() from __debugfs_remove_file() into caller __debugfs_remove() and rename helper for post remove file to __debugfs_file_removed().

This will simplify adding fsnotify_unlink() hook.

Cc Greg Kroah-Hartman <[email protected]>
---
 fs/debugfs/inode.c | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)
fs/debugfs/inode.c +3 -11
index acef14ad53db..d89874da9791 100644
@@ -617,13 +617,10 @@ struct dentry *debugfs_create_symlink(const char *name, struct dentry *parent,
 }
 EXPORT_SYMBOL_GPL(debugfs_create_symlink);
 
-static void __debugfs_remove_file(struct dentry *dentry, struct dentry *parent)
+static void __debugfs_file_removed(struct dentry *dentry)
 {
 	struct debugfs_fsdata *fsd;
 
-	simple_unlink(d_inode(parent), dentry);
-	d_delete(dentry);
-
 	/*
 	 * Paired with the closing smp_mb() implied by a successful
 	 * cmpxchg() in debugfs_file_get(): either
@@ -644,16 +641,15 @@ static int __debugfs_remove(struct dentry *dentry, struct dentry *parent)
 
 	if (simple_positive(dentry)) {
 		dget(dentry);
-		if (!d_is_reg(dentry)) {
-			if (d_is_dir(dentry))
-				ret = simple_rmdir(d_inode(parent), dentry);
-			else
-				simple_unlink(d_inode(parent), dentry);
-			if (!ret)
-				d_delete(dentry);
+		if (d_is_dir(dentry)) {
+			ret = simple_rmdir(d_inode(parent), dentry);
 		} else {

fs/debugfs/inode.c static int __debugfs_remove(struct dentry *dentry, struct dentry *parent) · line 655

Quoted 10 lines
fs/debugfs/inode.c +5 -1
-			__debugfs_remove_file(dentry, parent);
+			simple_unlink(d_inode(parent), dentry);
 		}
+		if (!ret)
+			d_delete(dentry);
+		if (d_is_reg(dentry))
+			__debugfs_file_removed(dentry);
 		dput(dentry);
 	}
 	return ret;

Ugh, I had to stare at this for a long time. I _think_ it all looks equalivant now, but if this breaks, I know who to blame :)

author reviewer
#18 lore raw

PATCH v3 7/10 debugfs: call fsnotify_{unlink,rmdir}() hooks

In reply to Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Greg Kroah-Hartman <[email protected]>
---
 fs/debugfs/inode.c | 3 +++
 1 file changed, 3 insertions(+)
fs/debugfs/inode.c +3 -0
index d89874da9791..1e444fe1f778 100644
@@ -643,8 +643,11 @@ static int __debugfs_remove(struct dentry *dentry, struct dentry *parent)
 		dget(dentry);
 		if (d_is_dir(dentry)) {
 			ret = simple_rmdir(d_inode(parent), dentry);
+			if (!ret)
+				fsnotify_rmdir(d_inode(parent), dentry);
 		} else {
 			simple_unlink(d_inode(parent), dentry);
+			fsnotify_unlink(d_inode(parent), dentry);
 		}
 		if (!ret)
 			d_delete(dentry);
Signature · 1 line
2.17.1
maintainer Reviewed-by on v3 7/10
#19 lore raw

Re: [PATCH v3 07/10] debugfs: call fsnotify_{unlink,rmdir}() hooks

Quoted 27 lines · Amir Goldstein

This will allow generating fsnotify delete events after the fsnotify_nameremove() hook is removed from d_delete().

Cc Greg Kroah-Hartman <[email protected]>
---
 fs/debugfs/inode.c | 3 +++
 1 file changed, 3 insertions(+)
fs/debugfs/inode.c +3 -0
index d89874da9791..1e444fe1f778 100644
@@ -643,8 +643,11 @@ static int __debugfs_remove(struct dentry *dentry, struct dentry *parent)
 		dget(dentry);
 		if (d_is_dir(dentry)) {
 			ret = simple_rmdir(d_inode(parent), dentry);
+			if (!ret)
+				fsnotify_rmdir(d_inode(parent), dentry);
 		} else {
 			simple_unlink(d_inode(parent), dentry);
+			fsnotify_unlink(d_inode(parent), dentry);
 		}
 		if (!ret)
 			d_delete(dentry);
Signature · 1 line
2.17.1
author reviewer
#20 lore raw

PATCH v3 8/10 configfs: call fsnotify_rmdir() hook

In reply to Amir Goldstein

This will allow generating fsnotify delete events on unregister of group/subsystem after the fsnotify_nameremove() hook is removed from d_delete().

The rest of the d_delete() calls from this filesystem are either called recursively from within debugfs_unregister_{group,subsystem}, called from a vfs function that already has delete hooks or are called from shutdown/cleanup code.

Cc Joel Becker <[email protected]>
Cc Christoph Hellwig <[email protected]>
---
 fs/configfs/dir.c | 3 +++
 1 file changed, 3 insertions(+)
fs/configfs/dir.c +3 -0
index 5e7932d668ab..ba17881a8d84 100644
@@ -27,6 +27,7 @@
 #undef DEBUG
 
 #include <linux/fs.h>
+#include <linux/fsnotify.h>
 #include <linux/mount.h>
 #include <linux/module.h>
 #include <linux/slab.h>
@@ -1804,6 +1805,7 @@ void configfs_unregister_group(struct config_group *group)
 	configfs_detach_group(&group->cg_item);
 	d_inode(dentry)->i_flags |= S_DEAD;
 	dont_mount(dentry);
+	fsnotify_rmdir(d_inode(parent), dentry);
 	d_delete(dentry);
 	inode_unlock(d_inode(parent));
 
@@ -1932,6 +1934,7 @@ void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
 	configfs_detach_group(&group->cg_item);
 	d_inode(dentry)->i_flags |= S_DEAD;
 	dont_mount(dentry);
+	fsnotify_rmdir(d_inode(root), dentry);
 	inode_unlock(d_inode(dentry));
 
 	d_delete(dentry);
Signature · 1 line
2.17.1
author reviewer
#21 lore raw

Re: [PATCH v3 08/10] configfs: call fsnotify_rmdir() hook

Quoted 5 lines · Amir Goldstein

This will allow generating fsnotify delete events on unregister of group/subsystem after the fsnotify_nameremove() hook is removed from d_delete().

Quoted 7 lines

The rest of the d_delete() calls from this filesystem are either called recursively from within debugfs_unregister_{group,subsystem}, called from a vfs function that already has delete hooks or are called from shutdown/cleanup code.

Cc Joel Becker <[email protected]>
Cc Christoph Hellwig <[email protected]>

Hi Christoph and Joel,

Per Christoph's request, I cc'ed you guys on the entire patch series for context,
so my discussion with Greg [1] about the special status of configfs in this patch set
should already be somewhere in your mailboxes...

Could I ask you to provide an ACK on this patch and on the chosen policy. To recap:
Before patch set:
1. User gets create/delete events when files/dirs created/removed via vfs_*()
2. User does *not* get create events when files/dirs created via debugfs_register_*()
3. User *does* get delete events when files/dirs removed via debugfs_unregister_*()

After patch set:
1. No change
2. No change
3. User will get delete events only on the root group/subsystem dir when tree is removed via debugfs_unregister_*()

For symmetry, we could also add create events for root group/subsystem dir when tree is created via debugfs_unregister_*(), but that would be a followup patch.
For users though, it may be that delete events are more important than create events
(i.e. for user cleanup tasks).

Thanks,
Amir.

[1] https://lore.kernel.org/linux-fsdevel/CAOQ4uxjyg5AVPrcR4bPm4zMY9BKmgV8g7TAuH--cfKNJv8pRYQ@mail.gmail.com/

Quoted 36 lines
---
 fs/configfs/dir.c | 3 +++
 1 file changed, 3 insertions(+)
fs/configfs/dir.c +3 -1
index 5e7932d668ab..ba17881a8d84 100644
@@ -27,6 +27,7 @@
 #undef DEBUG
 
 #include <linux/fs.h>
+#include <linux/fsnotify.h>
 #include <linux/mount.h>
 #include <linux/module.h>
 #include <linux/slab.h>
@@ -1804,6 +1805,7 @@ void configfs_unregister_group(struct config_group *group)
        configfs_detach_group(&group->cg_item);
        d_inode(dentry)->i_flags |= S_DEAD;
        dont_mount(dentry);
+       fsnotify_rmdir(d_inode(parent), dentry);
        d_delete(dentry);
        inode_unlock(d_inode(parent));
 
@@ -1932,6 +1934,7 @@ void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
        configfs_detach_group(&group->cg_item);
        d_inode(dentry)->i_flags |= S_DEAD;
        dont_mount(dentry);
+       fsnotify_rmdir(d_inode(root), dentry);
        inode_unlock(d_inode(dentry));
 
        d_delete(dentry);
--

2.17.1

author reviewer
#22 lore raw

Re: [PATCH v3 08/10] configfs: call fsnotify_rmdir() hook

Quoted 17 lines · Amir Goldstein
Quoted 5 lines · Amir Goldstein

This will allow generating fsnotify delete events on unregister of group/subsystem after the fsnotify_nameremove() hook is removed from d_delete().

Quoted 7 lines

The rest of the d_delete() calls from this filesystem are either called recursively from within debugfs_unregister_{group,subsystem}, called from a vfs function that already has delete hooks or are called from shutdown/cleanup code.

Cc Joel Becker <[email protected]>
Cc Christoph Hellwig <[email protected]>

Hi Christoph and Joel,

Quoted 7 lines

Per Christoph's request, I cc'ed you guys on the entire patch series for context,
so my discussion with Greg [1] about the special status of configfs in this patch set
should already be somewhere in your mailboxes...

Could I ask you to provide an ACK on this patch and on the chosen

PING.
Any comment from configfs maintainers?

Jan,

We can also drop this patch from the series given that configs has no explicit fsnotify create hooks. If users come shouting, we can add the missing create and delete hooks.

Thanks,
Amir.

Quoted 62 lines

policy. To recap:
Before patch set:
1. User gets create/delete events when files/dirs created/removed via vfs_*()
2. User does *not* get create events when files/dirs created via debugfs_register_*()
3. User *does* get delete events when files/dirs removed via debugfs_unregister_*()

After patch set:
1. No change
2. No change
3. User will get delete events only on the root group/subsystem dir when tree is removed via debugfs_unregister_*()

For symmetry, we could also add create events for root group/subsystem dir when tree is created via debugfs_unregister_*(), but that would be a followup patch.
For users though, it may be that delete events are more important than create events
(i.e. for user cleanup tasks).

Thanks,
Amir.

[1] https://lore.kernel.org/linux-fsdevel/CAOQ4uxjyg5AVPrcR4bPm4zMY9BKmgV8g7TAuH--cfKNJv8pRYQ@mail.gmail.com/

Quoted 36 lines
---
 fs/configfs/dir.c | 3 +++
 1 file changed, 3 insertions(+)
fs/configfs/dir.c +3 -1
index 5e7932d668ab..ba17881a8d84 100644
@@ -27,6 +27,7 @@
 #undef DEBUG
 
 #include <linux/fs.h>
+#include <linux/fsnotify.h>
 #include <linux/mount.h>
 #include <linux/module.h>
 #include <linux/slab.h>
@@ -1804,6 +1805,7 @@ void configfs_unregister_group(struct config_group *group)
        configfs_detach_group(&group->cg_item);
        d_inode(dentry)->i_flags |= S_DEAD;
        dont_mount(dentry);
+       fsnotify_rmdir(d_inode(parent), dentry);
        d_delete(dentry);
        inode_unlock(d_inode(parent));
 
@@ -1932,6 +1934,7 @@ void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
        configfs_detach_group(&group->cg_item);
        d_inode(dentry)->i_flags |= S_DEAD;
        dont_mount(dentry);
+       fsnotify_rmdir(d_inode(root), dentry);
        inode_unlock(d_inode(dentry));
 
        d_delete(dentry);
--

2.17.1

author reviewer
#23 lore raw

PATCH v3 9/10 fsnotify: move fsnotify_nameremove() hook out of d_delete()

In reply to Amir Goldstein

d_delete() was piggy backed for the fsnotify_nameremove() hook when in fact not all callers of d_delete() care about fsnotify events.

For all callers of d_delete() that may be interested in fsnotify events, we made sure to call one of fsnotify_{unlink,rmdir}() hooks before calling d_delete().

Now we can move the fsnotify_nameremove() call from d_delete() to the fsnotify_{unlink,rmdir}() hooks.

Two explicit calls to fsnotify_nameremove() from nfs/afs sillyrename are also removed. This will cause a change of behavior - nfs/afs will NOT generate an fsnotify delete event when renaming over a positive dentry. This change is desirable, because it is consistent with the behavior of all other filesystems.

---
 fs/afs/dir_silly.c       | 5 -----
 fs/dcache.c              | 2 --
 fs/nfs/unlink.c          | 6 ------
 include/linux/fsnotify.h | 2 ++
 4 files changed, 2 insertions(+), 13 deletions(-)
fs/afs/dir_silly.c +0 -5
index 28f4aa015229..312687223170 100644
@@ -64,11 +64,6 @@ static int afs_do_silly_rename(struct afs_vnode *dvnode, struct afs_vnode *vnode
 		if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
 			afs_edit_dir_add(dvnode, &new->d_name,
 					 &vnode->fid, afs_edit_dir_for_silly_1);
-
-		/* vfs_unlink and the like do not issue this when a file is
-		 * sillyrenamed, so do it here.
-		 */
-		fsnotify_nameremove(old, 0);
 	}
 
 	kfree(scb);
fs/dcache.c +0 -2
index 8136bda27a1f..ce131339410c 100644
@@ -2371,7 +2371,6 @@ EXPORT_SYMBOL(d_hash_and_lookup);
 void d_delete(struct dentry * dentry)
 {
 	struct inode *inode = dentry->d_inode;
-	int isdir = d_is_dir(dentry);
 
 	spin_lock(&inode->i_lock);
 	spin_lock(&dentry->d_lock);
@@ -2386,7 +2385,6 @@ void d_delete(struct dentry * dentry)
 		spin_unlock(&dentry->d_lock);
 		spin_unlock(&inode->i_lock);
 	}
-	fsnotify_nameremove(dentry, isdir);
 }
 EXPORT_SYMBOL(d_delete);
 
fs/nfs/unlink.c +0 -6
index 52d533967485..0effeee28352 100644
@@ -396,12 +396,6 @@ nfs_complete_sillyrename(struct rpc_task *task, struct nfs_renamedata *data)
 		nfs_cancel_async_unlink(dentry);
 		return;
 	}
-
-	/*
-	 * vfs_unlink and the like do not issue this when a file is
-	 * sillyrenamed, so do it here.
-	 */
-	fsnotify_nameremove(dentry, 0);
 }
 
 #define SILLYNAME_PREFIX ".nfs"
include/linux/fsnotify.h +2 -0
index 7f23eddefcd0..0145073c2b42 100644
@@ -199,6 +199,7 @@ static inline void fsnotify_unlink(struct inode *dir, struct dentry *dentry)
 	WARN_ON_ONCE(d_is_negative(dentry));
 
 	/* TODO: call fsnotify_dirent() */
+	fsnotify_nameremove(dentry, 0);
 }
 
 /*
@@ -222,6 +223,7 @@ static inline void fsnotify_rmdir(struct inode *dir, struct dentry *dentry)
 	WARN_ON_ONCE(d_is_negative(dentry));
 
 	/* TODO: call fsnotify_dirent() */
+	fsnotify_nameremove(dentry, 1);
 }
 
 /*
Signature · 1 line
2.17.1
author reviewer
#24 lore raw

PATCH v3 10/10 fsnotify: get rid of fsnotify_nameremove()

In reply to Amir Goldstein

For all callers of fsnotify_{unlink,rmdir}(), we made sure that d_parent and d_name are stable. Therefore, fsnotify_{unlink,rmdir}() do not need the safety measures in fsnotify_nameremove() to stabilize parent and name. We can now simplify those hooks and get rid of fsnotify_nameremove().

---
 fs/notify/fsnotify.c             | 41 --------------------------------
 include/linux/fsnotify.h         |  6 ++---
 include/linux/fsnotify_backend.h |  4 ----
 3 files changed, 2 insertions(+), 49 deletions(-)
fs/notify/fsnotify.c +0 -41
index 8c7cbac7183c..5433e37fb0c5 100644
@@ -107,47 +107,6 @@ void fsnotify_sb_delete(struct super_block *sb)
 	fsnotify_clear_marks_by_sb(sb);
 }
 
-/*
- * fsnotify_nameremove - a filename was removed from a directory
- *
- * This is mostly called under parent vfs inode lock so name and
- * dentry->d_parent should be stable. However there are some corner cases where
- * inode lock is not held. So to be on the safe side and be reselient to future
- * callers and out of tree users of d_delete(), we do not assume that d_parent
- * and d_name are stable and we use dget_parent() and
- * take_dentry_name_snapshot() to grab stable references.
- */
-void fsnotify_nameremove(struct dentry *dentry, int isdir)
-{
-	struct dentry *parent;
-	struct name_snapshot name;
-	__u32 mask = FS_DELETE;
-
-	/* d_delete() of pseudo inode? (e.g. __ns_get_path() playing tricks) */
-	if (IS_ROOT(dentry))
-		return;
-
-	if (isdir)
-		mask |= FS_ISDIR;
-
-	parent = dget_parent(dentry);
-	/* Avoid unneeded take_dentry_name_snapshot() */
-	if (!(d_inode(parent)->i_fsnotify_mask & FS_DELETE) &&
-	    !(dentry->d_sb->s_fsnotify_mask & FS_DELETE))
-		goto out_dput;
-
-	take_dentry_name_snapshot(&name, dentry);
-
-	fsnotify(d_inode(parent), mask, d_inode(dentry), FSNOTIFY_EVENT_INODE,
-		 &name.name, 0);
-
-	release_dentry_name_snapshot(&name);
-
-out_dput:
-	dput(parent);
-}
-EXPORT_SYMBOL(fsnotify_nameremove);
-
 /*
  * Given an inode, first check if we care what happens to our children.  Inotify
  * and dnotify both tell their parents about events.  If we care about any event
include/linux/fsnotify.h +2 -4
index 0145073c2b42..a2d5d175d3c1 100644
@@ -198,8 +198,7 @@ static inline void fsnotify_unlink(struct inode *dir, struct dentry *dentry)
 	/* Expected to be called before d_delete() */
 	WARN_ON_ONCE(d_is_negative(dentry));
 
-	/* TODO: call fsnotify_dirent() */
-	fsnotify_nameremove(dentry, 0);
+	fsnotify_dirent(dir, dentry, FS_DELETE);
 }
 
 /*
@@ -222,8 +221,7 @@ static inline void fsnotify_rmdir(struct inode *dir, struct dentry *dentry)
 	/* Expected to be called before d_delete() */
 	WARN_ON_ONCE(d_is_negative(dentry));
 
-	/* TODO: call fsnotify_dirent() */
-	fsnotify_nameremove(dentry, 1);
+	fsnotify_dirent(dir, dentry, FS_DELETE | FS_ISDIR);
 }
 
 /*
include/linux/fsnotify_backend.h +0 -4
index a9f9dcc1e515..c28f6ed1f59b 100644
@@ -355,7 +355,6 @@ extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u
 extern void __fsnotify_inode_delete(struct inode *inode);
 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
 extern void fsnotify_sb_delete(struct super_block *sb);
-extern void fsnotify_nameremove(struct dentry *dentry, int isdir);
 extern u32 fsnotify_get_cookie(void);
 
 static inline int fsnotify_inode_watches_children(struct inode *inode)
@@ -525,9 +524,6 @@ static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
 static inline void fsnotify_sb_delete(struct super_block *sb)
 {}
 
-static inline void fsnotify_nameremove(struct dentry *dentry, int isdir)
-{}
-
 static inline void fsnotify_update_flags(struct dentry *dentry)
 {}
 
Signature · 1 line
2.17.1
#25 lore raw
In reply to Amir Goldstein
Quoted 18 lines · Amir Goldstein

Jan,

For v3 I went with a straight forward approach. Filesystems that have fsnotify_{create,mkdir} hooks also get explicit fsnotify_{unlink,rmdir} hooks.

Hopefully, this approach is orthogonal to whatever changes Al is planning for recursive tree remove code, because in many of the cases, the hooks are added at the entry point for the recursive tree remove.

After looking closer at all the filesystems that were converted to simple_remove in v2, I decided to exempt another 3 filesystems from the fsnotify delete hooks: hypfs,qibfs and aafs.
hypfs is pure cleanup (*). qibfs and aafs can remove dentry on user configuration change, but they do not generate create events, so it is less likely that users depend on the delete events.

Quoted 2 lines

That leaves configfs the only filesystem that gets the new delete hooks even though it does not have create hooks.

why doesn't configfs have create hooks? That's what userspace does in configfs, shouldn't it be notified about it? Keeping it "unequal" seems odd to me.

thanks,

greg k-h

author reviewer
#26 lore raw
Quoted 20 lines · Greg Kroah-Hartman
Quoted 18 lines · Amir Goldstein

Jan,

For v3 I went with a straight forward approach. Filesystems that have fsnotify_{create,mkdir} hooks also get explicit fsnotify_{unlink,rmdir} hooks.

Hopefully, this approach is orthogonal to whatever changes Al is planning for recursive tree remove code, because in many of the cases, the hooks are added at the entry point for the recursive tree remove.

After looking closer at all the filesystems that were converted to simple_remove in v2, I decided to exempt another 3 filesystems from the fsnotify delete hooks: hypfs,qibfs and aafs.
hypfs is pure cleanup (*). qibfs and aafs can remove dentry on user configuration change, but they do not generate create events, so it is less likely that users depend on the delete events.

Quoted 7 lines
Quoted 2 lines

That leaves configfs the only filesystem that gets the new delete hooks even though it does not have create hooks.

why doesn't configfs have create hooks? That's what userspace does in configfs, shouldn't it be notified about it? Keeping it "unequal" seems odd to me.

So it's not exactly that configfs has no create hooks at all. For "normal" filesystems mkdir (for example) is only possible by mkdir(2) syscall and there is create hook in vfs_mkdir().

The configfs (as well as debugfs/tracefs/etc), there are other code paths that create directories, namely: configfs_register_grup/subsystem(). Those code paths have explicit fsnotify_mkdir() hook in debugfs/tracefs/etc, but not in configfs. Why? because nobody put the hooks and no user complained.

Should we add fsnotify_mkdir() hooks in configfs - probably yes. I can do it as followup, but this is not the purpose of this patch set. The purpose of this patch set (achieved in the last patch) is to simplify the implementation of the fsnotify delete hook.
Today it is overly complicated by the fact that the hooks was placed in d_delete() and d_delete() is called from some code paths that have no business with fsnotify notifications at all.

Once this patch set is done sprinkling fsnotify_rmdir/unlink() hooks in "proper" places, it removes the current fsnotify hook from d_delete(). d_delete() is called from configfs_unregister_group/subsystem(), so if we do not add fsnotify delete hooks to configfs we will regress the existing "unequal" behavior.
Maybe there are no users depending on fsnotify delete notifications from configfs - I do not know. If we believe that is the case, we can drop the configfs patch.
Maybe there *are* user depending on fsnotify delete notifications from aafs (apparmorfs), which I decided to exclude from v3 patch set.
If we believe that is the case, or if we find out later that in case - no problem
it is simple to add the missing fsnotify hooks in apparmorfs.

Thought? About inclusion of configfs? About exclusion of apparmorfs? Again this is only exclusion/inclusion of hooks from code path that does NOT come from vfs syscalls (e.g. aa_remove_profiles()).

Thanks,
Amir.

#27 lore raw
Quoted 25 lines · Amir Goldstein
Quoted 23 lines · Greg Kroah-Hartman
Quoted 20 lines · Amir Goldstein

Jan,

For v3 I went with a straight forward approach. Filesystems that have fsnotify_{create,mkdir} hooks also get explicit fsnotify_{unlink,rmdir} hooks.

Hopefully, this approach is orthogonal to whatever changes Al is planning for recursive tree remove code, because in many of the cases, the hooks are added at the entry point for the recursive tree remove.

After looking closer at all the filesystems that were converted to simple_remove in v2, I decided to exempt another 3 filesystems from the fsnotify delete hooks: hypfs,qibfs and aafs.
hypfs is pure cleanup (*). qibfs and aafs can remove dentry on user configuration change, but they do not generate create events, so it is less likely that users depend on the delete events.

That leaves configfs the only filesystem that gets the new delete hooks even though it does not have create hooks.

Quoted 8 lines
Quoted 4 lines

why doesn't configfs have create hooks? That's what userspace does in configfs, shouldn't it be notified about it? Keeping it "unequal" seems odd to me.

So it's not exactly that configfs has no create hooks at all. For "normal" filesystems mkdir (for example) is only possible by mkdir(2) syscall and there is create hook in vfs_mkdir().

Ah, ok, missed that.

thanks,

greg k-h

maintainer
#28 lore raw
In reply to Amir Goldstein

Hi Amir!

Quoted 16 lines · Amir Goldstein

For v3 I went with a straight forward approach. Filesystems that have fsnotify_{create,mkdir} hooks also get explicit fsnotify_{unlink,rmdir} hooks.

Hopefully, this approach is orthogonal to whatever changes Al is planning for recursive tree remove code, because in many of the cases, the hooks are added at the entry point for the recursive tree remove.

After looking closer at all the filesystems that were converted to simple_remove in v2, I decided to exempt another 3 filesystems from the fsnotify delete hooks: hypfs,qibfs and aafs.
hypfs is pure cleanup (*). qibfs and aafs can remove dentry on user configuration change, but they do not generate create events, so it is less likely that users depend on the delete events.

Quoted 2 lines

That leaves configfs the only filesystem that gets the new delete hooks even though it does not have create hooks.

OK, I went through the patches and they look OK to me. So once we get acks from respective maintainers I think we can merge them.

Honza

Quoted 84 lines

The following d_delete() call sites have been audited and will no longer generate fsnotify event after this series:

arch/s390/hypfs/inode.c:hypfs_remove() - cleanup (*) .../usb/gadget/function/f_fs.c:ffs_epfiles_destroy() - no create events .../infiniband/hw/qib/qib_fs.c:remove_device_files() - no create events fs/ceph/dir.c:ceph_unlink() - from vfs_unlink() fs/ceph/inode.c:ceph_fill_trace() - invalidate (**) fs/ceph/inode.c:ceph_readdir_prepopulate() - invalidate (**) fs/configfs/dir.c:detach_groups() - hooks added, from vfs or cleanup (*) fs/configfs/dir.c:configfs_attach_item() - cleanup (*) fs/configfs/dir.c:configfs_attach_group() - cleanup (*) fs/efivarfs/file.c:efivarfs_file_write() - invalidate (**) fs/fuse/dir.c:fuse_reverse_inval_entry() - invalidate (**) fs/nfs/dir.c:nfs_dentry_handle_enoent() - invalidate (**) fs/nsfs.c:__ns_get_path() - cleanup (*) fs/ocfs2/dlmglue.c:ocfs2_dentry_convert_worker() - invalidate (**) fs/reiserfs/xattr.c:xattr_{unlink,rmdir}() - hidden xattr inode security/apparmor/apparmorfs.c:aafs_remove() - no create events

(*) There are 2 "cleanup" use cases:
- Create;init;delete if init failed
- Batch delete of files within dir before removing dir Both those cases are not interesting for users that wish to observe configuration changes on pseudo filesystems. Often, there is already an fsnotify event generated on the directory removal which is what users should find interesting, for example:
configfs_unregister_{group,subsystem}().

(**) The different "invalidate" use cases differ, but they all share
one thing in common - user is not guarantied to get an event with current kernel. For example, when a file is deleted remotely on nfs server, nfs client is not guarantied to get an fsnotify delete event. On current kernel, nfs client could generate an fsnotify delete event if the local entry happens to be in cache and client finds out that entry is deleted on server during another user operation. Incidentally, this group of use cases is where most of the call sites are with "unstable" d_name, which is the reason for this patch series to begin with.

Thanks,
Amir.

Changes since v2:
- Drop simple_rename() conversions (add explicit hooks instead)
- Drop hooks from hypfs/qibfs/aafs
- Split out debugfs re-factoring patch

Changes since v1:
- Split up per filesystem patches
- New hook names fsnotify_{unlink,rmdir}()
- Simplify fsnotify code in separate final patch

Amir Goldstein (10):
fsnotify: add empty fsnotify_{unlink,rmdir}() hooks
btrfs: call fsnotify_rmdir() hook
rpc_pipefs: call fsnotify_{unlink,rmdir}() hooks
tracefs: call fsnotify_{unlink,rmdir}() hooks
devpts: call fsnotify_unlink() hook
debugfs: simplify __debugfs_remove_file()
debugfs: call fsnotify_{unlink,rmdir}() hooks
configfs: call fsnotify_rmdir() hook
fsnotify: move fsnotify_nameremove() hook out of d_delete()
fsnotify: get rid of fsnotify_nameremove()

 fs/afs/dir_silly.c               |  5 ----
 fs/btrfs/ioctl.c                 |  4 +++-
 fs/configfs/dir.c                |  3 +++
 fs/dcache.c                      |  2 --
 fs/debugfs/inode.c               | 21 ++++++++--------
 fs/devpts/inode.c                |  1 +
 fs/namei.c                       |  2 ++
 fs/nfs/unlink.c                  |  6 -----
 fs/notify/fsnotify.c             | 41 --------------------------------
 fs/tracefs/inode.c               |  3 +++
 include/linux/fsnotify.h         | 26 ++++++++++++++++++++
 include/linux/fsnotify_backend.h |  4 ----
 net/sunrpc/rpc_pipe.c            |  4 ++++
 13 files changed, 52 insertions(+), 70 deletions(-)
Signature · 1 line
2.17.1
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR