From: Amir Goldstein <amir73il@gmail.com>
To: Jan Kara <jack@suse.cz>
Cc: Matthew Bobrowski <mbobrowski@mbobrowski.org>,
linux-fsdevel <linux-fsdevel@vger.kernel.org>
Subject: Re: [PATCH v4 05/10] fsnotify: send MOVE_SELF event with parent/name info
Date: Tue, 14 Jul 2020 15:44:16 +0300 [thread overview]
Message-ID: <CAOQ4uxhs2Hc-qr9or2tYLgPkoB-tQ4w-pr0qPSYa8qrvcD3rVQ@mail.gmail.com> (raw)
In-Reply-To: <20200714121329.GF23073@quack2.suse.cz>
On Tue, Jul 14, 2020 at 3:13 PM Jan Kara <jack@suse.cz> wrote:
>
> On Thu 02-07-20 15:57:39, Amir Goldstein wrote:
> > MOVE_SELF event does not get reported to a parent watching children
> > when a child is moved, but it can be reported to sb/mount mark with
> > parent/name info if group is interested in parent/name info.
> >
> > Use the fsnotify_parent() helper to send a MOVE_SELF event and adjust
> > fsnotify() to handle the case of an event "on child" that should not
> > be sent to the watching parent's inode mark.
> >
> > Signed-off-by: Amir Goldstein <amir73il@gmail.com>
> > ---
> > fs/notify/fsnotify.c | 21 +++++++++++++++++----
> > include/linux/fsnotify.h | 5 +----
> > include/linux/fsnotify_backend.h | 2 +-
> > 3 files changed, 19 insertions(+), 9 deletions(-)
> >
> > diff --git a/fs/notify/fsnotify.c b/fs/notify/fsnotify.c
> > index 6683c77a5b13..0faf5b09a73e 100644
> > --- a/fs/notify/fsnotify.c
> > +++ b/fs/notify/fsnotify.c
> > @@ -352,6 +352,7 @@ int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_type,
> > struct super_block *sb = to_tell->i_sb;
> > struct inode *dir = S_ISDIR(to_tell->i_mode) ? to_tell : NULL;
> > struct mount *mnt = NULL;
> > + struct inode *inode = NULL;
> > struct inode *child = NULL;
> > int ret = 0;
> > __u32 test_mask, marks_mask;
> > @@ -362,6 +363,14 @@ int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_type,
> > if (mask & FS_EVENT_ON_CHILD)
> > child = fsnotify_data_inode(data, data_type);
> >
> > + /*
> > + * If event is "on child" then to_tell is a watching parent.
> > + * An event "on child" may be sent to mount/sb mark with parent/name
> > + * info, but not appropriate for watching parent (e.g. FS_MOVE_SELF).
> > + */
> > + if (!child || (mask & FS_EVENTS_POSS_ON_CHILD))
> > + inode = to_tell;
>
> I'm now confused. Don't you want to fill in FSNOTIFY_OBJ_TYPE_INODE below
> for FS_MOVE_SELF event? But this condition is false for it so you won't do
> it?
>
I do not.
For events with the flag FS_EVENT_ON_CHILD, the inode in
FSNOTIFY_OBJ_TYPE_INODE is always the parent and the inode in
FSNOTIFY_OBJ_TYPE_CHILD is always the child.
So FS_MOVE_SELF will be reported if sb/mount are watching
or if child inode is watching, but NOT if only parent inode is watching.
I realize I may have been able to make other choices, but seemed like
the most consistent choice to me.
If you see a better option, let me know.
> > +
> > /*
> > * Optimization: srcu_read_lock() has a memory barrier which can
> > * be expensive. It protects walking the *_fsnotify_marks lists.
> > @@ -369,14 +378,17 @@ int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_type,
> > * SRCU because we have no references to any objects and do not
> > * need SRCU to keep them "alive".
> > */
> > - if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks &&
> > + if (!sb->s_fsnotify_marks &&
> > (!mnt || !mnt->mnt_fsnotify_marks) &&
> > + (!inode || !inode->i_fsnotify_marks) &&
> > (!child || !child->i_fsnotify_marks))
> > return 0;
> >
> > - marks_mask = to_tell->i_fsnotify_mask | sb->s_fsnotify_mask;
> > + marks_mask = sb->s_fsnotify_mask;
> > if (mnt)
> > marks_mask |= mnt->mnt_fsnotify_mask;
> > + if (inode)
> > + marks_mask |= inode->i_fsnotify_mask;
> > if (child)
> > marks_mask |= child->i_fsnotify_mask;
> >
> > @@ -390,14 +402,15 @@ int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_type,
> >
> > iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
> >
> > - iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
> > - fsnotify_first_mark(&to_tell->i_fsnotify_marks);
> > iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
> > fsnotify_first_mark(&sb->s_fsnotify_marks);
> > if (mnt) {
> > iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
> > fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
> > }
> > + if (inode)
> > + iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
> > + fsnotify_first_mark(&inode->i_fsnotify_marks);
> > if (child) {
> > iter_info.marks[FSNOTIFY_OBJ_TYPE_CHILD] =
> > fsnotify_first_mark(&child->i_fsnotify_marks);
> > diff --git a/include/linux/fsnotify.h b/include/linux/fsnotify.h
> > index 044cae3a0628..61dccaf21e7b 100644
> > --- a/include/linux/fsnotify.h
> > +++ b/include/linux/fsnotify.h
> > @@ -131,7 +131,6 @@ static inline void fsnotify_move(struct inode *old_dir, struct inode *new_dir,
> > u32 fs_cookie = fsnotify_get_cookie();
> > __u32 old_dir_mask = FS_MOVED_FROM;
> > __u32 new_dir_mask = FS_MOVED_TO;
> > - __u32 mask = FS_MOVE_SELF;
> > const struct qstr *new_name = &moved->d_name;
> >
> > if (old_dir == new_dir)
> > @@ -140,7 +139,6 @@ static inline void fsnotify_move(struct inode *old_dir, struct inode *new_dir,
> > if (isdir) {
> > old_dir_mask |= FS_ISDIR;
> > new_dir_mask |= FS_ISDIR;
> > - mask |= FS_ISDIR;
> > }
> >
> > fsnotify_name(old_dir, old_dir_mask, source, old_name, fs_cookie);
> > @@ -149,8 +147,7 @@ static inline void fsnotify_move(struct inode *old_dir, struct inode *new_dir,
> > if (target)
> > fsnotify_link_count(target);
> >
> > - if (source)
> > - fsnotify(source, mask, source, FSNOTIFY_EVENT_INODE, NULL, 0);
> > + fsnotify_dentry(moved, FS_MOVE_SELF);
>
> I'm somewhat unsure about this. Does this mean that 'moved' is guaranteed
> to be positive or that you've made sure that all the code below
> fsnotify_dentry() is actually fine with a negative dentry? I don't find
> either trivial to verify so some note in a changelog or maybe even a
> separate patch for this would be useful.
>
Oh, it's true. I should've mentioned it or separate this change.
I guess my reaction was the opposite of yours - it seemed obvious to
me that moved
dentry is positive - vfs_rename() is called under lock_rename() and it seemed
obvious to me that callers verified positive source, but in any case,
vfs_rename()
starts with may_delete() that verifies positive rename victim and
debugfs_rename(), the other caller of fsnotify_move() also verifies
positive victim.
I will add this information to the commit message.
Thanks,
Amir.
next prev parent reply other threads:[~2020-07-14 12:44 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-07-02 12:57 [PATCH v4 00/10] fanotify events with name info Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 01/10] inotify: report both events on parent and child with single callback Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 02/10] fanotify: " Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 03/10] fsnotify: send event to " Amir Goldstein
2020-07-14 10:34 ` Jan Kara
2020-07-14 11:54 ` Amir Goldstein
2020-07-15 17:09 ` Jan Kara
2020-07-15 17:42 ` Amir Goldstein
2020-07-16 6:38 ` Amir Goldstein
2020-07-16 7:39 ` Amir Goldstein
2020-07-16 9:55 ` Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 04/10] fsnotify: send event with parent/name info to sb/mount/non-dir marks Amir Goldstein
2020-07-14 11:54 ` Jan Kara
2020-07-14 12:17 ` Amir Goldstein
2020-07-14 15:31 ` Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 05/10] fsnotify: send MOVE_SELF event with parent/name info Amir Goldstein
2020-07-14 12:13 ` Jan Kara
2020-07-14 12:44 ` Amir Goldstein [this message]
2020-07-02 12:57 ` [PATCH v4 06/10] fanotify: add basic support for FAN_REPORT_DIR_FID Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 07/10] fanotify: report events with parent dir fid to sb/mount/non-dir marks Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 08/10] fanotify: add support for FAN_REPORT_NAME Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 09/10] fanotify: report parent fid + name + child fid Amir Goldstein
2020-07-02 12:57 ` [PATCH v4 10/10] fanotify: report parent fid " Amir Goldstein
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAOQ4uxhs2Hc-qr9or2tYLgPkoB-tQ4w-pr0qPSYa8qrvcD3rVQ@mail.gmail.com \
--to=amir73il@gmail.com \
--cc=jack@suse.cz \
--cc=linux-fsdevel@vger.kernel.org \
--cc=mbobrowski@mbobrowski.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).