mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Christian Brauner <brauner@kernel.org>
To: Mateusz Guzik <mjguzik@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>,
	viro@zeniv.linux.org.uk, linux-fsdevel@vger.kernel.org,
	linux-kernel@vger.kernel.org, oleg@redhat.com,
	Matthew Wilcox <willy@infradead.org>,
	Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [PATCH v2 (kindof)] fs: use __fput_sync in close(2)
Date: Tue, 8 Aug 2023 19:00:13 +0200	[thread overview]
Message-ID: <20230808-urgestein-kurstadt-6333ccf292eb@brauner> (raw)
In-Reply-To: <20230808-lebst-vorgibt-75c3010b4e54@brauner>

On Tue, Aug 08, 2023 at 06:30:06PM +0200, Christian Brauner wrote:
> On Tue, Aug 08, 2023 at 05:07:22PM +0200, Mateusz Guzik wrote:
> > I slapped the following variant just for illustration purposes.
> > 
> > - adds __close_fd which returns a struct file
> > - adds __filp_close with a flag whether to fput
> > - makes close(2) use both
> > - transparent to everyone else
> > 
> > Downside is that __fput_sync still loses the assert. Instead of
> > losing, it could perhaps be extended with a hack to check syscall
> > number -- pass if either this is close (or binary compat close) or a
> > kthread, BUG out otherwise. Alternatively perhaps deref could be
> > opencoded along with a comment about real fput that this is taking
> > place. Or maybe some other cosmetic choice.
> > 
> > I cannot compile-test right now, so down below is a rough copy make
> > sure it is clear what I mean.
> > 
> > I feel compelled to note that simple patches get microbenchmarked all
> > the time, with these results being the only justification provided.
> > I'm confused why this patch is supposed to be an exception given its
> > simplicity.
> > 
> > Serious justification should be expected from tough calls --
> > complicated, invasive changes, maybe with numerous tradeoffs.
> > 
> > In contrast close(2) doing __fput_sync looks a clear cut thing to do,
> > at worst one can argue which way to do it.
> > 
> > diff --git a/fs/file.c b/fs/file.c
> > index 3fd003a8604f..c341b07533b0 100644
> > --- a/fs/file.c
> > +++ b/fs/file.c
> > @@ -651,20 +651,30 @@ static struct file *pick_file(struct
> > files_struct *files, unsigned fd)
> >         return file;
> >  }
> > 
> > -int close_fd(unsigned fd)
> > +struct file *__close_fd(unsigned fd, struct file_struct *files)
> >  {
> > -       struct files_struct *files = current->files;
> >         struct file *file;
> > 
> >         spin_lock(&files->file_lock);
> >         file = pick_file(files, fd);
> >         spin_unlock(&files->file_lock);
> > +
> > +       return file;
> > +}
> > +EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
> > +
> > +int close_fd(unsigned fd)
> > +{
> > +       struct files_struct *files = current->files;
> > +       struct file *file;
> > +
> > +       file = __close_fd(fd, files);
> >         if (!file)
> >                 return -EBADF;
> > 
> >         return filp_close(file, files);
> >  }
> > -EXPORT_SYMBOL(close_fd); /* for ksys_close() */
> > +EXPORT_SYMBOL(close_fd);
> > 
> >  /**
> >   * last_fd - return last valid index into fd table
> > diff --git a/fs/file_table.c b/fs/file_table.c
> > index fc7d677ff5ad..b7461f0b73f4 100644
> > --- a/fs/file_table.c
> > +++ b/fs/file_table.c
> > @@ -463,6 +463,11 @@ void __fput_sync(struct file *file)
> >  {
> >         if (atomic_long_dec_and_test(&file->f_count)) {
> >                 struct task_struct *task = current;
> > +               /*
> > +                * I see 2 basic options
> > +                * 1. just remove the assert
> > +                * 2. demand the flag *or* that the caller is close(2)
> > +                */
> >                 BUG_ON(!(task->flags & PF_KTHREAD));
> >                 __fput(file);
> >         }
> > diff --git a/fs/open.c b/fs/open.c
> > index e6ead0f19964..b1602307c1c3 100644
> > --- a/fs/open.c
> > +++ b/fs/open.c
> > @@ -1533,7 +1533,16 @@ EXPORT_SYMBOL(filp_close);
> >   */
> >  SYSCALL_DEFINE1(close, unsigned int, fd)
> >  {
> > -       int retval = close_fd(fd);
> > +       struct files_struct *files = current->files;
> > +       struct file *file;
> > +       int retval;
> > +
> > +       file = __close_fd(fd);
> > +       if (!file)
> > +               return -EBADF;
> > +
> > +       retval = __filp_close(file, files, false);
> > +       __fput_sync(file);
> > 
> >         /* can't restart close syscall because file table entry was cleared */
> >         if (unlikely(retval == -ERESTARTSYS ||
> > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > index 562f2623c9c9..e64c0238a65f 100644
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -2388,7 +2388,11 @@ static inline struct file
> > *file_clone_open(struct file *file)
> >  {
> >         return dentry_open(&file->f_path, file->f_flags, file->f_cred);
> >  }
> > -extern int filp_close(struct file *, fl_owner_t id);
> > +extern int __filp_close(struct file *file, fl_owner_t id, bool dofput);
> > +static inline int filp_close(struct file *file, fl_owner_t id)
> > +{
> > +       return __filp_close(file, id, true);
> > +}
> > 
> >  extern struct filename *getname_flags(const char __user *, int, int *);
> >  extern struct filename *getname_uflags(const char __user *, int);
> 
> At least make this really dumb and obvious and keep the ugliness to
> internal.h and open.c

Sorry, I only sent a portion of the patch...

  reply	other threads:[~2023-08-08 18:03 UTC|newest]

Thread overview: 23+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-08-06 23:06 [PATCH] " Mateusz Guzik
2023-08-07  3:18 ` Matthew Wilcox
2023-08-08  5:56 ` Eric W. Biederman
2023-08-08  7:32   ` Mateusz Guzik
2023-08-08  8:13   ` Christian Brauner
2023-08-08  8:23     ` Mateusz Guzik
2023-08-08  8:40       ` Christian Brauner
2023-08-08  9:21         ` Mateusz Guzik
2023-08-08 15:07           ` [PATCH v2 (kindof)] " Mateusz Guzik
2023-08-08 16:30             ` Christian Brauner
2023-08-08 17:00               ` Christian Brauner [this message]
2023-08-08 17:05               ` Linus Torvalds
2023-08-08 17:06                 ` Christian Brauner
2023-08-09  9:03                 ` David Laight
2023-08-08 16:57   ` [PATCH] " Linus Torvalds
2023-08-08 17:10     ` Mateusz Guzik
2023-08-08 17:18       ` Linus Torvalds
2023-08-08 17:24         ` Mateusz Guzik
2023-08-08 17:35           ` Christian Brauner
2023-08-08 17:48             ` Linus Torvalds
2023-08-08 17:15     ` Christian Brauner
2023-08-08 17:22       ` Linus Torvalds
2023-08-08 17:48         ` Eric W. Biederman

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=20230808-urgestein-kurstadt-6333ccf292eb@brauner \
    --to=brauner@kernel.org \
    --cc=ebiederm@xmission.com \
    --cc=linux-fsdevel@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mjguzik@gmail.com \
    --cc=oleg@redhat.com \
    --cc=torvalds@linux-foundation.org \
    --cc=viro@zeniv.linux.org.uk \
    --cc=willy@infradead.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

all inboxes | Powered by JetHome®