Re: [PATCH 1/2] open: add close_range()

[Christian Brauner]

On Tue, May 21, 2019 at 10:23 PM Linus Torvalds
 wrote:
>
> On Tue, May 21, 2019 at 9:41 AM Christian Brauner  
> wrote:
> >
> > Yeah, you mentioned this before. I do like being able to specify an
> > upper bound to have the ability to place fds strategically after said
> > upper bound.
>
> I suspect that's the case.
>
> And if somebody really wants to just close everything and uses a large
> upper bound, we can - if we really want to - just compare the upper
> bound to the file table size, and do an optimized case for that. We do
> that upper bound comparison anyway to limit the size of the walk, so
> *if* it's a big deal, that case could then do the whole "shrink
> fdtable" case too.

Makes sense.

>
> But I don't believe it's worth optimizing for unless somebody really
> has a load where that is shown to be a big deal.   Just do the silly
> and simple loop, and add a cond_resched() in the loop, like
> close_files() does for the "we have a _lot_ of files open" case.

Ok. I will resend a v1 later with the cond_resched() logic you and Al
suggested added.

Thanks!
Christian

Re: [PATCH 1/2] open: add close_range()

[Linus Torvalds]

On Tue, May 21, 2019 at 9:41 AM Christian Brauner  wrote:
>
> Yeah, you mentioned this before. I do like being able to specify an
> upper bound to have the ability to place fds strategically after said
> upper bound.

I suspect that's the case.

And if somebody really wants to just close everything and uses a large
upper bound, we can - if we really want to - just compare the upper
bound to the file table size, and do an optimized case for that. We do
that upper bound comparison anyway to limit the size of the walk, so
*if* it's a big deal, that case could then do the whole "shrink
fdtable" case too.

But I don't believe it's worth optimizing for unless somebody really
has a load where that is shown to be a big deal.   Just do the silly
and simple loop, and add a cond_resched() in the loop, like
close_files() does for the "we have a _lot_ of files open" case.

   Linus

Re: [PATCH 1/2] open: add close_range()

[Matthew Wilcox]

On Tue, May 21, 2019 at 08:20:09PM +0100, Al Viro wrote:
> On Tue, May 21, 2019 at 05:30:27PM +0100, David Howells wrote:
> 
> > If we can live with close_from(int first) rather than close_range(), then 
> > this
> > can perhaps be done a lot more efficiently by:
> > 
> > new = alloc_fdtable(first);
> > spin_lock(>file_lock);
> > old = files_fdtable(files);
> > copy_fds(new, old, 0, first - 1);
> > rcu_assign_pointer(files->fdt, new);
> > spin_unlock(>file_lock);
> > clear_fds(old, 0, first - 1);
> > close_fdt_from(old, first);
> > kfree_rcu(old);
> 
> I really hate to think how that would interact with POSIX locks...

POSIX locks store current->files in fl_owner; David's resizing the
underlying files->fdt, just like growing from 64 to 256 fds.

Re: [PATCH 1/2] open: add close_range()

[Al Viro]

On Tue, May 21, 2019 at 05:30:27PM +0100, David Howells wrote:

> If we can live with close_from(int first) rather than close_range(), then this
> can perhaps be done a lot more efficiently by:
> 
>   new = alloc_fdtable(first);
>   spin_lock(>file_lock);
>   old = files_fdtable(files);
>   copy_fds(new, old, 0, first - 1);
>   rcu_assign_pointer(files->fdt, new);
>   spin_unlock(>file_lock);
>   clear_fds(old, 0, first - 1);
>   close_fdt_from(old, first);
>   kfree_rcu(old);

I really hate to think how that would interact with POSIX locks...

Re: [PATCH 1/2] open: add close_range()

[Christian Brauner]

On Tue, May 21, 2019 at 04:00:06PM +0100, Al Viro wrote:
> On Tue, May 21, 2019 at 01:34:47PM +0200, Christian Brauner wrote:
> 
> > This adds the close_range() syscall. It allows to efficiently close a range
> > of file descriptors up to all file descriptors of a calling task.
> > 
> > The syscall came up in a recent discussion around the new mount API and
> > making new file descriptor types cloexec by default. During this
> > discussion, Al suggested the close_range() syscall (cf. [1]). Note, a
> > syscall in this manner has been requested by various people over time.
> > 
> > First, it helps to close all file descriptors of an exec()ing task. This
> > can be done safely via (quoting Al's example from [1] verbatim):
> > 
> > /* that exec is sensitive */
> > unshare(CLONE_FILES);
> > /* we don't want anything past stderr here */
> > close_range(3, ~0U);
> > execve();
> > 
> > The code snippet above is one way of working around the problem that file
> > descriptors are not cloexec by default. This is aggravated by the fact that
> > we can't just switch them over without massively regressing userspace. For
> > a whole class of programs having an in-kernel method of closing all file
> > descriptors is very helpful (e.g. demons, service managers, programming
> > language standard libraries, container managers etc.).
> > (Please note, unshare(CLONE_FILES) should only be needed if the calling
> >  task is multi-threaded and shares the file descriptor table with another
> >  thread in which case two threads could race with one thread allocating
> >  file descriptors and the other one closing them via close_range(). For the
> >  general case close_range() before the execve() is sufficient.)
> > 
> > Second, it allows userspace to avoid implementing closing all file
> > descriptors by parsing through /proc//fd/* and calling close() on each
> > file descriptor. From looking at various large(ish) userspace code bases
> > this or similar patterns are very common in:
> > - service managers (cf. [4])
> > - libcs (cf. [6])
> > - container runtimes (cf. [5])
> > - programming language runtimes/standard libraries
> >   - Python (cf. [2])
> >   - Rust (cf. [7], [8])
> > As Dmitry pointed out there's even a long-standing glibc bug about missing
> > kernel support for this task (cf. [3]).
> > In addition, the syscall will also work for tasks that do not have procfs
> > mounted and on kernels that do not have procfs support compiled in. In such
> > situations the only way to make sure that all file descriptors are closed
> > is to call close() on each file descriptor up to UINT_MAX or RLIMIT_NOFILE,
> > OPEN_MAX trickery (cf. comment [8] on Rust).
> > 
> > The performance is striking. For good measure, comparing the following
> > simple close_all_fds() userspace implementation that is essentially just
> > glibc's version in [6]:
> > 
> > static int close_all_fds(void)
> > {
> > DIR *dir;
> > struct dirent *direntp;
> > 
> > dir = opendir("/proc/self/fd");
> > if (!dir)
> > return -1;
> > 
> > while ((direntp = readdir(dir))) {
> > int fd;
> > if (strcmp(direntp->d_name, ".") == 0)
> > continue;
> > if (strcmp(direntp->d_name, "..") == 0)
> > continue;
> > fd = atoi(direntp->d_name);
> > if (fd == 0 || fd == 1 || fd == 2)
> > continue;
> > close(fd);
> > }
> > 
> > closedir(dir); /* cannot fail */
> > return 0;
> > }
> > 
> > to close_range() yields:
> > 1. closing 4 open files:
> >- close_all_fds(): ~280 us
> >- close_range():~24 us
> > 
> > 2. closing 1000 open files:
> >- close_all_fds(): ~5000 us
> >- close_range():   ~800 us
> > 
> > close_range() is designed to allow for some flexibility. Specifically, it
> > does not simply always close all open file descriptors of a task. Instead,
> > callers can specify an upper bound.
> > This is e.g. useful for scenarios where specific file descriptors are
> > created with well-known numbers that are supposed to be excluded from
> > getting closed.
> > For extra paranoia close_range() comes with a flags argument. This can e.g.
> > be used to implement extension. Once can imagine userspace wanting to stop
> > at the first error instead of ignoring errors under certain circumstances.
> > There might be other valid ideas in the future. In any case, a flag
> > argument doesn't hurt and keeps us on the safe side.
> > 
> > >From an implementation side this is kept rather dumb. It saw some input
> > from David and Jann but all nonsense is obviously my own!
> > - Errors to close file descriptors are currently ignored. (Could be changed
> >   by setting a flag in the future if needed.)
> > - __close_range() is a rather simplistic wrapper around __close_fd().
> >   My reasoning behind 

Re: [PATCH 1/2] open: add close_range()

[Christian Brauner]

On Tue, May 21, 2019 at 05:30:27PM +0100, David Howells wrote:
> Al Viro  wrote:
> 
> > Umm...  That's going to be very painful if you dup2() something to MAX_INT 
> > and
> > then run that; roughly 2G iterations of bouncing ->file_lock up and down,
> > without anything that would yield CPU in process.
> > 
> > If anything, I would suggest something like
> > 
> > fd = *start_fd;
> > grab the lock
> > fdt = files_fdtable(files);
> > more:
> > look for the next eviction candidate in ->open_fds, starting at fd
> > if there's none up to max_fd
> > drop the lock
> > return NULL
> > *start_fd = fd + 1;
> > if the fscker is really opened and not just reserved
> > rcu_assign_pointer(fdt->fd[fd], NULL);
> > __put_unused_fd(files, fd);
> > drop the lock
> > return the file we'd got
> > if (unlikely(need_resched()))
> > drop lock
> > cond_resched();
> > grab lock
> > fdt = files_fdtable(files);
> > goto more;
> > 
> > with the main loop being basically
> > while ((file = pick_next(files, _fd, max_fd)) != NULL)
> > filp_close(file, files);
> 
> If we can live with close_from(int first) rather than close_range(), then this
> can perhaps be done a lot more efficiently by:

Yeah, you mentioned this before. I do like being able to specify an
upper bound to have the ability to place fds strategically after said
upper bound.
I have used this quite a few times where I know that given task may have
inherited up to m fds and I want to inherit a specific pipe who's fd I
know. Then I'd dup2(pipe_fd, ) and then close all
other fds. Is that too much of a corner case?

Christian

Re: [PATCH 1/2] open: add close_range()

[David Howells]

Al Viro  wrote:

> Umm...  That's going to be very painful if you dup2() something to MAX_INT and
> then run that; roughly 2G iterations of bouncing ->file_lock up and down,
> without anything that would yield CPU in process.
> 
> If anything, I would suggest something like
> 
>   fd = *start_fd;
>   grab the lock
> fdt = files_fdtable(files);
> more:
>   look for the next eviction candidate in ->open_fds, starting at fd
>   if there's none up to max_fd
>   drop the lock
>   return NULL
>   *start_fd = fd + 1;
>   if the fscker is really opened and not just reserved
>   rcu_assign_pointer(fdt->fd[fd], NULL);
>   __put_unused_fd(files, fd);
>   drop the lock
>   return the file we'd got
>   if (unlikely(need_resched()))
>   drop lock
>   cond_resched();
>   grab lock
>   fdt = files_fdtable(files);
>   goto more;
> 
> with the main loop being basically
>   while ((file = pick_next(files, _fd, max_fd)) != NULL)
>   filp_close(file, files);

If we can live with close_from(int first) rather than close_range(), then this
can perhaps be done a lot more efficiently by:

new = alloc_fdtable(first);
spin_lock(>file_lock);
old = files_fdtable(files);
copy_fds(new, old, 0, first - 1);
rcu_assign_pointer(files->fdt, new);
spin_unlock(>file_lock);
clear_fds(old, 0, first - 1);
close_fdt_from(old, first);
kfree_rcu(old);

David

Re: [PATCH 1/2] open: add close_range()

[Al Viro]

On Tue, May 21, 2019 at 01:34:47PM +0200, Christian Brauner wrote:

> This adds the close_range() syscall. It allows to efficiently close a range
> of file descriptors up to all file descriptors of a calling task.
> 
> The syscall came up in a recent discussion around the new mount API and
> making new file descriptor types cloexec by default. During this
> discussion, Al suggested the close_range() syscall (cf. [1]). Note, a
> syscall in this manner has been requested by various people over time.
> 
> First, it helps to close all file descriptors of an exec()ing task. This
> can be done safely via (quoting Al's example from [1] verbatim):
> 
> /* that exec is sensitive */
> unshare(CLONE_FILES);
> /* we don't want anything past stderr here */
> close_range(3, ~0U);
> execve();
> 
> The code snippet above is one way of working around the problem that file
> descriptors are not cloexec by default. This is aggravated by the fact that
> we can't just switch them over without massively regressing userspace. For
> a whole class of programs having an in-kernel method of closing all file
> descriptors is very helpful (e.g. demons, service managers, programming
> language standard libraries, container managers etc.).
> (Please note, unshare(CLONE_FILES) should only be needed if the calling
>  task is multi-threaded and shares the file descriptor table with another
>  thread in which case two threads could race with one thread allocating
>  file descriptors and the other one closing them via close_range(). For the
>  general case close_range() before the execve() is sufficient.)
> 
> Second, it allows userspace to avoid implementing closing all file
> descriptors by parsing through /proc//fd/* and calling close() on each
> file descriptor. From looking at various large(ish) userspace code bases
> this or similar patterns are very common in:
> - service managers (cf. [4])
> - libcs (cf. [6])
> - container runtimes (cf. [5])
> - programming language runtimes/standard libraries
>   - Python (cf. [2])
>   - Rust (cf. [7], [8])
> As Dmitry pointed out there's even a long-standing glibc bug about missing
> kernel support for this task (cf. [3]).
> In addition, the syscall will also work for tasks that do not have procfs
> mounted and on kernels that do not have procfs support compiled in. In such
> situations the only way to make sure that all file descriptors are closed
> is to call close() on each file descriptor up to UINT_MAX or RLIMIT_NOFILE,
> OPEN_MAX trickery (cf. comment [8] on Rust).
> 
> The performance is striking. For good measure, comparing the following
> simple close_all_fds() userspace implementation that is essentially just
> glibc's version in [6]:
> 
> static int close_all_fds(void)
> {
> DIR *dir;
> struct dirent *direntp;
> 
> dir = opendir("/proc/self/fd");
> if (!dir)
> return -1;
> 
> while ((direntp = readdir(dir))) {
> int fd;
> if (strcmp(direntp->d_name, ".") == 0)
> continue;
> if (strcmp(direntp->d_name, "..") == 0)
> continue;
> fd = atoi(direntp->d_name);
> if (fd == 0 || fd == 1 || fd == 2)
> continue;
> close(fd);
> }
> 
> closedir(dir); /* cannot fail */
> return 0;
> }
> 
> to close_range() yields:
> 1. closing 4 open files:
>- close_all_fds(): ~280 us
>- close_range():~24 us
> 
> 2. closing 1000 open files:
>- close_all_fds(): ~5000 us
>- close_range():   ~800 us
> 
> close_range() is designed to allow for some flexibility. Specifically, it
> does not simply always close all open file descriptors of a task. Instead,
> callers can specify an upper bound.
> This is e.g. useful for scenarios where specific file descriptors are
> created with well-known numbers that are supposed to be excluded from
> getting closed.
> For extra paranoia close_range() comes with a flags argument. This can e.g.
> be used to implement extension. Once can imagine userspace wanting to stop
> at the first error instead of ignoring errors under certain circumstances.
> There might be other valid ideas in the future. In any case, a flag
> argument doesn't hurt and keeps us on the safe side.
> 
> >From an implementation side this is kept rather dumb. It saw some input
> from David and Jann but all nonsense is obviously my own!
> - Errors to close file descriptors are currently ignored. (Could be changed
>   by setting a flag in the future if needed.)
> - __close_range() is a rather simplistic wrapper around __close_fd().
>   My reasoning behind this is based on the nature of how __close_fd() needs
>   to release an fd. But maybe I misunderstood specifics:
>   We take the files_lock and rcu-dereference the fdtable of the calling
>   task, we find the entry in the fdtable, get the file and need to release
>  

Re: [PATCH 1/2] open: add close_range()

[Christian Brauner]

On Tue, May 21, 2019 at 03:10:11PM +0200, Florian Weimer wrote:
> * Christian Brauner:
> 
> >> Solaris has an fdwalk function:
> >> 
> >>   
> >> 
> >> So a different way to implement this would expose a nextfd system call
> >
> > Meh. If nextfd() then I would like it to be able to:
> > - get the nextfd(fd) >= fd
> > - get highest open fd e.g. nextfd(-1)
> 
> The highest open descriptor isn't istering for fdwalk because nextfd
> would just fail.

Sure. I was thinking about other usecases. For example, sometimes in
userspace you want to do the following:
save_fd = dup(fd,  
> > But then I wonder if nextfd() needs to be a syscall and isn't just
> > either:
> > fcntl(fd, F_GET_NEXT)?
> > or
> > prctl(PR_GET_NEXT)?
> 
> I think the fcntl route is a bit iffy because you might need it to get
> the *first* valid descriptor.
> 
> >> to userspace, so that we can use that to implement both fdwalk and
> >> closefrom.  But maybe fdwalk is just too obscure, given the existence of
> >> /proc.
> >
> > Yeah we probably don't need fdwalk.
> 
> Agreed.  Just wanted to bring it up for completeness.  I certainly don't
> want to derail the implementation of close_range.
> 
> Thanks,
> Florian

Re: [PATCH 1/2] open: add close_range()

[Christian Brauner]

On Tue, May 21, 2019 at 03:10:11PM +0200, Florian Weimer wrote:
> * Christian Brauner:
> 
> >> Solaris has an fdwalk function:
> >> 
> >>   
> >> 
> >> So a different way to implement this would expose a nextfd system call
> >
> > Meh. If nextfd() then I would like it to be able to:
> > - get the nextfd(fd) >= fd
> > - get highest open fd e.g. nextfd(-1)
> 
> The highest open descriptor isn't istering for fdwalk because nextfd
> would just fail.
> 
> > But then I wonder if nextfd() needs to be a syscall and isn't just
> > either:
> > fcntl(fd, F_GET_NEXT)?
> > or
> > prctl(PR_GET_NEXT)?
> 
> I think the fcntl route is a bit iffy because you might need it to get
> the *first* valid descriptor.

Oh, how would that be difficult? Maybe I'm missing context.
Couldn't you just do

fcntl(0, F_GET_NEXT)

> 
> >> to userspace, so that we can use that to implement both fdwalk and
> >> closefrom.  But maybe fdwalk is just too obscure, given the existence of
> >> /proc.
> >
> > Yeah we probably don't need fdwalk.
> 
> Agreed.  Just wanted to bring it up for completeness.  I certainly don't
> want to derail the implementation of close_range.

No, that's perfectly fine. I mean, you clearly need this and are one of
the major stakeholders. For example, Rust (probably also Python) will
call down into libc and not use the syscall directly. They kinda do this
with getfdtable rn already.
So what you say makes sense for libc has some relevance for the other
tools as well.

Christian

Re: [PATCH 1/2] open: add close_range()

[Florian Weimer]

* Christian Brauner:

>> Solaris has an fdwalk function:
>> 
>>   
>> 
>> So a different way to implement this would expose a nextfd system call
>
> Meh. If nextfd() then I would like it to be able to:
> - get the nextfd(fd) >= fd
> - get highest open fd e.g. nextfd(-1)

The highest open descriptor isn't istering for fdwalk because nextfd
would just fail.

> But then I wonder if nextfd() needs to be a syscall and isn't just
> either:
> fcntl(fd, F_GET_NEXT)?
> or
> prctl(PR_GET_NEXT)?

I think the fcntl route is a bit iffy because you might need it to get
the *first* valid descriptor.

>> to userspace, so that we can use that to implement both fdwalk and
>> closefrom.  But maybe fdwalk is just too obscure, given the existence of
>> /proc.
>
> Yeah we probably don't need fdwalk.

Agreed.  Just wanted to bring it up for completeness.  I certainly don't
want to derail the implementation of close_range.

Thanks,
Florian

Re: [PATCH 1/2] open: add close_range()

[Christian Brauner]

On Tue, May 21, 2019 at 02:09:29PM +0200, Florian Weimer wrote:
> * Christian Brauner:
> 
> > +/**
> > + * __close_range() - Close all file descriptors in a given range.
> > + *
> > + * @fd: starting file descriptor to close
> > + * @max_fd: last file descriptor to close
> > + *
> > + * This closes a range of file descriptors. All file descriptors
> > + * from @fd up to and including @max_fd are closed.
> > + */
> > +int __close_range(struct files_struct *files, unsigned fd, unsigned max_fd)
> > +{
> > +   unsigned int cur_max;
> > +
> > +   if (fd > max_fd)
> > +   return -EINVAL;
> > +
> > +   rcu_read_lock();
> > +   cur_max = files_fdtable(files)->max_fds;
> > +   rcu_read_unlock();
> > +
> > +   /* cap to last valid index into fdtable */
> > +   if (max_fd >= cur_max)
> > +   max_fd = cur_max - 1;
> > +
> > +   while (fd <= max_fd)
> > +   __close_fd(files, fd++);
> > +
> > +   return 0;
> > +}
> 
> This seems rather drastic.  How long does this block in kernel mode?
> Maybe it's okay as long as the maximum possible value for cur_max stays
> around 4 million or so.

That's probably valid concern when you reach very high numbers though I
wonder how relevant this is in practice.
Also, you would only be blocking yourself I imagine, i.e. you can't DOS
another task with this unless your multi-threaded.

> 
> Solaris has an fdwalk function:
> 
>   
> 
> So a different way to implement this would expose a nextfd system call

Meh. If nextfd() then I would like it to be able to:
- get the nextfd(fd) >= fd
- get highest open fd e.g. nextfd(-1)

But then I wonder if nextfd() needs to be a syscall and isn't just
either:
fcntl(fd, F_GET_NEXT)?
or
prctl(PR_GET_NEXT)?

Technically, one could also do:

fd_range(unsigned fd, unsigend end_fd, unsigned flags);

fd_range(3, 50, FD_RANGE_CLOSE);

/* return highest fd within the range [3, 50] */
fd_range(3, 50, FD_RANGE_NEXT);

/* return highest fd */
fd_range(3, UINT_MAX, FD_RANGE_NEXT);

This syscall could also reasonably be extended.

> to userspace, so that we can use that to implement both fdwalk and
> closefrom.  But maybe fdwalk is just too obscure, given the existence of
> /proc.

Yeah we probably don't need fdwalk.

> 
> I'll happily implement closefrom on top of close_range in glibc (plus
> fallback for older kernels based on /proc—with an abort in case that
> doesn't work because the RLIMIT_NOFILE hack is unreliable
> unfortunately).
> 
> Thanks,
> Florian

Re: [PATCH 1/2] open: add close_range()

[Florian Weimer]

* Christian Brauner:

> +/**
> + * __close_range() - Close all file descriptors in a given range.
> + *
> + * @fd: starting file descriptor to close
> + * @max_fd: last file descriptor to close
> + *
> + * This closes a range of file descriptors. All file descriptors
> + * from @fd up to and including @max_fd are closed.
> + */
> +int __close_range(struct files_struct *files, unsigned fd, unsigned max_fd)
> +{
> + unsigned int cur_max;
> +
> + if (fd > max_fd)
> + return -EINVAL;
> +
> + rcu_read_lock();
> + cur_max = files_fdtable(files)->max_fds;
> + rcu_read_unlock();
> +
> + /* cap to last valid index into fdtable */
> + if (max_fd >= cur_max)
> + max_fd = cur_max - 1;
> +
> + while (fd <= max_fd)
> + __close_fd(files, fd++);
> +
> + return 0;
> +}

This seems rather drastic.  How long does this block in kernel mode?
Maybe it's okay as long as the maximum possible value for cur_max stays
around 4 million or so.

Solaris has an fdwalk function:

  

So a different way to implement this would expose a nextfd system call
to userspace, so that we can use that to implement both fdwalk and
closefrom.  But maybe fdwalk is just too obscure, given the existence of
/proc.

I'll happily implement closefrom on top of close_range in glibc (plus
fallback for older kernels based on /proc—with an abort in case that
doesn't work because the RLIMIT_NOFILE hack is unreliable
unfortunately).

Thanks,
Florian

[PATCH 1/2] open: add close_range()

[Christian Brauner]

This adds the close_range() syscall. It allows to efficiently close a range
of file descriptors up to all file descriptors of a calling task.

The syscall came up in a recent discussion around the new mount API and
making new file descriptor types cloexec by default. During this
discussion, Al suggested the close_range() syscall (cf. [1]). Note, a
syscall in this manner has been requested by various people over time.

First, it helps to close all file descriptors of an exec()ing task. This
can be done safely via (quoting Al's example from [1] verbatim):

/* that exec is sensitive */
unshare(CLONE_FILES);
/* we don't want anything past stderr here */
close_range(3, ~0U);
execve();

The code snippet above is one way of working around the problem that file
descriptors are not cloexec by default. This is aggravated by the fact that
we can't just switch them over without massively regressing userspace. For
a whole class of programs having an in-kernel method of closing all file
descriptors is very helpful (e.g. demons, service managers, programming
language standard libraries, container managers etc.).
(Please note, unshare(CLONE_FILES) should only be needed if the calling
 task is multi-threaded and shares the file descriptor table with another
 thread in which case two threads could race with one thread allocating
 file descriptors and the other one closing them via close_range(). For the
 general case close_range() before the execve() is sufficient.)

Second, it allows userspace to avoid implementing closing all file
descriptors by parsing through /proc//fd/* and calling close() on each
file descriptor. From looking at various large(ish) userspace code bases
this or similar patterns are very common in:
- service managers (cf. [4])
- libcs (cf. [6])
- container runtimes (cf. [5])
- programming language runtimes/standard libraries
  - Python (cf. [2])
  - Rust (cf. [7], [8])
As Dmitry pointed out there's even a long-standing glibc bug about missing
kernel support for this task (cf. [3]).
In addition, the syscall will also work for tasks that do not have procfs
mounted and on kernels that do not have procfs support compiled in. In such
situations the only way to make sure that all file descriptors are closed
is to call close() on each file descriptor up to UINT_MAX or RLIMIT_NOFILE,
OPEN_MAX trickery (cf. comment [8] on Rust).

The performance is striking. For good measure, comparing the following
simple close_all_fds() userspace implementation that is essentially just
glibc's version in [6]:

static int close_all_fds(void)
{
DIR *dir;
struct dirent *direntp;

dir = opendir("/proc/self/fd");
if (!dir)
return -1;

while ((direntp = readdir(dir))) {
int fd;
if (strcmp(direntp->d_name, ".") == 0)
continue;
if (strcmp(direntp->d_name, "..") == 0)
continue;
fd = atoi(direntp->d_name);
if (fd == 0 || fd == 1 || fd == 2)
continue;
close(fd);
}

closedir(dir); /* cannot fail */
return 0;
}

to close_range() yields:
1. closing 4 open files:
   - close_all_fds(): ~280 us
   - close_range():~24 us

2. closing 1000 open files:
   - close_all_fds(): ~5000 us
   - close_range():   ~800 us

close_range() is designed to allow for some flexibility. Specifically, it
does not simply always close all open file descriptors of a task. Instead,
callers can specify an upper bound.
This is e.g. useful for scenarios where specific file descriptors are
created with well-known numbers that are supposed to be excluded from
getting closed.
For extra paranoia close_range() comes with a flags argument. This can e.g.
be used to implement extension. Once can imagine userspace wanting to stop
at the first error instead of ignoring errors under certain circumstances.
There might be other valid ideas in the future. In any case, a flag
argument doesn't hurt and keeps us on the safe side.

>From an implementation side this is kept rather dumb. It saw some input
from David and Jann but all nonsense is obviously my own!
- Errors to close file descriptors are currently ignored. (Could be changed
  by setting a flag in the future if needed.)
- __close_range() is a rather simplistic wrapper around __close_fd().
  My reasoning behind this is based on the nature of how __close_fd() needs
  to release an fd. But maybe I misunderstood specifics:
  We take the files_lock and rcu-dereference the fdtable of the calling
  task, we find the entry in the fdtable, get the file and need to release
  files_lock before calling filp_close().
  In the meantime the fdtable might have been altered so we can't just
  retake the spinlock and keep the old rcu-reference of the fdtable
  around. Instead we need to grab a fresh reference to the fdtable.
  If my reasoning is correct then