fs: fix fd table size alignment properly
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 30 Mar 2022 06:29:18 +0000 (23:29 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 8 Apr 2022 12:23:54 +0000 (14:23 +0200)
commitc331c9d1d2b7a5ce875f8f1ac53e3987f73ecc75
tree8cb7ae922ae088dcb5a1843805521c26e6f4d6fc
parent611170142b4244d7d6edba1e3400517d8625296b
fs: fix fd table size alignment properly

[ Upstream commit d888c83fcec75194a8a48ccd283953bdba7b2550 ]

Jason Donenfeld reports that my commit 1c24a186398f ("fs: fd tables have
to be multiples of BITS_PER_LONG") doesn't work, and the reason is an
embarrassing brown-paper-bag bug.

Yes, we want to align the number of fds to BITS_PER_LONG, and yes, the
reason they might not be aligned is because the incoming 'max_fd'
argument might not be aligned.

But aligining the argument - while simple - will cause a "infinitely
big" maxfd (eg NR_OPEN_MAX) to just overflow to zero.  Which most
definitely isn't what we want either.

The obvious fix was always just to do the alignment last, but I had
moved it earlier just to make the patch smaller and the code look
simpler.  Duh.  It certainly made _me_ look simple.

Fixes: 1c24a186398f ("fs: fd tables have to be multiples of BITS_PER_LONG")
Reported-and-tested-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Fedor Pchelkin <aissur0002@gmail.com>
Cc: Alexey Khoroshilov <khoroshilov@ispras.ru>
Cc: Christian Brauner <brauner@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/file.c