[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20160114222544.GB17997@ZenIV.linux.org.uk>
Date: Thu, 14 Jan 2016 22:25:44 +0000
From: Al Viro <viro@...IV.linux.org.uk>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: Tomeu Vizoso <tomeu@...euvizoso.net>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Neil Brown <neilb@...e.com>,
linux-fsdevel <linux-fsdevel@...r.kernel.org>
Subject: Re: [PATCH v2 06/11] don't put symlink bodies in pagecache into
highmem
On Thu, Jan 14, 2016 at 01:40:32PM -0800, Linus Torvalds wrote:
> On Thu, Jan 14, 2016 at 1:02 PM, Al Viro <viro@...iv.linux.org.uk> wrote:
> >
> > Arrrgh. Try this:
>
> Yeah, that would do it.
>
> Al, did you check any other filesystems do this?
There's one more turd like that - shmem should've done inode_nohighmem()
a bit earlier in shmem_symlink(). The rest is OK.
> Also, I'm wondering if we should perhaps revert the "don't use
> highmem". Do we actually have examples of running out of kmaps? Do we
> care?
For one thing, we'll lose RCU ->get_link() for those. For another... yes,
it was a nasty bug (I missed the possibility that filesystem might seed
the page cache on ->symlink() directly and use a highmem page - mea culpa),
but we can easily catch it at runtime. We really shouldn't put highmem
pages into address_space without __GFP_HIGHMEM, and catching those in
__add_to_page_cache_locked() isn't costly.
Anyway, mm/shmem.c bit follows. With that + NFS one we ought to be OK
wrt that class of bogosities. I'll write the bits for
Documentation/filesystems/porting (basically, "if you preseed the pagecache
at ->symlink() time, don't put highmem pages there; page_symlink() will
take care of that, provided that inode_nohighmem() is called first") and
push the combined patch to #for-linus.
diff --git a/mm/shmem.c b/mm/shmem.c
index 5813b7f..642471b 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -2469,6 +2469,7 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
inode->i_op = &shmem_short_symlink_operations;
inode->i_link = info->symlink;
} else {
+ inode_nohighmem(inode);
error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
if (error) {
iput(inode);
@@ -2476,7 +2477,6 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
}
inode->i_mapping->a_ops = &shmem_aops;
inode->i_op = &shmem_symlink_inode_operations;
- inode_nohighmem(inode);
memcpy(page_address(page), symname, len);
SetPageUptodate(page);
set_page_dirty(page);
Powered by blists - more mailing lists