linux-kernel - Re: 4.9.0 regression in pipe-backed iov

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20170113215919.GU1555@ZenIV.linux.org.uk>
Date:   Fri, 13 Jan 2017 21:59:19 +0000
From:   Al Viro <viro@...IV.linux.org.uk>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     "Alan J. Wylie" <alan@...ie.me.uk>,
        Thorsten Leemhuis <regressions@...mhuis.info>,
        linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: 4.9.0 regression in pipe-backed iov_iter with systemd-nspawn

On Fri, Jan 13, 2017 at 09:55:04PM +0000, Al Viro wrote:
> On Fri, Jan 13, 2017 at 08:47:31PM +0000, Al Viro wrote:
> > On Fri, Jan 13, 2017 at 12:32:37PM -0800, Linus Torvalds wrote:
> > 
> > > Ugh. I still think your patch is butt-ugly, and the index comparisons
> > > make me nervous, but..
> > 
> > No arguments here - 6am on 20-odd hours of uptime is _not_ a good time
> > for writing, especially since the data structure needs better documentation
> > and probably a couple of inlined helpers.  I'll try to massage the damn
> > thing into more readable form.
> 
> FWIW, I think it will be more readable if we separate the "advance" and
> "truncate" parts like this (warning: not even build-tested).  Comments?

diff --git a/lib/iov_iter.c b/lib/iov_iter.c
index 25f572303801..dae1ac940d5f 100644
--- a/lib/iov_iter.c
+++ b/lib/iov_iter.c
@@ -730,43 +730,60 @@ size_t iov_iter_copy_from_user_atomic(struct page *page,
 }
 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
 
+static inline void pipe_truncate(struct iov_iter *i)
+{
+	struct pipe_inode_info *pipe = i->pipe;
+	if (pipe->nrbufs) {
+		size_t off = i->iov_offset;
+		int idx = i->idx;
+		int n;
+
+		n = (pipe->curbuf + pipe->nrbufs - idx) & (pipe->buffers - 1);
+		if (off) {
+			pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
+			/* free all after idx; n can't be 0 */
+			idx = next_idx(idx, pipe);
+			n--;
+		} else {
+			/* free all _starting_ at idx.
+			 * n is 0 when we have nothing to do
+			 * *or* when we are truncating full pipe to empty.
+			 */
+			if (pipe->nrbufs == pipe->buffers && !n)
+				n = pipe->buffers;
+		}
+		while (n--) {
+			pipe_buf_release(pipe, &pipe->bufs[idx]);
+			idx = next_idx(idx, pipe);
+			pipe->nrbufs--;
+		}
+	}
+}
+
 static void pipe_advance(struct iov_iter *i, size_t size)
 {
 	struct pipe_inode_info *pipe = i->pipe;
-	struct pipe_buffer *buf;
-	int idx = i->idx;
-	size_t off = i->iov_offset, orig_sz;
-	
 	if (unlikely(i->count < size))
 		size = i->count;
-	orig_sz = size;
-
 	if (size) {
+		struct pipe_buffer *buf;
+		size_t off = i->iov_offset, left = size;
+		int idx = i->idx;
 		if (off) /* make it relative to the beginning of buffer */
-			size += off - pipe->bufs[idx].offset;
+			left += off - pipe->bufs[idx].offset;
 		while (1) {
 			buf = &pipe->bufs[idx];
-			if (size <= buf->len)
+			if (left <= buf->len)
 				break;
-			size -= buf->len;
+			left -= buf->len;
 			idx = next_idx(idx, pipe);
 		}
-		buf->len = size;
 		i->idx = idx;
-		off = i->iov_offset = buf->offset + size;
-	}
-	if (off)
-		idx = next_idx(idx, pipe);
-	if (pipe->nrbufs) {
-		int unused = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
-		/* [curbuf,unused) is in use.  Free [idx,unused) */
-		while (idx != unused) {
-			pipe_buf_release(pipe, &pipe->bufs[idx]);
-			idx = next_idx(idx, pipe);
-			pipe->nrbufs--;
-		}
+		i->iov_offset = buf->offset + left;
 	}
-	i->count -= orig_sz;
+	i->count -= size;
+	/* ... and discard everything past that point */
+	pipe_truncate(i);
 }
 
 void iov_iter_advance(struct iov_iter *i, size_t size)
@@ -826,6 +843,7 @@ void iov_iter_pipe(struct iov_iter *i, int direction,
 			size_t count)
 {
 	BUG_ON(direction != ITER_PIPE);
+	WARN_ON(pipe->nrbufs == pipe->buffers);
 	i->type = direction;
 	i->pipe = pipe;
 	i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);