]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/commitdiff
3.18-stable patches
authorGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 18 May 2018 07:24:10 +0000 (09:24 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 18 May 2018 07:24:10 +0000 (09:24 +0200)
added patches:
pipe-cap-initial-pipe-capacity-according-to-pipe-max-size-limit.patch

queue-3.18/pipe-cap-initial-pipe-capacity-according-to-pipe-max-size-limit.patch [new file with mode: 0644]
queue-3.18/series

diff --git a/queue-3.18/pipe-cap-initial-pipe-capacity-according-to-pipe-max-size-limit.patch b/queue-3.18/pipe-cap-initial-pipe-capacity-according-to-pipe-max-size-limit.patch
new file mode 100644 (file)
index 0000000..a206611
--- /dev/null
@@ -0,0 +1,67 @@
+From 086e774a57fba4695f14383c0818994c0b31da7c Mon Sep 17 00:00:00 2001
+From: "Michael Kerrisk (man-pages)" <mtk.manpages@gmail.com>
+Date: Tue, 11 Oct 2016 13:53:43 -0700
+Subject: pipe: cap initial pipe capacity according to pipe-max-size limit
+
+From: Michael Kerrisk (man-pages) <mtk.manpages@gmail.com>
+
+commit 086e774a57fba4695f14383c0818994c0b31da7c upstream.
+
+This is a patch that provides behavior that is more consistent, and
+probably less surprising to users. I consider the change optional, and
+welcome opinions about whether it should be applied.
+
+By default, pipes are created with a capacity of 64 kiB.  However,
+/proc/sys/fs/pipe-max-size may be set smaller than this value.  In this
+scenario, an unprivileged user could thus create a pipe whose initial
+capacity exceeds the limit. Therefore, it seems logical to cap the
+initial pipe capacity according to the value of pipe-max-size.
+
+The test program shown earlier in this patch series can be used to
+demonstrate the effect of the change brought about with this patch:
+
+    # cat /proc/sys/fs/pipe-max-size
+    1048576
+    # sudo -u mtk ./test_F_SETPIPE_SZ 1
+    Initial pipe capacity: 65536
+    # echo 10000 > /proc/sys/fs/pipe-max-size
+    # cat /proc/sys/fs/pipe-max-size
+    16384
+    # sudo -u mtk ./test_F_SETPIPE_SZ 1
+    Initial pipe capacity: 16384
+    # ./test_F_SETPIPE_SZ 1
+    Initial pipe capacity: 65536
+
+The last two executions of 'test_F_SETPIPE_SZ' show that pipe-max-size
+caps the initial allocation for a new pipe for unprivileged users, but
+not for privileged users.
+
+Link: http://lkml.kernel.org/r/31dc7064-2a17-9c5b-1df1-4e3012ee992c@gmail.com
+Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
+Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com>
+Cc: Willy Tarreau <w@1wt.eu>
+Cc: <socketpair@gmail.com>
+Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
+Cc: Jens Axboe <axboe@fb.com>
+Cc: Al Viro <viro@zeniv.linux.org.uk>
+Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
+Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
+Signed-off-by: Daniel Sangorrin <daniel.sangorrin@toshiba.co.jp>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ fs/pipe.c |    3 +++
+ 1 file changed, 3 insertions(+)
+
+--- a/fs/pipe.c
++++ b/fs/pipe.c
+@@ -618,6 +618,9 @@ struct pipe_inode_info *alloc_pipe_info(
+               unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
+               struct user_struct *user = get_current_user();
++              if (pipe_bufs * PAGE_SIZE > pipe_max_size && !capable(CAP_SYS_RESOURCE))
++                      pipe_bufs = pipe_max_size >> PAGE_SHIFT;
++
+               if (!too_many_pipe_buffers_hard(user)) {
+                       if (too_many_pipe_buffers_soft(user))
+                               pipe_bufs = 1;
index f73f8d053d789012379231d032a1f243b8495a0a..d71e2bca686f1031473fa0c24a9a568955fe86d4 100644 (file)
@@ -18,3 +18,4 @@ qmi_wwan-do-not-steal-interfaces-from-class-drivers.patch
 lockd-lost-rollback-of-set_grace_period-in-lockd_down_net.patch
 revert-arm-dts-imx6qdl-wandboard-fix-audio-channel-swap.patch
 l2tp-revert-l2tp-fix-missing-print-session-offset-info.patch
+pipe-cap-initial-pipe-capacity-according-to-pipe-max-size-limit.patch