/*
 * netsniff-ng - the packet sniffing beast
 * Copyright 2009, 2010 Daniel Borkmann.
 * Subject to the GPL, version 2.
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <arpa/inet.h>
#include <linux/if_ether.h>

#include "xmalloc.h"
#include "die.h"
#include "ring_rx.h"
#include "built_in.h"

void destroy_rx_ring(int sock, struct ring *ring)
{
	int ret;

	munmap(ring->mm_space, ring->mm_len);
	ring->mm_len = 0;

	fmemset(&ring->layout, 0, sizeof(ring->layout));
	ret = setsockopt(sock, SOL_PACKET, PACKET_RX_RING, &ring->layout,
			 sizeof(ring->layout));
	if (unlikely(ret))
		panic("Cannot destroy the RX_RING: %s!\n", strerror(errno));

	xfree(ring->frames);
}

void setup_rx_ring_layout(int sock, struct ring *ring, unsigned int size,
			  int jumbo_support)
{
	fmemset(&ring->layout, 0, sizeof(ring->layout));

	ring->layout.tp_block_size = (jumbo_support ?
				      getpagesize() << 4 :
				      getpagesize() << 2);
	ring->layout.tp_frame_size = (jumbo_support ?
				      TPACKET_ALIGNMENT << 12 :
				      TPACKET_ALIGNMENT << 7);
	ring->layout.tp_block_nr = size / ring->layout.tp_block_size;
	ring->layout.tp_frame_nr = ring->layout.tp_block_size /
				   ring->layout.tp_frame_size *
				   ring->layout.tp_block_nr;

	bug_on(ring->layout.tp_block_size < ring->layout.tp_frame_size);
	bug_on((ring->layout.tp_block_size % ring->layout.tp_frame_size) != 0);
	bug_on((ring->layout.tp_block_size % getpagesize()) != 0);
}

void create_rx_ring(int sock, struct ring *ring, int verbose)
{
	int ret;

	set_sockopt_tpacket(sock);
retry:
	ret = setsockopt(sock, SOL_PACKET, PACKET_RX_RING, &ring->layout,
			 sizeof(ring->layout));
	if (errno == ENOMEM && ring->layout.tp_block_nr > 1) {
		ring->layout.tp_block_nr >>= 1;
		ring->layout.tp_frame_nr = ring->layout.tp_block_size / 
					   ring->layout.tp_frame_size * 
					   ring->layout.tp_block_nr;
		goto retry;
	}

	if (ret < 0)
		panic("Cannot allocate RX_RING!\n");

	ring->mm_len = ring->layout.tp_block_size * ring->layout.tp_block_nr;

	if (verbose) {
		printf("RX: %.2Lf MiB, %u Frames, each %u Byte allocated\n",
		       (long double) ring->mm_len / (1 << 20),
		       ring->layout.tp_frame_nr, ring->layout.tp_frame_size);
	}
}

void mmap_rx_ring(int sock, struct ring *ring)
{
	ring->mm_space = mmap(0, ring->mm_len, PROT_READ | PROT_WRITE,
			      MAP_SHARED | MAP_LOCKED | MAP_POPULATE, sock, 0);
	if (ring->mm_space == MAP_FAILED) {
		destroy_rx_ring(sock, ring);
		panic("Cannot mmap RX_RING!\n");
	}
}

void alloc_rx_ring_frames(struct ring *ring)
{
	int i;
	size_t len = ring->layout.tp_frame_nr * sizeof(*ring->frames);

	ring->frames = xmalloc_aligned(len, CO_CACHE_LINE_SIZE);
	fmemset(ring->frames, 0, len);

	for (i = 0; i < ring->layout.tp_frame_nr; ++i) {
		ring->frames[i].iov_len = ring->layout.tp_frame_size;
		ring->frames[i].iov_base = ring->mm_space +
					   (i * ring->layout.tp_frame_size);
	}
}

void bind_rx_ring(int sock, struct ring *ring, int ifindex)
{
	int ret;
	/*
	 * The RX_RING registers itself to the networking stack with
	 * dev_add_pack(), so we have one single RX_RING for all devs
	 * otherwise you'll get the packet twice.
	 */
	fmemset(&ring->s_ll, 0, sizeof(ring->s_ll));

	ring->s_ll.sll_family = AF_PACKET;
	ring->s_ll.sll_protocol = htons(ETH_P_ALL);
	ring->s_ll.sll_ifindex = ifindex;
	ring->s_ll.sll_hatype = 0;
	ring->s_ll.sll_halen = 0;
	ring->s_ll.sll_pkttype = 0;

	ret = bind(sock, (struct sockaddr *) &ring->s_ll, sizeof(ring->s_ll));
	if (ret < 0) {
		destroy_rx_ring(sock, ring);
		panic("Cannot bind RX_RING!\n");
	}
}
ion value='1'>ssdiff</option><option value='2'>stat only</option></select></td></tr><tr><td/><td class='ctrl'><noscript><input type='submit' value='reload'/></noscript></td></tr></table></form></div><table summary='commit info' class='commit-info'>
<tr><th>author</th><td>Anand Jain &lt;anand.jain@oracle.com&gt;</td><td class='right'>2016-09-22 12:56:13 +0800</td></tr>
<tr><th>committer</th><td>David Sterba &lt;dsterba@suse.com&gt;</td><td class='right'>2016-09-26 19:37:06 +0200</td></tr>
<tr><th>commit</th><td colspan='2' class='oid'><a href='/cgit.cgi/linux/net-next.git/commit/?id=0ccd05285e7f5a8e297e1d6dfc41e7c65757d6fa'>0ccd05285e7f5a8e297e1d6dfc41e7c65757d6fa</a> (<a href='/cgit.cgi/linux/net-next.git/patch/?id=0ccd05285e7f5a8e297e1d6dfc41e7c65757d6fa'>patch</a>)</td></tr>
<tr><th>tree</th><td colspan='2' class='oid'><a href='/cgit.cgi/linux/net-next.git/tree/?id=0ccd05285e7f5a8e297e1d6dfc41e7c65757d6fa'>43a35000feb9d8610afd4c74b81a4160131f0a78</a></td></tr>
<tr><th>parent</th><td colspan='2' class='oid'><a href='/cgit.cgi/linux/net-next.git/commit/?id=a958eab0ed7fdc1b977bc25d3af6efedaa945488'>a958eab0ed7fdc1b977bc25d3af6efedaa945488</a> (<a href='/cgit.cgi/linux/net-next.git/diff/?id=0ccd05285e7f5a8e297e1d6dfc41e7c65757d6fa&amp;id2=a958eab0ed7fdc1b977bc25d3af6efedaa945488'>diff</a>)</td></tr></table>
<div class='commit-subject'>btrfs: fix a possible umount deadlock</div><div class='commit-msg'>btrfs_show_devname() is using the device_list_mutex, sometimes
a call to blkdev_put() leads vfs calling into this func. So
call blkdev_put() outside of device_list_mutex, as of now.

[  983.284212] ======================================================
[  983.290401] [ INFO: possible circular locking dependency detected ]
[  983.296677] 4.8.0-rc5-ceph-00023-g1b39cec2 #1 Not tainted
[  983.302081] -------------------------------------------------------
[  983.308357] umount/21720 is trying to acquire lock:
[  983.313243]  (&amp;bdev-&gt;bd_mutex){+.+.+.}, at: [&lt;ffffffff9128ec51&gt;] blkdev_put+0x31/0x150
[  983.321264]
[  983.321264] but task is already holding lock:
[  983.327101]  (&amp;fs_devs-&gt;device_list_mutex){+.+...}, at: [&lt;ffffffffc033d6f6&gt;] __btrfs_close_devices+0x46/0x200 [btrfs]
[  983.337839]
[  983.337839] which lock already depends on the new lock.
[  983.337839]
[  983.346024]
[  983.346024] the existing dependency chain (in reverse order) is:
[  983.353512]
-&gt; #4 (&amp;fs_devs-&gt;device_list_mutex){+.+...}:
[  983.359096]        [&lt;ffffffff910dfd0c&gt;] lock_acquire+0x1bc/0x1f0
[  983.365143]        [&lt;ffffffff91823125&gt;] mutex_lock_nested+0x65/0x350
[  983.371521]        [&lt;ffffffffc02d8116&gt;] btrfs_show_devname+0x36/0x1f0 [btrfs]
[  983.378710]        [&lt;ffffffff9129523e&gt;] show_vfsmnt+0x4e/0x150
[  983.384593]        [&lt;ffffffff9126ffc7&gt;] m_show+0x17/0x20
[  983.389957]        [&lt;ffffffff91276405&gt;] seq_read+0x2b5/0x3b0
[  983.395669]        [&lt;ffffffff9124c808&gt;] __vfs_read+0x28/0x100
[  983.401464]        [&lt;ffffffff9124eb3b&gt;] vfs_read+0xab/0x150
[  983.407080]        [&lt;ffffffff9124ec32&gt;] SyS_read+0x52/0xb0
[  983.412609]        [&lt;ffffffff91825fc0&gt;] entry_SYSCALL_64_fastpath+0x23/0xc1
[  983.419617]
-&gt; #3 (namespace_sem){++++++}:
[  983.424024]        [&lt;ffffffff910dfd0c&gt;] lock_acquire+0x1bc/0x1f0
[  983.430074]        [&lt;ffffffff918239e9&gt;] down_write+0x49/0x80
[  983.435785]        [&lt;ffffffff91272457&gt;] lock_mount+0x67/0x1c0
[  983.441582]        [&lt;ffffffff91272ab2&gt;] do_add_mount+0x32/0xf0
[  983.447458]        [&lt;ffffffff9127363a&gt;] finish_automount+0x5a/0xc0
[  983.453682]        [&lt;ffffffff91259513&gt;] follow_managed+0x1b3/0x2a0
[  983.459912]        [&lt;ffffffff9125b750&gt;] lookup_fast+0x300/0x350
[  983.465875]        [&lt;ffffffff9125d6e7&gt;] path_openat+0x3a7/0xaa0
[  983.471846]        [&lt;ffffffff9125ef75&gt;] do_filp_open+0x85/0xe0
[  983.477731]        [&lt;ffffffff9124c41c&gt;] do_sys_open+0x14c/0x1f0
[  983.483702]        [&lt;ffffffff9124c4de&gt;] SyS_open+0x1e/0x20
[  983.489240]        [&lt;ffffffff91825fc0&gt;] entry_SYSCALL_64_fastpath+0x23/0xc1
[  983.496254]
-&gt; #2 (&amp;sb-&gt;s_type-&gt;i_mutex_key#3){+.+.+.}:
[  983.501798]        [&lt;ffffffff910dfd0c&gt;] lock_acquire+0x1bc/0x1f0
[  983.507855]        [&lt;ffffffff918239e9&gt;] down_write+0x49/0x80
[  983.513558]        [&lt;ffffffff91366237&gt;] start_creating+0x87/0x100
[  983.519703]        [&lt;ffffffff91366647&gt;] debugfs_create_dir+0x17/0x100
[  983.526195]        [&lt;ffffffff911df153&gt;] bdi_register+0x93/0x210
[  983.532165]        [&lt;ffffffff911df313&gt;] bdi_register_owner+0x43/0x70
[  983.538570]        [&lt;ffffffff914080fb&gt;] device_add_disk+0x1fb/0x450
[  983.544888]        [&lt;ffffffff91580226&gt;] loop_add+0x1e6/0x290
[  983.550596]        [&lt;ffffffff91fec358&gt;] loop_init+0x10b/0x14f
[  983.556394]        [&lt;ffffffff91002207&gt;] do_one_initcall+0xa7/0x180
[  983.562618]        [&lt;ffffffff91f932e0&gt;] kernel_init_freeable+0x1cc/0x266
[  983.569370]        [&lt;ffffffff918174be&gt;] kernel_init+0xe/0x100
[  983.575166]        [&lt;ffffffff9182620f&gt;] ret_from_fork+0x1f/0x40
[  983.581131]
-&gt; #1 (loop_index_mutex){+.+.+.}:
[  983.585801]        [&lt;ffffffff910dfd0c&gt;] lock_acquire+0x1bc/0x1f0
[  983.591858]        [&lt;ffffffff91823125&gt;] mutex_lock_nested+0x65/0x350
[  983.598256]        [&lt;ffffffff9157ed3f&gt;] lo_open+0x1f/0x60
[  983.603704]        [&lt;ffffffff9128eec3&gt;] __blkdev_get+0x123/0x400
[  983.609757]        [&lt;ffffffff9128f4ea&gt;] blkdev_get+0x34a/0x350
[  983.615639]        [&lt;ffffffff9128f554&gt;] blkdev_open+0x64/0x80
[  983.621428]        [&lt;ffffffff9124aff6&gt;] do_dentry_open+0x1c6/0x2d0
[  983.627651]        [&lt;ffffffff9124c029&gt;] vfs_open+0x69/0x80
[  983.633181]        [&lt;ffffffff9125db74&gt;] path_openat+0x834/0xaa0
[  983.639152]        [&lt;ffffffff9125ef75&gt;] do_filp_open+0x85/0xe0
[  983.645035]        [&lt;ffffffff9124c41c&gt;] do_sys_open+0x14c/0x1f0
[  983.650999]        [&lt;ffffffff9124c4de&gt;] SyS_open+0x1e/0x20
[  983.656535]        [&lt;ffffffff91825fc0&gt;] entry_SYSCALL_64_fastpath+0x23/0xc1
[  983.663541]
-&gt; #0 (&amp;bdev-&gt;bd_mutex){+.+.+.}:
[  983.668107]        [&lt;ffffffff910def43&gt;] __lock_acquire+0x1003/0x17b0
[  983.674510]        [&lt;ffffffff910dfd0c&gt;] lock_acquire+0x1bc/0x1f0
[  983.680561]        [&lt;ffffffff91823125&gt;] mutex_lock_nested+0x65/0x350
[  983.686967]        [&lt;ffffffff9128ec51&gt;] blkdev_put+0x31/0x150
[  983.692761]        [&lt;ffffffffc033481f&gt;] btrfs_close_bdev+0x4f/0x60 [btrfs]
[  983.699699]        [&lt;ffffffffc033d77b&gt;] __btrfs_close_devices+0xcb/0x200 [btrfs]
[  983.707178]        [&lt;ffffffffc033d8db&gt;] btrfs_close_devices+0x2b/0xa0 [btrfs]
[  983.714380]        [&lt;ffffffffc03081c5&gt;] close_ctree+0x265/0x340 [btrfs]
[  983.721061]        [&lt;ffffffffc02d7959&gt;] btrfs_put_super+0x19/0x20 [btrfs]
[  983.727908]        [&lt;ffffffff91250e2f&gt;] generic_shutdown_super+0x6f/0x100
[  983.734744]        [&lt;ffffffff91250f56&gt;] kill_anon_super+0x16/0x30
[  983.740888]        [&lt;ffffffffc02da97e&gt;] btrfs_kill_super+0x1e/0x130 [btrfs]
[  983.747909]        [&lt;ffffffff91250fe9&gt;] deactivate_locked_super+0x49/0x80
[  983.754745]        [&lt;ffffffff912515fd&gt;] deactivate_super+0x5d/0x70
[  983.760977]        [&lt;ffffffff91270a1c&gt;] cleanup_mnt+0x5c/0x80
[  983.766773]        [&lt;ffffffff91270a92&gt;] __cleanup_mnt+0x12/0x20
[  983.772738]        [&lt;ffffffff910aa2fe&gt;] task_work_run+0x7e/0xc0
[  983.778708]        [&lt;ffffffff91081b5a&gt;] exit_to_usermode_loop+0x7e/0xb4
[  983.785373]        [&lt;ffffffff910039eb&gt;] syscall_return_slowpath+0xbb/0xd0
[  983.792212]        [&lt;ffffffff9182605c&gt;] entry_SYSCALL_64_fastpath+0xbf/0xc1
[  983.799225]
[  983.799225] other info that might help us debug this:
[  983.799225]
[  983.807291] Chain exists of:
  &amp;bdev-&gt;bd_mutex --&gt; namespace_sem --&gt; &amp;fs_devs-&gt;device_list_mutex

[  983.816521]  Possible unsafe locking scenario:
[  983.816521]
[  983.822489]        CPU0                    CPU1
[  983.827043]        ----                    ----
[  983.831599]   lock(&amp;fs_devs-&gt;device_list_mutex);
[  983.836289]                                lock(namespace_sem);
[  983.842268]                                lock(&amp;fs_devs-&gt;device_list_mutex);
[  983.849478]   lock(&amp;bdev-&gt;bd_mutex);
[  983.853127]
[  983.853127]  *** DEADLOCK ***
[  983.853127]
[  983.859113] 3 locks held by umount/21720:
[  983.863145]  #0:  (&amp;type-&gt;s_umount_key#35){++++..}, at: [&lt;ffffffff912515f5&gt;] deactivate_super+0x55/0x70
[  983.872713]  #1:  (uuid_mutex){+.+.+.}, at: [&lt;ffffffffc033d8d3&gt;] btrfs_close_devices+0x23/0xa0 [btrfs]
[  983.882206]  #2:  (&amp;fs_devs-&gt;device_list_mutex){+.+...}, at: [&lt;ffffffffc033d6f6&gt;] __btrfs_close_devices+0x46/0x200 [btrfs]
[  983.893422]
[  983.893422] stack backtrace:
[  983.897824] CPU: 6 PID: 21720 Comm: umount Not tainted 4.8.0-rc5-ceph-00023-g1b39cec2 #1
[  983.905958] Hardware name: Supermicro SYS-5018R-WR/X10SRW-F, BIOS 1.0c 09/07/2015
[  983.913492]  0000000000000000 ffff8c8a53c17a38 ffffffff91429521 ffffffff9260f4f0
[  983.921018]  ffffffff92642760 ffff8c8a53c17a88 ffffffff911b2b04 0000000000000050
[  983.928542]  ffffffff9237d620 ffff8c8a5294aee0 ffff8c8a5294aeb8 ffff8c8a5294aee0
[  983.936072] Call Trace:
[  983.938545]  [&lt;ffffffff91429521&gt;] dump_stack+0x85/0xc4
[  983.943715]  [&lt;ffffffff911b2b04&gt;] print_circular_bug+0x1fb/0x20c
[  983.949748]  [&lt;ffffffff910def43&gt;] __lock_acquire+0x1003/0x17b0
[  983.955613]  [&lt;ffffffff910dfd0c&gt;] lock_acquire+0x1bc/0x1f0
[  983.961123]  [&lt;ffffffff9128ec51&gt;] ? blkdev_put+0x31/0x150
[  983.966550]  [&lt;ffffffff91823125&gt;] mutex_lock_nested+0x65/0x350
[  983.972407]  [&lt;ffffffff9128ec51&gt;] ? blkdev_put+0x31/0x150
[  983.977832]  [&lt;ffffffff9128ec51&gt;] blkdev_put+0x31/0x150
[  983.983101]  [&lt;ffffffffc033481f&gt;] btrfs_close_bdev+0x4f/0x60 [btrfs]
[  983.989500]  [&lt;ffffffffc033d77b&gt;] __btrfs_close_devices+0xcb/0x200 [btrfs]
[  983.996415]  [&lt;ffffffffc033d8db&gt;] btrfs_close_devices+0x2b/0xa0 [btrfs]
[  984.003068]  [&lt;ffffffffc03081c5&gt;] close_ctree+0x265/0x340 [btrfs]
[  984.009189]  [&lt;ffffffff9126cc5e&gt;] ? evict_inodes+0x15e/0x170
[  984.014881]  [&lt;ffffffffc02d7959&gt;] btrfs_put_super+0x19/0x20 [btrfs]
[  984.021176]  [&lt;ffffffff91250e2f&gt;] generic_shutdown_super+0x6f/0x100
[  984.027476]  [&lt;ffffffff91250f56&gt;] kill_anon_super+0x16/0x30
[  984.033082]  [&lt;ffffffffc02da97e&gt;] btrfs_kill_super+0x1e/0x130 [btrfs]
[  984.039548]  [&lt;ffffffff91250fe9&gt;] deactivate_locked_super+0x49/0x80
[  984.045839]  [&lt;ffffffff912515fd&gt;] deactivate_super+0x5d/0x70
[  984.051525]  [&lt;ffffffff91270a1c&gt;] cleanup_mnt+0x5c/0x80
[  984.056774]  [&lt;ffffffff91270a92&gt;] __cleanup_mnt+0x12/0x20
[  984.062201]  [&lt;ffffffff910aa2fe&gt;] task_work_run+0x7e/0xc0
[  984.067625]  [&lt;ffffffff91081b5a&gt;] exit_to_usermode_loop+0x7e/0xb4
[  984.073747]  [&lt;ffffffff910039eb&gt;] syscall_return_slowpath+0xbb/0xd0
[  984.080038]  [&lt;ffffffff9182605c&gt;] entry_SYSCALL_64_fastpath+0xbf/0xc1

Reported-by: Ilya Dryomov &lt;idryomov@gmail.com&gt;
Signed-off-by: Anand Jain &lt;anand.jain@oracle.com&gt;
Reviewed-by: David Sterba &lt;dsterba@suse.com&gt;
Signed-off-by: David Sterba &lt;dsterba@suse.com&gt;
</div><div class='diffstat-header'><a href='/cgit.cgi/linux/net-next.git/diff/?id=0ccd05285e7f5a8e297e1d6dfc41e7c65757d6fa'>Diffstat</a></div><table summary='diffstat' class='diffstat'><tr><td class='mode'>-rw-r--r--</td><td class='upd'><a href='/cgit.cgi/linux/net-next.git/diff/fs/btrfs/volumes.c?id=0ccd05285e7f5a8e297e1d6dfc41e7c65757d6fa'>fs/btrfs/volumes.c</a></td><td class='right'>26</td><td class='graph'><table summary='file diffstat' width='26%'><tr><td class='add' style='width: 76.9%;'/><td class='rem' style='width: 23.1%;'/><td class='none' style='width: 0.0%;'/></tr></table></td></tr>
</table><div class='diffstat-summary'>1 files changed, 20 insertions, 6 deletions</div><table summary='diff' class='diff'><tr><td><div class='head'>diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c