net, swap: Remove a warning and clarify why sk_mem_reclaim is required when deactivat...
authorMel Gorman <mgorman@suse.de>
Thu, 11 Jun 2015 01:02:04 +0000 (21:02 -0400)
committerDavid S. Miller <davem@davemloft.net>
Thu, 11 Jun 2015 06:02:31 +0000 (23:02 -0700)
Jeff Layton reported the following;

 [   74.232485] ------------[ cut here ]------------
 [   74.233354] WARNING: CPU: 2 PID: 754 at net/core/sock.c:364 sk_clear_memalloc+0x51/0x80()
 [   74.234790] Modules linked in: cts rpcsec_gss_krb5 nfsv4 dns_resolver nfs fscache xfs libcrc32c snd_hda_codec_generic snd_hda_intel snd_hda_controller snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device nfsd snd_pcm snd_timer snd e1000 ppdev parport_pc joydev parport pvpanic soundcore floppy serio_raw i2c_piix4 pcspkr nfs_acl lockd virtio_balloon acpi_cpufreq auth_rpcgss grace sunrpc qxl drm_kms_helper ttm drm virtio_console virtio_blk virtio_pci ata_generic virtio_ring pata_acpi virtio
 [   74.243599] CPU: 2 PID: 754 Comm: swapoff Not tainted 4.1.0-rc6+ #5
 [   74.244635] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
 [   74.245546]  0000000000000000 0000000079e69e31 ffff8800d066bde8 ffffffff8179263d
 [   74.246786]  0000000000000000 0000000000000000 ffff8800d066be28 ffffffff8109e6fa
 [   74.248175]  0000000000000000 ffff880118d48000 ffff8800d58f5c08 ffff880036e380a8
 [   74.249483] Call Trace:
 [   74.249872]  [<ffffffff8179263d>] dump_stack+0x45/0x57
 [   74.250703]  [<ffffffff8109e6fa>] warn_slowpath_common+0x8a/0xc0
 [   74.251655]  [<ffffffff8109e82a>] warn_slowpath_null+0x1a/0x20
 [   74.252585]  [<ffffffff81661241>] sk_clear_memalloc+0x51/0x80
 [   74.253519]  [<ffffffffa0116c72>] xs_disable_swap+0x42/0x80 [sunrpc]
 [   74.254537]  [<ffffffffa01109de>] rpc_clnt_swap_deactivate+0x7e/0xc0 [sunrpc]
 [   74.255610]  [<ffffffffa03e4fd7>] nfs_swap_deactivate+0x27/0x30 [nfs]
 [   74.256582]  [<ffffffff811e99d4>] destroy_swap_extents+0x74/0x80
 [   74.257496]  [<ffffffff811ecb52>] SyS_swapoff+0x222/0x5c0
 [   74.258318]  [<ffffffff81023f27>] ? syscall_trace_leave+0xc7/0x140
 [   74.259253]  [<ffffffff81798dae>] system_call_fastpath+0x12/0x71
 [   74.260158] ---[ end trace 2530722966429f10 ]---

The warning in question was unnecessary but with Jeff's series the rules
are also clearer.  This patch removes the warning and updates the comment
to explain why sk_mem_reclaim() may still be called.

[jlayton: remove if (sk->sk_forward_alloc) conditional. As Leon
          points out that it's not needed.]

Cc: Leon Romanovsky <leon@leon.nu>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/sock.c

index 292f42228bfb361b5748998bbcc538b1e16a2f22..469d6039c7f5dcad90427793161a73c2747baf6e 100644 (file)
@@ -354,15 +354,12 @@ void sk_clear_memalloc(struct sock *sk)
 
        /*
         * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
-        * progress of swapping. However, if SOCK_MEMALLOC is cleared while
-        * it has rmem allocations there is a risk that the user of the
-        * socket cannot make forward progress due to exceeding the rmem
-        * limits. By rights, sk_clear_memalloc() should only be called
-        * on sockets being torn down but warn and reset the accounting if
-        * that assumption breaks.
+        * progress of swapping. SOCK_MEMALLOC may be cleared while
+        * it has rmem allocations due to the last swapfile being deactivated
+        * but there is a risk that the socket is unusable due to exceeding
+        * the rmem limits. Reclaim the reserves and obey rmem limits again.
         */
-       if (WARN_ON(sk->sk_forward_alloc))
-               sk_mem_reclaim(sk);
+       sk_mem_reclaim(sk);
 }
 EXPORT_SYMBOL_GPL(sk_clear_memalloc);