tmpfs: occasional WARNING at mm/shmem.c:609 shmem_evict_inode()

tmpfs: occasional WARNING at mm/shmem.c:609 shmem_evict_inode()

[Holger Hoffstätte]

Not sure if this should go here (-mm?), but it is FS related, so..

This is with 3.19.1, but I think I saw it before as well. Every once in a while
(approx. every couple of weeks or so) I get something like the following:

Mar  9 18:25:01 tux kernel: ------------[ cut here ]------------
Mar  9 18:25:01 tux kernel: WARNING: CPU: 1 PID: 24856 at mm/shmem.c:609 shmem_evict_inode+0x10a/0x140()
Mar  9 18:25:01 tux kernel: Modules linked in: btrfs nfsd auth_rpcgss oid_registry lockd grace sunrpc sch_fq_codel xor raid6_pq snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic x86_pkg_temp_thermal coretemp crc32_pclmul i915 crc32c_intel aesni_intel snd_hda_intel aes_x86_64 intel_gtt snd_hda_controller glue_helper i2c_algo_bit lrw gf128mul snd_hda_codec ablk_helper drm_kms_helper cryptd snd_pcm drm r8169 snd_timer i2c_i801 snd mii soundcore i2c_core video usbhid [last unloaded: btrfs]
Mar  9 18:25:01 tux kernel: CPU: 1 PID: 24856 Comm: rm Not tainted 3.19.1 #1
Mar  9 18:25:01 tux kernel: Hardware name: System manufacturer System Product Name/P8Z68-V LX, BIOS 4105 07/01/2013
Mar  9 18:25:01 tux kernel:  ffffffff8173f375 ffff8800b65d7d78 ffffffff815420fa 0000000000000007
Mar  9 18:25:01 tux kernel:  0000000000000000 ffff8800b65d7db8 ffffffff8104cfea ffff8800c0af4680
Mar  9 18:25:01 tux kernel:  ffff8800c0af4560 ffff8800c0af4560 ffff8800c0af4560 ffff8800c0af4578
Mar  9 18:25:01 tux kernel: Call Trace:
Mar  9 18:25:01 tux kernel:  [<ffffffff815420fa>] dump_stack+0x45/0x57
Mar  9 18:25:01 tux kernel:  [<ffffffff8104cfea>] warn_slowpath_common+0x8a/0xc0
Mar  9 18:25:01 tux kernel:  [<ffffffff8104d0da>] warn_slowpath_null+0x1a/0x20
Mar  9 18:25:01 tux kernel:  [<ffffffff8111faaa>] shmem_evict_inode+0x10a/0x140
Mar  9 18:25:01 tux kernel:  [<ffffffff811809ec>] evict+0xbc/0x1b0
Mar  9 18:25:01 tux kernel:  [<ffffffff811812b5>] iput+0xf5/0x1a0
Mar  9 18:25:01 tux kernel:  [<ffffffff811764ba>] do_unlinkat+0x1ba/0x310
Mar  9 18:25:01 tux kernel:  [<ffffffff81179049>] ? iterate_dir+0x79/0x130
Mar  9 18:25:01 tux kernel:  [<ffffffff811a5612>] ? dnotify_flush+0x42/0x100
Mar  9 18:25:01 tux kernel:  [<ffffffff81163dda>] ? filp_close+0x5a/0x80
Mar  9 18:25:01 tux kernel:  [<ffffffff81176dcb>] SyS_unlinkat+0x1b/0x40
Mar  9 18:25:01 tux kernel:  [<ffffffff81547392>] system_call_fastpath+0x12/0x17
Mar  9 18:25:01 tux kernel: ---[ end trace 4d758c9126672397 ]---

This seems to be related to building kernels & packages on tmpfs; in this case
it was a package post-build cleanup rm that hit it.

The relevant line in shmem_evict_inode() according to [1] is:
..
608    simple_xattrs_free(&info->xattrs);
609    WARN_ON(inode->i_blocks);
610    shmem_free_inode(inode->i_sb);
611    clear_inode(inode);
..

Why would rm trip over existing blocks?

Thanks!

Holger

[1] http://lxr.free-electrons.com/source/mm/shmem.c#L592