[4.7-rc3 June 16th code][EXT4] Rename file corruption?

[4.7-rc3 June 16th code][EXT4] Rename file corruption?

[Shawn Starr]

Hello Kernel devs,

I built a snapshot of Linus's kernel (master)

Triggered fault in EXT4 compiling LLVM/clang/Mesa:

Jun 15 12:36:41 segfault kernel: [34407.966896] Modules linked in: bnep bluetooth cpufreq_stats ebtable_filter ebtables ip6table_filter ip6_tables bridge amdkfd amd_iommu_v2 amdgpu intel_rapl x86_pkg_temp_thermal intel_powerclamp kvm_intel arc4 iwlmvm kvm coretemp snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic mac80211 uvcvideo snd_hda_intel snd_hda_codec videobuf2_vmalloc videobuf2_memops videobuf2_v4l2 vfat snd_usb_audio irqbypass crct10dif_pclmul videobuf2_core snd_soc_rt5640 snd_soc_rl6231 i2c_algo_bit snd_soc_core snd_usbmidi_lib snd_compress crc32_pclmul videodev ghash_clmulni_intel fat ttm snd_hda_core snd_pcm_dmaengine ac97_bus snd_rawmidi snd_hwdep snd_seq intel_cstate drm_kms_helper intel_uncore drm dell_wmi mei_wdt sparse_keymap iwlwifi snd_seq_device media snd_pcm iTCO_wdt dell_laptop dell_smbios iTCO_vendor_support snd_timer dcdbas ppdev dell_smm_hwmon cfg80211 snd pcspkr joydev elan_i2c ie31200_edac wmi i2c_designware_platform edac_core mei_me shpchp dw_dmac parport_pc parport mei i2c_designware_core lpc_ich video snd_soc_sst_acpi snd_soc_sst_match soundcore dw_dmac_core i2c_i801 tpm_tis spi_pxa2xx_platform tpm dell_rbtn rfkill dell_smo8800 nfsd auth_rpcgss nfs_acl lockd grace sunrpc binfmt_misc 8021q garp stp llc netconsole mrp sdhci_pci crc32c_intel sdhci_acpi sdhci serio_raw mmc_core fjes i2c_hid e1000e ptp pps_core
Jun 15 12:36:41 segfault kernel: [34407.973782] CPU: 3 PID: 12388 Comm: gcc Not tainted 4.7.0-rc3+ #1
Jun 15 12:36:41 segfault kernel: [34407.974026] Hardware name: Dell Inc. Precision M6800/05NG6V, BIOS A16 12/01/2015
Jun 15 12:36:41 segfault kernel: [34407.974289]  0000000000000286 0000000040a52c61 ffff88073fe9bc90 ffffffff813d07af
Jun 15 12:36:41 segfault kernel: [34407.974576]  0000000000000000 0000000000000000 ffff88073fe9bcd0 ffffffff8109f53b
Jun 15 12:36:41 segfault kernel: [34407.974864]  000001183fe9bcf8 ffff880487740928 ffff880487740928 0000000000000000
Jun 15 12:36:41 segfault kernel: [34407.975150] Call Trace:
Jun 15 12:36:41 segfault kernel: [34407.975246]  [<ffffffff813d07af>] dump_stack+0x63/0x84
Jun 15 12:36:41 segfault kernel: [34407.975432]  [<ffffffff8109f53b>] __warn+0xcb/0xf0
Jun 15 12:36:41 segfault kernel: [34407.975605]  [<ffffffff8109f66d>] warn_slowpath_null+0x1d/0x20
Jun 15 12:36:41 segfault kernel: [34407.975817]  [<ffffffff8125b76e>] drop_nlink+0x3e/0x50
Jun 15 12:36:41 segfault kernel: [34407.976068]  [<ffffffff812e2080>] ext4_rename+0x530/0x910
Jun 15 12:36:41 segfault kernel: [34407.976332]  [<ffffffff81249a7e>] ? terminate_walk+0x8e/0xf0
Jun 15 12:36:41 segfault kernel: [34407.976607]  [<ffffffff812e247d>] ext4_rename2+0x1d/0x30
Jun 15 12:36:41 segfault kernel: [34407.976866]  [<ffffffff8124dcd3>] vfs_rename+0x5a3/0x900
Jun 15 12:36:41 segfault kernel: [34407.977125]  [<ffffffff8124ad01>] ? trailing_symlink+0x181/0x220
Jun 15 12:36:41 segfault kernel: [34407.977416]  [<ffffffff81252337>] SyS_rename+0x3a7/0x3d0
Jun 15 12:36:41 segfault kernel: [34407.977679]  [<ffffffff817d88b2>] entry_SYSCALL_64_fastpath+0x1a/0xa4
Jun 15 12:36:41 segfault kernel: [34407.978014] ---[ end trace 812bc2d188b90844 ]---
Jun 15 12:36:45 segfault kernel: [34412.076658] ------------[ cut here ]------------
Jun 15 12:36:45 segfault kernel: [34412.076897] WARNING: CPU: 2 PID: 13368 at fs/inode.c:280 drop_nlink+0x3e/0x50

On shutdown, had inode corruption, rebooted had to fsck to repair.

Thanks,
Shawn

Re: [4.7-rc3 June 16th code][EXT4] Rename file corruption?

[Christoph Hellwig]

We see somewhat similar rename corruption on XFS, also under heavy
compile workloads.  The reporters have already ruled out the XFS changes
from the 4.7 merge window, so this looks a lot like the VFS changes
that went into this merge window.

Re: [4.7-rc3 June 16th code][EXT4] Rename file corruption?

[Al Viro]

On Thu, Jun 16, 2016 at 01:47:57AM -0700, Christoph Hellwig wrote:
> We see somewhat similar rename corruption on XFS, also under heavy
> compile workloads.  The reporters have already ruled out the XFS changes
> from the 4.7 merge window, so this looks a lot like the VFS changes
> that went into this merge window.

Can that be triggered on 4e82901 + cherry-pick of e7d6ef979?
Same for 9f541801 + e7d6ef979?