[Linux-cachefs] [PATCH 0/8] ceph: fscache support & upstream changes

Sun Sep 8 21:21:37 UTC 2013

David,

I think that change does the trick. I had it running on the same
machine for 5 hours and had the kernel forcefully drop some of the
inodes in the cache (via drop caches) without a crash. I'll send a
proper patch email after you take a look and make sure I did the right
thing.

Thanks,
- Milosz

On Sat, Sep 7, 2013 at 11:07 PM, Milosz Tanski <milosz at adfin.com> wrote:
> David,
>
> I ran into another issue that caused one my machines to hang on a
> bunch of tasks and then hard lock. Here's the backtrace of the hang:
>
> INFO: task kworker/1:2:4214 blocked for more than 120 seconds.
> "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> kworker/1:2     D ffff880443513fc0     0  4214      2 0x00000000
> Workqueue: ceph-msgr con_work [libceph]
>  ffff88042b093868 0000000000000246 ffff88042b8e5bc0 ffffffff81569fc6
>  ffff88042c51dbc0 ffff88042b093fd8 ffff88042b093fd8 ffff88042b093fd8
>  ffff88042c518000 ffff88042c51dbc0 ffff8804266b8d10 ffff8804439d7188
> Call Trace:
>  [<ffffffff81569fc6>] ? _raw_spin_unlock_irqrestore+0x16/0x20
>  [<ffffffffa0016570>] ? fscache_wait_bit_interruptible+0x30/0x30 [fscache]
>  [<ffffffff81568d09>] schedule+0x29/0x70
>  [<ffffffffa001657e>] fscache_wait_atomic_t+0xe/0x20 [fscache]
>  [<ffffffff815665cf>] out_of_line_wait_on_atomic_t+0x9f/0xe0
>  [<ffffffff81083560>] ? autoremove_wake_function+0x40/0x40
>  [<ffffffffa0015a9c>] __fscache_relinquish_cookie+0x15c/0x310 [fscache]
>  [<ffffffffa00a4fae>] ceph_fscache_unregister_inode_cookie+0x3e/0x50 [ceph]
>  [<ffffffffa007e373>] ceph_destroy_inode+0x33/0x200 [ceph]
>  [<ffffffff811c13ae>] ? __fsnotify_inode_delete+0xe/0x10
>  [<ffffffff8119ba1c>] destroy_inode+0x3c/0x70
>  [<ffffffff8119bb69>] evict+0x119/0x1b0
>  [<ffffffff8119c3f3>] iput+0x103/0x190
>  [<ffffffffa009aaed>] iterate_session_caps+0x7d/0x240 [ceph]
>  [<ffffffffa009b170>] ? remove_session_caps_cb+0x270/0x270 [ceph]
>  [<ffffffffa00a1fc5>] dispatch+0x725/0x1b40 [ceph]
>  [<ffffffff81459466>] ? kernel_recvmsg+0x46/0x60
>  [<ffffffffa002c0e8>] ? ceph_tcp_recvmsg+0x48/0x60 [libceph]
>  [<ffffffffa002ecbe>] try_read+0xc1e/0x1e70 [libceph]
>  [<ffffffffa0030015>] con_work+0x105/0x1920 [libceph]
>  [<ffffffff8100349e>] ? xen_end_context_switch+0x1e/0x30
>  [<ffffffff8108dbca>] ? finish_task_switch+0x5a/0xc0
>  [<ffffffff8107aa59>] process_one_work+0x179/0x490
>  [<ffffffff8107bf5b>] worker_thread+0x11b/0x370
>  [<ffffffff8107be40>] ? manage_workers.isra.21+0x2e0/0x2e0
>  [<ffffffff81082a80>] kthread+0xc0/0xd0
>  [<ffffffff81010000>] ? perf_trace_xen_mmu_set_pud+0xd0/0xd0
>  [<ffffffff810829c0>] ? flush_kthread_worker+0xb0/0xb0
>  [<ffffffff81572cec>] ret_from_fork+0x7c/0xb0
>  [<ffffffff810829c0>] ? flush_kthread_worker+0xb0/0xb0
>
> It looks like it's waiting for the the cookie's n_active to drop down
> to 0 ... but it isn't. After spending a bunch of hours reading the
> code, then having a some beers (it is Saturday night after all), then
> looking at code again... I think that the
> __fscache_check_consistency() function increments the n_active counter
> but never lowers it. I think the solution to this is the bellow diff
> but I'm not a 100% sure. Can you let me know if I'm on the right
> track... of it's beer googles?
>
> diff --git a/fs/fscache/cookie.c b/fs/fscache/cookie.c
> index 318e843..b2a86e3 100644
> --- a/fs/fscache/cookie.c
> +++ b/fs/fscache/cookie.c
> @@ -586,7 +586,8 @@ int __fscache_check_consistency(struct
> fscache_cookie *cookie)
>
>         fscache_operation_init(op, NULL, NULL);
>         op->flags = FSCACHE_OP_MYTHREAD |
> -               (1 << FSCACHE_OP_WAITING);
> +               (1 << FSCACHE_OP_WAITING) |
> +               (1 << FSCACHE_OP_UNUSE_COOKIE);
>
>         spin_lock(&cookie->lock);
>
> Thanks,
> - Milosz
>
> On Fri, Sep 6, 2013 at 4:03 PM, Sage Weil <sage at inktank.com> wrote:
>> On Fri, 6 Sep 2013, Milosz Tanski wrote:
>>> Sage,
>>>
>>> I've taken David's latest changes and per his request merged his
>>> 'fscache-fixes-for-ceph' tag then applied my changes on top of that.
>>> In addition to the pervious changes I also added a fix for the
>>> warnings the linux-next build bot found.
>>>
>>> I've given the results a quick test to make sure it builds, boots and
>>> runs okay. The code is located in my repository:
>>>
>>>   https://adfin@bitbucket.org/adfin/linux-fs.git in the wip-fscache-v2 branch
>>>
>>> I hope that this is the final go for now and thanks for everyone's patience.
>>
>> Looks good; I'll send this to Linus along with the other ceph patches
>> shortly.
>>
>> Thanks, everyone!
>> sage
>>
>>
>>>
>>>  - Milosz
>>>
>>> On Fri, Sep 6, 2013 at 11:59 AM, David Howells <dhowells at redhat.com> wrote:
>>> > Milosz Tanski <milosz at adfin.com> wrote:
>>> >
>>> >> After running this for a day on some loaded machines I ran into what
>>> >> looks like an old issue with the new code. I remember you saw an issue
>>> >> that manifested it self in a similar way a while back.
>>> >>
>>> >> [13837253.462779] FS-Cache: Assertion failed
>>> >> [13837253.462782] 3 == 5 is false
>>> >> [13837253.462807] ------------[ cut here ]------------
>>> >> [13837253.462811] kernel BUG at fs/fscache/operation.c:414!
>>> >
>>> > Bah.
>>> >
>>> > I forgot to call fscache_op_complete().  Patch updated and repushed.
>>> >
>>> > Btw, I've reordered the patches to put the CIFS patch last.  Can you merge the
>>> > patches prior to the CIFS commit from my branch rather than cherry picking
>>> > them so that if they go via two different routes, GIT will handle the merge
>>> > correctly?  I've stuck a tag on it (fscache-fixes-for-ceph) to make that
>>> > easier for you.
>>> >
>>> > I've also asked another RH engineer to try doing some basic testing on the
>>> > CIFS stuff - which may validate the fscache_readpages_cancel patch.
>>> >
>>> > David
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
>>> the body of a message to majordomo at vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>>>