Possible memory allocation deadlock

[edillmann]

Hi,

Is this a new bug ?
I found a lot of oops like this on the receiving side of a snapshot transfert.
I'm using latest master on kernel 3.18.3

[41374.265342] Possible memory allocation deadlock: size=1048576 lflags=0x42d0
[41374.265345] CPU: 2 PID: 71096 Comm: zfs Tainted: P OE 3.18.3-jave #1
[41374.265347] Hardware name: /DH67BL, BIOS BLH6710H.86A.0160.2012.1204.1156 12/04/2012
[41374.265349] 0000000000002000 ffff8801e431ba48 ffffffff8171de28 0000000000000a3c
[41374.265352] 00000000000042d0 ffff8801e431ba88 ffffffffa0427f74 ffff8801e431ba58
[41374.265355] ffff8801e431bcb0 ffff8800b7127550 0000000000000000 ffff8801e431bca0
[41374.265358] Call Trace:
[41374.265363] [] dump_stack+0x46/0x58
[41374.265371] [] spl_kmem_alloc_debug+0x184/0x190 [spl]
[41374.265378] [] spl_vmem_alloc+0x19/0x20 [spl]
[41374.265403] [] dmu_recv_stream+0xa3/0xb30 [zfs]
[41374.265411] [] ? nvlist_common.part.102+0xe2/0x1e0 [znvpair]
[41374.265419] [] ? nvlist_xpack+0xde/0x110 [znvpair]
[41374.265425] [] ? nvlist_common.part.102+0xe2/0x1e0 [znvpair]
[41374.265431] [] ? spl_kmem_free+0x32/0x50 [spl]
[41374.265438] [] ? fnvlist_pack_free+0xe/0x10 [znvpair]
[41374.265476] [] ? put_nvlist+0x91/0xa0 [zfs]
[41374.265513] [] zfs_ioc_recv+0x1ec/0xbd0 [zfs]
[41374.265532] [] ? dbuf_rele_and_unlock+0x2a0/0x390 [zfs]
[41374.265538] [] ? spl_kmem_free+0x32/0x50 [spl]
[41374.265547] [] ? tsd_set+0x67/0x2e0 [spl]
[41374.265554] [] ? tsd_hash_search.isra.1+0x77/0xa0 [spl]
[41374.265585] [] ? rrw_exit+0x51/0x160 [zfs]
[41374.265589] [] ? __kmalloc+0x55/0x230
[41374.265595] [] ? strdup+0x3c/0x60 [spl]
[41374.265631] [] zfsdev_ioctl+0x446/0x470 [zfs]
[41374.265635] [] do_vfs_ioctl+0x2e0/0x4c0
[41374.265638] [] ? vtime_account_user+0x54/0x60
[41374.265641] [] SyS_ioctl+0x81/0xa0
[41374.265644] [] ? int_check_syscall_exit_work+0x34/0x3d
[41374.265666] [] system_call_fastpath+0x16/0x1b

[behlendorf]

It's new, but it's not an OOPS. Just a warning introduced by the recent kmem rework. We expected a couple of these to put up, thanks for reporting it so we can resolve it before the tag.

[behlendorf]

@edillmann I should mention if you need a fix now you could set spl_kmem_alloc_max=65536 on the live system which should resolve this.

[behlendorf]

@edillmann are you using the large block patches as well?

[edillmann]

@behlendorf I don't use large block patches for now, I did set spl_kmem_alloc_max to 65536 and will report if it resolv this.

[behlendorf]

This should be resolved by openzfs/spl@c7db36a which has been merged to master. The excessively large allocation will automatically retry and use vmalloc() as required.