[LU-13680] large allocations in osd_bufs_get() failing

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.14.0
• Severity: 3
• Assignee: Andreas Dilger
• Reporter: Andreas Dilger

Description

Large allocations in osd_bufs_get() can fail if the OSS memory is fragmented after use, if that thread has not serviced a cacheless IO since startup or is newly started:

kernel: ll_ost_io04_052: page allocation failure: order:5, mode:0x10c050
kernel: CPU: 4 PID: 1980 Comm: ll_ost_io04_052 
Call Trace:
dump_stack+0x19/0x1b
warn_alloc_failed+0x110/0x180
__alloc_pages_slowpath+0x6b6/0x724
__alloc_pages_nodemask+0x404/0x420
alloc_pages_current+0x98/0x110
__get_free_pages+0xe/0x40
kmalloc_order_trace+0x2e/0xa0
osd_bufs_get+0x7b7/0x870 [osd_ldiskfs]
ofd_preprw_read+0x2ea/0x1110 [ofd]
ofd_preprw+0x499/0x8c0 [ofd]
tgt_brw_read+0x9e3/0x1e40 [ptlrpc]
tgt_request_handle+0xada/0x1570 [ptlrpc]
ptlrpc_server_handle_request+0x24b/0xab0 [ptlrpc]
ptlrpc_main+0xb34/0x1470 [ptlrpc]

The order 5 allocation is 128KB, so osd_bufs_get() should be using OBD_ALLOC_LARGE() or equivalent.

Issue Links

• is related to: LU-11347 Do not use pagecache for SSD I/O when read/write cache are disabled
• is related to: LU-12071 bypass pagecache for large files