[LU-12140] DOM: limitaion of data size

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.13.0
• Affects Version/s: Lustre 2.12.1
• Environment: configuration with router, MDS_FS_MKFS_OPTS="-O large_xattr", sanity-dom/sanityn

Epic/Theme

lnet

• Severity: 3
• Assignee: Mikhail Pershin
• Reporter: Alexander Boyko

Description

During sanity-dom testing the next issue was appeared:

 [ 1644.726837] LNetError: 3137:0:(lib-move.c:4143:lnet_parse()) 192.168.8.1@tcp, src 192.168.8.1@tcp: bad PUT payload 1051832 (1048576 max expected)

I've added a bit debug to take vmcores from a sender.

Here is analyze from crash

 
md = {
     start = 0xffff880098200100,
     length = 1051832,
     threshold = 0,
     max_size = -1742733056,
     options = 0,
     user_ptr = 0xffff880098200000,
     eq_handle = {
       cookie = 23
     },
     bulk_handle = {
       cookie = 0
     }
   },
msg_niov = 1,
 msg_iov = 0xffff88009995aba0,
 
msg_kiov = 0x0,
 
ffff880135be9800
rc_fmt = 0xffffffffc095d080 <RQF_LDLM_INTENT_OPEN>,
 
static const struct req_msg_field *ldlm_intent_open_server[] = {
       &RMF_PTLRPC_BODY,
       &RMF_DLM_REP,
       &RMF_MDT_BODY,
       &RMF_MDT_MD,
       &RMF_ACL,
       &RMF_CAPA1,
       &RMF_CAPA2,
       &RMF_NIOBUF_INLINE,
};
rc_area = {{4294967295, 4294967295, 4294967295, 4294967295, 4294967295, 4294967295, 4294967295, 4294967295, 4294967295, 4294967295}, {184, 112, 216, 2432, 260, 0, 0, 1048592, 4294967295, 4294967295}}
 }
 
crash> p 184+112+216+2432+260+1048592
$15 = 1051796
 

The DOM size during open was 1Mb, the total length of lnet request was 1051796, and it doesn't fit at LNET_MTU limit. So the router shows error.
This brings us to problem when we cannot handle 1Mb stripe size DOM at LNET layer. I think it is a problem for PFL when a first stripe located at MDS, probably.
The workaround for sanity-dom testing is to decrease DOM_SIZE at sanity-dom.sh
Also MDS should limit this size to prevent such misbehavior.
I've assigned this to Mikhail, I'm not sure.