[LU-5894] DNE II Testing: 2.4/2.5 clients unable to create remote directories against current master servers

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.6.0
• Fix Version/s: Lustre 2.7.0
• Type: Bug
• Priority: Major
• Reporter: Patrick Farrell (Inactive)
• Assignee: Di Wang
• Resolution: Fixed

Environment

2.4/2.5 clients, master servers (from last week)

• Severity: 3

Description

2.4 and 2.5 clients receive -524, which is ENOTSUPP, when trying to create remote (NOT striped) directories on current master.

Centclient19 is running 2.5.0, the servers are running reasonably current master:
[root@centclient19 test]# lfs mkdir -i 1 test
error on LL_IOC_LMV_SETSTRIPE 'test' (3): Unknown error 524
error: mkdir: create stripe dir 'test' failed

Attached is the log from the MDS responsible for returning the error and the log from the client receiving it. (centss02 is the server, centclient19 is the client)

Comments

[Di Wang]

The problem probably comes from here

static int mdt_remote_permission(struct mdt_thread_info *info,
                                 struct mdt_object *parent,
                                 struct mdt_object *child)
{
............

 if (S_ISDIR(attr->la_mode) && spec->u.sp_ea.eadata != NULL &&
            spec->u.sp_ea.eadatalen != 0 && !mdt_is_striped_client(exp))
                return -ENOTSUPP;                        
}

Hmm, we should allow if the stripe_count == 0, I will cook a patch.

[Di Wang]

http://review.whamcloud.com/12715

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/12715/
Subject: LU-5894 mds: allow 2.4/2.5 clients create remote dir
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 1a165c02f5e97c830800d6dcbc11e86790fc015d

[Jodi Levi (Inactive)]

Patch landed to Master.

[Frank Heckes (Inactive)]

A similar error (also using non-root account) appeared during soak testing using build https://build.hpdd.intel.com/job/lustre-b2_5_fe/38/ :

[root@lola-16 soaked]# mkdir parent
[root@lola-16 soaked]# lfs mkdir -i 0 parent/child_0
error on LL_IOC_LMV_SETSTRIPE 'parent/child_0' (3): Invalid argument
error: mkdir: create stripe dir 'parent/child_0' failed
[root@lola-16 soaked]# lfs mkdir -i 1 parent/child_1
error on LL_IOC_LMV_SETSTRIPE 'parent/child_1' (3): Invalid argument
error: mkdir: create stripe dir 'parent/child_1' failed
[root@lola-16 soaked]# Write failed: Broken pipe

dmesg entry on client related to the event, reads as

LustreError: 135586:0:(dir.c:1166:ll_dir_ioctl()) child_0: wrong lum magic cd20cd0 or size 48: rc = -14

No error messages on the server side.

[Patrick Farrell (Inactive)]

Frank - I suspect you'll want to open a new ticket for that error. This ticket is for a specific error which was seen with particular client versions & was resolved. (Also, the original report is with root, not a non-root user.)

[Frank Heckes (Inactive)]

Please ignore my comment above as it was caused by installation error for a single client. Actually the node was running build artifacts of master branch still.
I'm sorry for the spam and any inconvenience.