[LU-16376] lprocfs_job_stats_log() Invalid jobid size (37), expect(32)

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.12.8, Lustre 2.15.2
• Fix Version/s: Lustre 2.16.0, Lustre 2.15.3
• Type: Bug
• Priority: Minor
• Reporter: Andreas Dilger
• Assignee: Andreas Dilger
• Resolution: Fixed
• Severity: 3

Description

Regular error messages on the console:

kernel: LustreError: 8537:0:(lprocfs_jobstats.c:283:lprocfs_job_stats_log()) Invalid jobid size (37), expect(32)

While this ostensibly seems like the JobID is too large, the problem is that lprocfs_job_stats_log() is only called on the MDS and OSS using the jobid extracted from the RPC in ptlrpc_body->pb_jobid[32] so it is impossible to be larger than 32 bytes.

/* 31 usable bytes string + null terminator. */ 
#define LUSTRE_JOBID_SIZE       32

int lprocfs_job_stats_log(struct obd_device *obd, char *jobid,
                          int event, long amount)
{
        if (strlen(jobid) >= LUSTRE_JOBID_SIZE) {
                CERROR("Invalid jobid size (%lu), expect(%d)\n",
                       (unsigned long)strlen(jobid) + 1, LUSTRE_JOBID_SIZE);
                RETURN(-EINVAL);
        }
}

void mdt_counter_incr(struct ptlrpc_request *req, int opcode, long amount)
{
        if (exp->exp_obd && obd2obt(exp->exp_obd)->obt_jobstats.ojs_hash &&
            (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
                lprocfs_job_stats_log(exp->exp_obd,
                                      lustre_msg_get_jobid(req->rq_reqmsg),
                                      opcode, amount);
}

static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
                                    char *jobid, long amount)
{
        if (exp->exp_obd && obd2obt(exp->exp_obd)->obt_jobstats.ojs_hash &&
            (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
                lprocfs_job_stats_log(exp->exp_obd, jobid, opcode, amount);
}

One likely possibility is that pb_jobid is not NUL terminated, and the strlen() overflows into the next bytes in the RPC.

Rather than repeatedly complain about this and not do anything, it would be better to just NUL terminate the string (as it should be) and then proceed with the truncated JobId

Issue Links

• is related to: LU-17512 add conditional operator for 'jobid_n...
• is related to: LU-16599 clearing jobstats should match output...

Comments

[Gerrit Updater]

"Andreas Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/49351
Subject: LU-16376 obdclass: NUL terminate log jobid strings
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 754e05e41da401627652bb88571ab101baba590e

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/49351/
Subject: LU-16376 obdclass: NUL terminate long jobid strings
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 9eba5d57297f807fddf046356c846478bbf232f4

[Peter Jones]

Landed for 2.16

[Gerrit Updater]

"Andreas Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/49490
Subject: LU-16376 obdclass: NUL terminate long jobid strings
Project: fs/lustre-release
Branch: b2_15
Current Patch Set: 1
Commit: 3ce08283b9a5ea0b50c68874a6acd38b84c23dc6

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/49490/
Subject: LU-16376 obdclass: NUL terminate long jobid strings
Project: fs/lustre-release
Branch: b2_15
Current Patch Set:
Commit: 727638a1d0e72a043c2798f081f166a0e3a39268