lustre/obdclass/jobid.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2014, Intel Corporation.
  27  *
  28  * Copyright 2017 Cray Inc, all rights reserved.
  29  * Author: Ben Evans.
  30  *
  31  * Store PID->JobID mappings
  32  */
  33
  34 #define DEBUG_SUBSYSTEM S_RPC
  35 #include <linux/user_namespace.h>
  36 #ifdef HAVE_UIDGID_HEADER
  37 #include <linux/uidgid.h>
  38 #endif
  39 #include <linux/utsname.h>
  40
  41 #include <libcfs/libcfs.h>
  42 #include <obd_support.h>
  43 #include <obd_class.h>
  44 #include <lustre_net.h>
  45
  46 static struct cfs_hash *jobid_hash;
  47 static struct cfs_hash_ops jobid_hash_ops;
  48 spinlock_t jobid_hash_lock;
  49
  50 #define RESCAN_INTERVAL 30
  51 #define DELETE_INTERVAL 300
  52
  53 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
  54 char obd_jobid_name[LUSTRE_JOBID_SIZE] = "%e.%u";
  55
  56 /**
  57  * Structure to store a single PID->JobID mapping
  58  */
  59 struct jobid_pid_map {
  60         struct hlist_node       jp_hash;
  61         time64_t                jp_time;
  62         spinlock_t              jp_lock; /* protects jp_jobid */
  63         char                    jp_jobid[LUSTRE_JOBID_SIZE];
  64         unsigned int            jp_joblen;
  65         atomic_t                jp_refcount;
  66         pid_t                   jp_pid;
  67 };
  68
  69 /*
  70  * Jobid can be set for a session (see setsid(2)) by writing to
  71  * a sysfs file from any process in that session.
  72  * The jobids are stored in a hash table indexed by the relevant
  73  * struct pid.  We periodically look for entries where the pid has
  74  * no PIDTYPE_SID tasks any more, and prune them.  This happens within
  75  * 5 seconds of a jobid being added, and every 5 minutes when jobids exist,
  76  * but none are added.
  77  */
  78 #define JOBID_EXPEDITED_CLEAN (5)
  79 #define JOBID_BACKGROUND_CLEAN (5 * 60)
  80
  81 struct session_jobid {
  82         struct pid              *sj_session;
  83         struct rhash_head       sj_linkage;
  84         struct rcu_head         sj_rcu;
  85         char                    sj_jobid[1];
  86 };
  87
  88 static const struct rhashtable_params jobid_params = {
  89         .key_len        = sizeof(struct pid *),
  90         .key_offset     = offsetof(struct session_jobid, sj_session),
  91         .head_offset    = offsetof(struct session_jobid, sj_linkage),
  92 };
  93
  94 static struct rhashtable session_jobids;
  95
  96 /*
  97  * jobid_current must be called with rcu_read_lock held.
  98  * if it returns non-NULL, the string can only be used
  99  * until rcu_read_unlock is called.
 100  */
 101 char *jobid_current(void)
 102 {
 103         struct pid *sid = task_session(current);
 104         struct session_jobid *sj;
 105
 106         sj = rhashtable_lookup_fast(&session_jobids, &sid, jobid_params);
 107         if (sj)
 108                 return sj->sj_jobid;
 109         return NULL;
 110 }
 111
 112 static void jobid_prune_expedite(void);
 113 /*
 114  * jobid_set_current will try to add a new entry
 115  * to the table.  If one exists with the same key, the
 116  * jobid will be replaced
 117  */
 118 int jobid_set_current(char *jobid)
 119 {
 120         struct pid *sid;
 121         struct session_jobid *sj, *origsj;
 122         int ret;
 123         int len = strlen(jobid);
 124
 125         sj = kmalloc(sizeof(*sj) + len, GFP_KERNEL);
 126         if (!sj)
 127                 return -ENOMEM;
 128         rcu_read_lock();
 129         sid = task_session(current);
 130         sj->sj_session = get_pid(sid);
 131         strncpy(sj->sj_jobid, jobid, len+1);
 132         origsj = rhashtable_lookup_get_insert_fast(&session_jobids,
 133                                                    &sj->sj_linkage,
 134                                                    jobid_params);
 135         if (origsj == NULL) {
 136                 /* successful insert */
 137                 rcu_read_unlock();
 138                 jobid_prune_expedite();
 139                 return 0;
 140         }
 141
 142         if (IS_ERR(origsj)) {
 143                 put_pid(sj->sj_session);
 144                 kfree(sj);
 145                 rcu_read_unlock();
 146                 return PTR_ERR(origsj);
 147         }
 148         ret = rhashtable_replace_fast(&session_jobids,
 149                                       &origsj->sj_linkage,
 150                                       &sj->sj_linkage,
 151                                       jobid_params);
 152         if (ret) {
 153                 put_pid(sj->sj_session);
 154                 kfree(sj);
 155                 rcu_read_unlock();
 156                 return ret;
 157         }
 158         put_pid(origsj->sj_session);
 159         rcu_read_unlock();
 160         kfree_rcu(origsj, sj_rcu);
 161         jobid_prune_expedite();
 162
 163         return 0;
 164 }
 165
 166 static void jobid_free(void *vsj, void *arg)
 167 {
 168         struct session_jobid *sj = vsj;
 169
 170         put_pid(sj->sj_session);
 171         kfree(sj);
 172 }
 173
 174 static void jobid_prune(struct work_struct *work);
 175 static DECLARE_DELAYED_WORK(jobid_prune_work, jobid_prune);
 176 static int jobid_prune_expedited;
 177 static void jobid_prune(struct work_struct *work)
 178 {
 179         int remaining = 0;
 180         struct rhashtable_iter iter;
 181         struct session_jobid *sj;
 182
 183         jobid_prune_expedited = 0;
 184         rhashtable_walk_enter(&session_jobids, &iter);
 185         rhashtable_walk_start(&iter);
 186         while ((sj = rhashtable_walk_next(&iter)) != NULL) {
 187                 if (!hlist_empty(&sj->sj_session->tasks[PIDTYPE_SID])) {
 188                         remaining++;
 189                         continue;
 190                 }
 191                 if (rhashtable_remove_fast(&session_jobids,
 192                                            &sj->sj_linkage,
 193                                            jobid_params) == 0) {
 194                         put_pid(sj->sj_session);
 195                         kfree_rcu(sj, sj_rcu);
 196                 }
 197         }
 198         rhashtable_walk_stop(&iter);
 199         rhashtable_walk_exit(&iter);
 200         if (remaining)
 201                 schedule_delayed_work(&jobid_prune_work,
 202                                       cfs_time_seconds(JOBID_BACKGROUND_CLEAN));
 203 }
 204
 205 static void jobid_prune_expedite(void)
 206 {
 207         if (!jobid_prune_expedited) {
 208                 jobid_prune_expedited = 1;
 209                 mod_delayed_work(system_wq, &jobid_prune_work,
 210                                  cfs_time_seconds(JOBID_EXPEDITED_CLEAN));
 211         }
 212 }
 213
 214 /*
 215  * Get jobid of current process by reading the environment variable
 216  * stored in between the "env_start" & "env_end" of task struct.
 217  *
 218  * If some job scheduler doesn't store jobid in the "env_start/end",
 219  * then an upcall could be issued here to get the jobid by utilizing
 220  * the userspace tools/API. Then, the jobid must be cached.
 221  */
 222 int jobid_get_from_environ(char *jobid_var, char *jobid, int *jobid_len)
 223 {
 224         int rc;
 225
 226         rc = cfs_get_environ(jobid_var, jobid, jobid_len);
 227         if (!rc)
 228                 goto out;
 229
 230         if (rc == -EOVERFLOW) {
 231                 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
 232                  * variable length strings instead of just numbers), it
 233                  * might make sense to keep the unique parts for JobID,
 234                  * instead of just returning an error.  That means a
 235                  * larger temp buffer for cfs_get_environ(), then
 236                  * truncating the string at some separator to fit into
 237                  * the specified jobid_len.  Fix later if needed. */
 238                 static ktime_t printed;
 239
 240                 if (unlikely(ktime_to_ns(printed) == 0 ||
 241                              ktime_after(ktime_get(),
 242                                          ktime_add_ns(printed,
 243                                                       3600*24*NSEC_PER_SEC)))) {
 244                         LCONSOLE_WARN("jobid: '%s' value too large (%d)\n",
 245                                       obd_jobid_var, *jobid_len);
 246                         printed = ktime_get();
 247                 }
 248
 249                 rc = 0;
 250         } else {
 251                 CDEBUG_LIMIT((rc == -ENOENT || rc == -EINVAL ||
 252                               rc == -EDEADLK) ? D_INFO : D_ERROR,
 253                              "jobid: get '%s' failed: rc = %d\n",
 254                              obd_jobid_var, rc);
 255         }
 256
 257 out:
 258         return rc;
 259 }
 260
 261 /*
 262  * jobid_should_free_item
 263  *
 264  * Each item is checked to see if it should be released
 265  * Removed from hash table by caller
 266  * Actually freed in jobid_put_locked
 267  *
 268  * Returns 1 if item is to be freed, 0 if it is to be kept
 269  */
 270
 271 static int jobid_should_free_item(void *obj, void *data)
 272 {
 273         char *jobid = data;
 274         struct jobid_pid_map *pidmap = obj;
 275         int rc = 0;
 276
 277         if (obj == NULL)
 278                 return 0;
 279
 280         if (jobid == NULL) {
 281                 WARN_ON_ONCE(atomic_read(&pidmap->jp_refcount) != 1);
 282                 return 1;
 283         }
 284
 285         spin_lock(&pidmap->jp_lock);
 286         /* prevent newly inserted items from deleting */
 287         if (jobid[0] == '\0' && atomic_read(&pidmap->jp_refcount) == 1)
 288                 rc = 1;
 289         else if (ktime_get_real_seconds() - pidmap->jp_time > DELETE_INTERVAL)
 290                 rc = 1;
 291         else if (strcmp(pidmap->jp_jobid, jobid) == 0)
 292                 rc = 1;
 293         spin_unlock(&pidmap->jp_lock);
 294
 295         return rc;
 296 }
 297
 298 /*
 299  * jobid_name_is_valid
 300  *
 301  * Checks if the jobid is a Lustre process
 302  *
 303  * Returns true if jobid is valid
 304  * Returns false if jobid looks like it's a Lustre process
 305  */
 306 static bool jobid_name_is_valid(char *jobid)
 307 {
 308         const char *const lustre_reserved[] = { "ll_ping", "ptlrpc",
 309                                                 "ldlm", "ll_sa", NULL };
 310         int i;
 311
 312         if (jobid[0] == '\0')
 313                 return false;
 314
 315         for (i = 0; lustre_reserved[i] != NULL; i++) {
 316                 if (strncmp(jobid, lustre_reserved[i],
 317                             strlen(lustre_reserved[i])) == 0)
 318                         return false;
 319         }
 320         return true;
 321 }
 322
 323 /*
 324  * jobid_get_from_cache()
 325  *
 326  * Returns contents of jobid_var from process environment for current PID,
 327  * or from the per-session jobid table.
 328  * Values fetch from process environment will be cached for some time to avoid
 329  * the overhead of scanning the environment.
 330  *
 331  * Return: -ENOMEM if allocating a new pidmap fails
 332  *         -ENOENT if no entry could be found
 333  *         +ve string length for success (something was returned in jobid)
 334  */
 335 static int jobid_get_from_cache(char *jobid, size_t joblen)
 336 {
 337         static time64_t last_expire;
 338         bool expire_cache = false;
 339         pid_t pid = current_pid();
 340         struct jobid_pid_map *pidmap = NULL;
 341         time64_t now = ktime_get_real_seconds();
 342         int rc = 0;
 343         ENTRY;
 344
 345         if (strcmp(obd_jobid_var, JOBSTATS_SESSION) == 0) {
 346                 char *jid;
 347
 348                 rcu_read_lock();
 349                 jid = jobid_current();
 350                 if (jid) {
 351                         strlcpy(jobid, jid, joblen);
 352                         joblen = strlen(jobid);
 353                 } else {
 354                         rc = -ENOENT;
 355                 }
 356                 rcu_read_unlock();
 357                 GOTO(out, rc);
 358         }
 359
 360         LASSERT(jobid_hash != NULL);
 361
 362         /* scan hash periodically to remove old PID entries from cache */
 363         spin_lock(&jobid_hash_lock);
 364         if (unlikely(last_expire + DELETE_INTERVAL <= now)) {
 365                 expire_cache = true;
 366                 last_expire = now;
 367         }
 368         spin_unlock(&jobid_hash_lock);
 369
 370         if (expire_cache)
 371                 cfs_hash_cond_del(jobid_hash, jobid_should_free_item,
 372                                   "intentionally_bad_jobid");
 373
 374         /* first try to find PID in the hash and use that value */
 375         pidmap = cfs_hash_lookup(jobid_hash, &pid);
 376         if (pidmap == NULL) {
 377                 struct jobid_pid_map *pidmap2;
 378
 379                 OBD_ALLOC_PTR(pidmap);
 380                 if (pidmap == NULL)
 381                         GOTO(out, rc = -ENOMEM);
 382
 383                 pidmap->jp_pid = pid;
 384                 pidmap->jp_time = 0;
 385                 pidmap->jp_jobid[0] = '\0';
 386                 spin_lock_init(&pidmap->jp_lock);
 387                 INIT_HLIST_NODE(&pidmap->jp_hash);
 388                 /*
 389                  * @pidmap might be reclaimed just after it is added into
 390                  * hash list, init @jp_refcount as 1 to make sure memory
 391                  * could be not freed during access.
 392                  */
 393                 atomic_set(&pidmap->jp_refcount, 1);
 394
 395                 /*
 396                  * Add the newly created map to the hash, on key collision we
 397                  * lost a racing addition and must destroy our newly allocated
 398                  * map.  The object which exists in the hash will be returned.
 399                  */
 400                 pidmap2 = cfs_hash_findadd_unique(jobid_hash, &pid,
 401                                                   &pidmap->jp_hash);
 402                 if (unlikely(pidmap != pidmap2)) {
 403                         CDEBUG(D_INFO, "jobid: duplicate found for PID=%u\n",
 404                                pid);
 405                         OBD_FREE_PTR(pidmap);
 406                         pidmap = pidmap2;
 407                 }
 408         }
 409
 410         /*
 411          * If pidmap is old (this is always true for new entries) refresh it.
 412          * If obd_jobid_var is not found, cache empty entry and try again
 413          * later, to avoid repeat lookups for PID if obd_jobid_var missing.
 414          */
 415         spin_lock(&pidmap->jp_lock);
 416         if (pidmap->jp_time + RESCAN_INTERVAL <= now) {
 417                 char env_jobid[LUSTRE_JOBID_SIZE] = "";
 418                 int env_len = sizeof(env_jobid);
 419
 420                 pidmap->jp_time = now;
 421
 422                 spin_unlock(&pidmap->jp_lock);
 423                 rc = jobid_get_from_environ(obd_jobid_var, env_jobid, &env_len);
 424
 425                 CDEBUG(D_INFO, "jobid: PID mapping established: %d->%s\n",
 426                        pidmap->jp_pid, env_jobid);
 427                 spin_lock(&pidmap->jp_lock);
 428                 if (!rc) {
 429                         pidmap->jp_joblen = env_len;
 430                         strlcpy(pidmap->jp_jobid, env_jobid,
 431                                 sizeof(pidmap->jp_jobid));
 432                         rc = 0;
 433                 } else if (rc == -ENOENT) {
 434                         /* It might have been deleted, clear out old entry */
 435                         pidmap->jp_joblen = 0;
 436                         pidmap->jp_jobid[0] = '\0';
 437                 }
 438         }
 439
 440         /*
 441          * Regardless of how pidmap was found, if it contains a valid entry
 442          * use that for now.  If there was a technical error (e.g. -ENOMEM)
 443          * use the old cached value until it can be looked up again properly.
 444          * If a cached missing entry was found, return -ENOENT.
 445          */
 446         if (pidmap->jp_joblen) {
 447                 strlcpy(jobid, pidmap->jp_jobid, joblen);
 448                 joblen = pidmap->jp_joblen;
 449                 rc = 0;
 450         } else if (!rc) {
 451                 rc = -ENOENT;
 452         }
 453         spin_unlock(&pidmap->jp_lock);
 454
 455         cfs_hash_put(jobid_hash, &pidmap->jp_hash);
 456
 457         EXIT;
 458 out:
 459         return rc < 0 ? rc : joblen;
 460 }
 461
 462 /*
 463  * jobid_interpret_string()
 464  *
 465  * Interpret the jobfmt string to expand specified fields, like coredumps do:
 466  *   %e = executable
 467  *   %g = gid
 468  *   %h = hostname
 469  *   %j = jobid from environment
 470  *   %p = pid
 471  *   %u = uid
 472  *
 473  * Unknown escape strings are dropped.  Other characters are copied through,
 474  * excluding whitespace (to avoid making jobid parsing difficult).
 475  *
 476  * Return: -EOVERFLOW if the expanded string does not fit within @joblen
 477  *         0 for success
 478  */
 479 static int jobid_interpret_string(const char *jobfmt, char *jobid,
 480                                   ssize_t joblen)
 481 {
 482         char c;
 483
 484         while ((c = *jobfmt++) && joblen > 1) {
 485                 char f;
 486                 int l;
 487
 488                 if (isspace(c)) /* Don't allow embedded spaces */
 489                         continue;
 490
 491                 if (c != '%') {
 492                         *jobid = c;
 493                         joblen--;
 494                         jobid++;
 495                         continue;
 496                 }
 497
 498                 switch ((f = *jobfmt++)) {
 499                 case 'e': /* executable name */
 500                         l = snprintf(jobid, joblen, "%s", current_comm());
 501                         break;
 502                 case 'g': /* group ID */
 503                         l = snprintf(jobid, joblen, "%u",
 504                                      from_kgid(&init_user_ns, current_fsgid()));
 505                         break;
 506                 case 'h': /* hostname */
 507                         l = snprintf(jobid, joblen, "%s",
 508                                      init_utsname()->nodename);
 509                         break;
 510                 case 'j': /* jobid stored in process environment */
 511                         l = jobid_get_from_cache(jobid, joblen);
 512                         if (l < 0)
 513                                 l = 0;
 514                         break;
 515                 case 'p': /* process ID */
 516                         l = snprintf(jobid, joblen, "%u", current_pid());
 517                         break;
 518                 case 'u': /* user ID */
 519                         l = snprintf(jobid, joblen, "%u",
 520                                      from_kuid(&init_user_ns, current_fsuid()));
 521                         break;
 522                 case '\0': /* '%' at end of format string */
 523                         l = 0;
 524                         goto out;
 525                 default: /* drop unknown %x format strings */
 526                         l = 0;
 527                         break;
 528                 }
 529                 jobid += l;
 530                 joblen -= l;
 531         }
 532         /*
 533          * This points at the end of the buffer, so long as jobid is always
 534          * incremented the same amount as joblen is decremented.
 535          */
 536 out:
 537         jobid[joblen - 1] = '\0';
 538
 539         return joblen < 0 ? -EOVERFLOW : 0;
 540 }
 541
 542 /*
 543  * Hash initialization, copied from server-side job stats bucket sizes
 544  */
 545 #define HASH_JOBID_BKT_BITS 5
 546 #define HASH_JOBID_CUR_BITS 7
 547 #define HASH_JOBID_MAX_BITS 12
 548
 549 int jobid_cache_init(void)
 550 {
 551         int rc = 0;
 552         ENTRY;
 553
 554         if (jobid_hash)
 555                 return 0;
 556
 557         spin_lock_init(&jobid_hash_lock);
 558         jobid_hash = cfs_hash_create("JOBID_HASH", HASH_JOBID_CUR_BITS,
 559                                      HASH_JOBID_MAX_BITS, HASH_JOBID_BKT_BITS,
 560                                      0, CFS_HASH_MIN_THETA, CFS_HASH_MAX_THETA,
 561                                      &jobid_hash_ops, CFS_HASH_DEFAULT);
 562         if (!jobid_hash) {
 563                 rc = -ENOMEM;
 564         } else {
 565                 rc = rhashtable_init(&session_jobids, &jobid_params);
 566                 if (rc) {
 567                         cfs_hash_putref(jobid_hash);
 568                         jobid_hash = NULL;
 569                 }
 570         }
 571
 572         RETURN(rc);
 573 }
 574 EXPORT_SYMBOL(jobid_cache_init);
 575
 576 void jobid_cache_fini(void)
 577 {
 578         struct cfs_hash *tmp_hash;
 579         ENTRY;
 580
 581         spin_lock(&jobid_hash_lock);
 582         tmp_hash = jobid_hash;
 583         jobid_hash = NULL;
 584         spin_unlock(&jobid_hash_lock);
 585
 586         cancel_delayed_work_sync(&jobid_prune_work);
 587
 588         if (tmp_hash != NULL) {
 589                 cfs_hash_cond_del(tmp_hash, jobid_should_free_item, NULL);
 590                 cfs_hash_putref(tmp_hash);
 591
 592                 rhashtable_free_and_destroy(&session_jobids, jobid_free, NULL);
 593         }
 594
 595
 596         EXIT;
 597 }
 598 EXPORT_SYMBOL(jobid_cache_fini);
 599
 600 /*
 601  * Hash operations for pid<->jobid
 602  */
 603 static unsigned jobid_hashfn(struct cfs_hash *hs, const void *key,
 604                              unsigned mask)
 605 {
 606         return cfs_hash_djb2_hash(key, sizeof(pid_t), mask);
 607 }
 608
 609 static void *jobid_key(struct hlist_node *hnode)
 610 {
 611         struct jobid_pid_map *pidmap;
 612
 613         pidmap = hlist_entry(hnode, struct jobid_pid_map, jp_hash);
 614         return &pidmap->jp_pid;
 615 }
 616
 617 static int jobid_keycmp(const void *key, struct hlist_node *hnode)
 618 {
 619         const pid_t *pid_key1;
 620         const pid_t *pid_key2;
 621
 622         LASSERT(key != NULL);
 623         pid_key1 = (pid_t *)key;
 624         pid_key2 = (pid_t *)jobid_key(hnode);
 625
 626         return *pid_key1 == *pid_key2;
 627 }
 628
 629 static void *jobid_object(struct hlist_node *hnode)
 630 {
 631         return hlist_entry(hnode, struct jobid_pid_map, jp_hash);
 632 }
 633
 634 static void jobid_get(struct cfs_hash *hs, struct hlist_node *hnode)
 635 {
 636         struct jobid_pid_map *pidmap;
 637
 638         pidmap = hlist_entry(hnode, struct jobid_pid_map, jp_hash);
 639
 640         atomic_inc(&pidmap->jp_refcount);
 641 }
 642
 643 static void jobid_put_locked(struct cfs_hash *hs, struct hlist_node *hnode)
 644 {
 645         struct jobid_pid_map *pidmap;
 646
 647         if (hnode == NULL)
 648                 return;
 649
 650         pidmap = hlist_entry(hnode, struct jobid_pid_map, jp_hash);
 651         LASSERT(atomic_read(&pidmap->jp_refcount) > 0);
 652         if (atomic_dec_and_test(&pidmap->jp_refcount)) {
 653                 CDEBUG(D_INFO, "Freeing: %d->%s\n",
 654                        pidmap->jp_pid, pidmap->jp_jobid);
 655
 656                 OBD_FREE_PTR(pidmap);
 657         }
 658 }
 659
 660 static struct cfs_hash_ops jobid_hash_ops = {
 661         .hs_hash        = jobid_hashfn,
 662         .hs_keycmp      = jobid_keycmp,
 663         .hs_key         = jobid_key,
 664         .hs_object      = jobid_object,
 665         .hs_get         = jobid_get,
 666         .hs_put         = jobid_put_locked,
 667         .hs_put_locked  = jobid_put_locked,
 668 };
 669
 670 /**
 671  * Generate the job identifier string for this process for tracking purposes.
 672  *
 673  * Fill in @jobid string based on the value of obd_jobid_var:
 674  * JOBSTATS_DISABLE:      none
 675  * JOBSTATS_NODELOCAL:    content of obd_jobid_name (jobid_interpret_string())
 676  * JOBSTATS_PROCNAME_UID: process name/UID
 677  * JOBSTATS_SESSION       per-session value set by
 678  *                            /sys/fs/lustre/jobid_this_session
 679  * anything else:         look up obd_jobid_var in the processes environment
 680  *
 681  * Return -ve error number, 0 on success.
 682  */
 683 int lustre_get_jobid(char *jobid, size_t joblen)
 684 {
 685         int rc = 0;
 686         ENTRY;
 687
 688         if (unlikely(joblen < 2)) {
 689                 if (joblen == 1)
 690                         jobid[0] = '\0';
 691                 RETURN(-EINVAL);
 692         }
 693
 694         if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0) {
 695                 /* Jobstats isn't enabled */
 696                 memset(jobid, 0, joblen);
 697         } else if (strcmp(obd_jobid_var, JOBSTATS_NODELOCAL) == 0) {
 698                 /* Whole node dedicated to single job */
 699                 rc = jobid_interpret_string(obd_jobid_name, jobid, joblen);
 700         } else if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
 701                 rc = jobid_interpret_string("%e.%u", jobid, joblen);
 702         } else if (strcmp(obd_jobid_var, JOBSTATS_SESSION) == 0) {
 703                 char *jid;
 704
 705                 rcu_read_lock();
 706                 jid = jobid_current();
 707                 if (jid)
 708                         strlcpy(jobid, jid, sizeof(jobid));
 709                 rcu_read_unlock();
 710         } else if (jobid_name_is_valid(current_comm())) {
 711                 /*
 712                  * obd_jobid_var holds the jobid environment variable name.
 713                  * Skip initial check if obd_jobid_name already uses "%j",
 714                  * otherwise try just "%j" first, then fall back to whatever
 715                  * is in obd_jobid_name if obd_jobid_var is not found.
 716                  */
 717                 rc = -EAGAIN;
 718                 if (!strnstr(obd_jobid_name, "%j", joblen))
 719                         rc = jobid_get_from_cache(jobid, joblen);
 720
 721                 /* fall back to jobid_node if jobid_var not in environment */
 722                 if (rc < 0) {
 723                         int rc2 = jobid_interpret_string(obd_jobid_name,
 724                                                          jobid, joblen);
 725                         if (!rc2)
 726                                 rc = 0;
 727                 }
 728         }
 729
 730         RETURN(rc);
 731 }
 732 EXPORT_SYMBOL(lustre_get_jobid);
 733
 734 /*
 735  * lustre_jobid_clear
 736  *
 737  * Search cache for JobID given by @find_jobid.
 738  * If any entries in the hash table match the value, they are removed
 739  */
 740 void lustre_jobid_clear(const char *find_jobid)
 741 {
 742         char jobid[LUSTRE_JOBID_SIZE];
 743         char *end;
 744
 745         if (jobid_hash == NULL)
 746                 return;
 747
 748         strlcpy(jobid, find_jobid, sizeof(jobid));
 749         /* trim \n off the end of the incoming jobid */
 750         end = strchr(jobid, '\n');
 751         if (end && *end == '\n')
 752                 *end = '\0';
 753
 754         CDEBUG(D_INFO, "Clearing Jobid: %s\n", jobid);
 755         cfs_hash_cond_del(jobid_hash, jobid_should_free_item, jobid);
 756
 757         CDEBUG(D_INFO, "%d items remain in jobID table\n",
 758                atomic_read(&jobid_hash->hs_count));
 759 }