lustre/quota/qsd_handler.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, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2012, 2017, Intel Corporation.
  25  * Use is subject to license terms.
  26  *
  27  * Author: Johann Lombardi <johann.lombardi@intel.com>
  28  * Author: Niu    Yawei    <yawei.niu@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LQUOTA
  32
  33 #include "qsd_internal.h"
  34
  35 /**
  36  * helper function bumping lqe_pending_req if there is no quota request in
  37  * flight for the lquota entry \a lqe. Otherwise, EBUSY is returned.
  38  */
  39 static inline int qsd_request_enter(struct lquota_entry *lqe)
  40 {
  41         /* is there already a quota request in flight? */
  42         if (lqe->lqe_pending_req != 0) {
  43                 LQUOTA_DEBUG(lqe, "already a request in flight");
  44                 return -EBUSY;
  45         }
  46
  47         if (lqe->lqe_pending_rel != 0) {
  48                 LQUOTA_ERROR(lqe, "no request in flight with pending_rel=%llu",
  49                              lqe->lqe_pending_rel);
  50                 LBUG();
  51         }
  52
  53         lqe->lqe_pending_req++;
  54         return 0;
  55 }
  56
  57 /**
  58  * Companion of qsd_request_enter() dropping lqe_pending_req to 0.
  59  */
  60 static inline void qsd_request_exit(struct lquota_entry *lqe)
  61 {
  62         if (lqe->lqe_pending_req != 1) {
  63                 LQUOTA_ERROR(lqe, "lqe_pending_req != 1!!!");
  64                 LBUG();
  65         }
  66         lqe->lqe_pending_req--;
  67         lqe->lqe_pending_rel = 0;
  68         wake_up(&lqe->lqe_waiters);
  69 }
  70
  71 /**
  72  * Check whether a qsd instance is all set to send quota request to master.
  73  * This includes checking whether:
  74  * - the connection to master is set up and usable,
  75  * - the qsd isn't stopping
  76  * - reintegration has been successfully completed and all indexes are
  77  *   up-to-date
  78  *
  79  * \param lqe - is the lquota entry for which we would like to send an quota
  80  *              request
  81  * \param lockh - is the remote handle of the global lock returned on success
  82  *
  83  * \retval 0 on success, appropriate error on failure
  84  */
  85 static int qsd_ready(struct lquota_entry *lqe, struct lustre_handle *lockh)
  86 {
  87         struct qsd_qtype_info   *qqi = lqe2qqi(lqe);
  88         struct qsd_instance     *qsd = qqi->qqi_qsd;
  89         struct obd_import       *imp = NULL;
  90         struct ldlm_lock        *lock;
  91         ENTRY;
  92
  93         read_lock(&qsd->qsd_lock);
  94         /* is the qsd about to shut down? */
  95         if (qsd->qsd_stopping) {
  96                 read_unlock(&qsd->qsd_lock);
  97                 LQUOTA_DEBUG(lqe, "dropping quota req since qsd is stopping");
  98                 /* Target is about to shut down, client will retry */
  99                 RETURN(-EINPROGRESS);
 100         }
 101
 102         /* is the connection to the quota master ready? */
 103         if (qsd->qsd_exp_valid)
 104                 imp = class_exp2cliimp(qsd->qsd_exp);
 105         if (imp == NULL || imp->imp_invalid) {
 106                 read_unlock(&qsd->qsd_lock);
 107                 LQUOTA_DEBUG(lqe, "connection to master not ready");
 108                 RETURN(-ENOTCONN);
 109         }
 110
 111         /* In most case, reintegration must have been triggered (when enable
 112          * quota or on OST start), however, in rare race condition (enabling
 113          * quota when starting OSTs), we might miss triggering reintegration
 114          * for some qqi.
 115          *
 116          * If the previous reintegration failed for some reason, we'll
 117          * re-trigger it here as well. */
 118         if (!qqi->qqi_glb_uptodate || !qqi->qqi_slv_uptodate) {
 119                 read_unlock(&qsd->qsd_lock);
 120                 LQUOTA_DEBUG(lqe, "not up-to-date, dropping request and "
 121                              "kicking off reintegration");
 122                 qsd_start_reint_thread(qqi);
 123                 RETURN(-EINPROGRESS);
 124         }
 125
 126         /* Fill the remote global lock handle, master will check this handle
 127          * to see if the slave is sending request with stale lock */
 128         lustre_handle_copy(lockh, &qqi->qqi_lockh);
 129         read_unlock(&qsd->qsd_lock);
 130
 131         if (!lustre_handle_is_used(lockh))
 132                 RETURN(-ENOLCK);
 133
 134         lock = ldlm_handle2lock(lockh);
 135         if (lock == NULL)
 136                 RETURN(-ENOLCK);
 137
 138         /* return remote lock handle to be packed in quota request */
 139         lustre_handle_copy(lockh, &lock->l_remote_handle);
 140         LDLM_LOCK_PUT(lock);
 141
 142         RETURN(0);
 143 }
 144
 145 /**
 146  * Check whether any quota space adjustment (pre-acquire/release/report) is
 147  * needed for a given quota ID. If a non-null \a qbody is passed, then the
 148  * \a qbody structure (qb_count/flags/usage) is filled with appropriate data
 149  * to be packed in the quota request.
 150  *
 151  * \param lqe   - is the lquota entry for which we would like to adjust quota
 152  *                space.
 153  * \param qbody - is the quota body to fill, if not NULL.
 154  *
 155  * \retval true  - space adjustment is required and \a qbody is filled, if not
 156  *                 NULL
 157  * \retval false - no space adjustment required
 158  */
 159 static bool qsd_calc_adjust(struct lquota_entry *lqe, struct quota_body *qbody)
 160 {
 161         __u64   usage, granted;
 162         ENTRY;
 163
 164         usage   = lqe->lqe_usage;
 165         usage  += lqe->lqe_pending_write + lqe->lqe_waiting_write;
 166         granted = lqe->lqe_granted;
 167
 168         if (qbody != NULL)
 169                 qbody->qb_flags = 0;
 170
 171         if (!lqe->lqe_enforced) {
 172                 /* quota not enforced any more for this ID */
 173                 if (granted != 0) {
 174                         /* release all quota space unconditionally */
 175                         LQUOTA_DEBUG(lqe, "not enforced, releasing all space");
 176                         if (qbody != NULL) {
 177                                 qbody->qb_count = granted;
 178                                 qbody->qb_flags = QUOTA_DQACQ_FL_REL;
 179                         }
 180                         RETURN(true);
 181                 }
 182                 RETURN(false);
 183         }
 184
 185         if (!lustre_handle_is_used(&lqe->lqe_lockh)) {
 186                 /* No valid per-ID lock
 187                  * When reporting quota (during reintegration or on setquota
 188                  * glimpse), we should release granted space if usage is 0.
 189                  * Otherwise, if the usage is less than granted, we need to
 190                  * acquire the per-ID lock to make sure the unused grant can be
 191                  * reclaimed by per-ID lock glimpse. */
 192                 if (usage == 0) {
 193                         /* no on-disk usage and no outstanding activity, release
 194                          * space */
 195                         if (granted != 0) {
 196                                 LQUOTA_DEBUG(lqe, "no usage, releasing all "
 197                                              "space");
 198                                 if (qbody != NULL) {
 199                                         qbody->qb_count = granted;
 200                                         qbody->qb_flags = QUOTA_DQACQ_FL_REL;
 201                                 }
 202                                 RETURN(true);
 203                         }
 204                         LQUOTA_DEBUG(lqe, "no usage + no granted, nothing to "
 205                                      "do");
 206                         RETURN(false);
 207                 }
 208
 209                 if (lqe->lqe_usage < lqe->lqe_granted) {
 210                         /* holding quota space w/o any lock, enqueue per-ID lock
 211                          * again */
 212                         LQUOTA_DEBUG(lqe, "(re)acquiring per-ID lock");
 213                         if (qbody != NULL) {
 214                                 qbody->qb_count = 0;
 215                                 qbody->qb_flags = QUOTA_DQACQ_FL_ACQ;
 216                         }
 217                         RETURN(true);
 218                 }
 219
 220                 if (lqe->lqe_usage > lqe->lqe_granted) {
 221                         /* quota overrun, report usage */
 222                         LQUOTA_DEBUG(lqe, "overrun, reporting usage");
 223                         if (qbody != NULL) {
 224                                 qbody->qb_usage = lqe->lqe_usage;
 225                                 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
 226                         }
 227                         RETURN(true);
 228                 }
 229                 LQUOTA_DEBUG(lqe, "granted matches usage, nothing to do");
 230                 RETURN(false);
 231         }
 232
 233         /* valid per-ID lock
 234          * Apply good old quota qunit adjustment logic which has been around
 235          * since lustre 1.4:
 236          * 1. release spare quota space? */
 237         if (granted > usage + lqe->lqe_qunit) {
 238                 /* pre-release quota space */
 239                 if (qbody == NULL)
 240                         RETURN(true);
 241                 qbody->qb_count = granted - usage;
 242                 /* if usage == 0, release all granted space */
 243                 if (usage) {
 244                         /* try to keep one qunit of quota space */
 245                         qbody->qb_count -= lqe->lqe_qunit;
 246                         /* but don't release less than qtune to avoid releasing
 247                          * space too often */
 248                         if (qbody->qb_count < lqe->lqe_qtune)
 249                                 qbody->qb_count = lqe->lqe_qtune;
 250                 }
 251                 qbody->qb_flags = QUOTA_DQACQ_FL_REL;
 252                 RETURN(true);
 253         }
 254
 255         /* 2. Any quota overrun? */
 256         if (lqe->lqe_usage > lqe->lqe_granted) {
 257                 /* we overconsumed quota space, we report usage in request so
 258                  * that master can adjust it unconditionally */
 259                 if (qbody == NULL)
 260                         RETURN(true);
 261                 qbody->qb_usage = lqe->lqe_usage;
 262                 granted         = lqe->lqe_usage;
 263                 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
 264         }
 265
 266         /* 3. Time to pre-acquire? */
 267         if (!lqe->lqe_edquot && !lqe->lqe_nopreacq && usage > 0 &&
 268             lqe->lqe_qunit != 0 && granted < usage + lqe->lqe_qtune) {
 269                 /* To pre-acquire quota space, we report how much spare quota
 270                  * space the slave currently owns, then the master will grant us
 271                  * back how much we can pretend given the current state of
 272                  * affairs */
 273                 if (qbody == NULL)
 274                         RETURN(true);
 275                 if (granted <= usage)
 276                         qbody->qb_count = 0;
 277                 else
 278                         qbody->qb_count = granted - usage;
 279                 qbody->qb_flags |= QUOTA_DQACQ_FL_PREACQ;
 280                 RETURN(true);
 281         }
 282
 283         if (qbody != NULL)
 284                 RETURN(qbody->qb_flags != 0);
 285         else
 286                 RETURN(false);
 287 }
 288
 289 /**
 290  * Helper function returning true when quota space need to be adjusted (some
 291  * unused space should be free or pre-acquire) and false otherwise.
 292  */
 293 static inline bool qsd_adjust_needed(struct lquota_entry *lqe)
 294 {
 295         return qsd_calc_adjust(lqe, NULL);
 296 }
 297
 298 /**
 299  * Callback function called when an acquire/release request sent to the master
 300  * is completed
 301  */
 302 static void qsd_req_completion(const struct lu_env *env,
 303                                struct qsd_qtype_info *qqi,
 304                                struct quota_body *reqbody,
 305                                struct quota_body *repbody,
 306                                struct lustre_handle *lockh,
 307                                struct lquota_lvb *lvb,
 308                                void *arg, int ret)
 309 {
 310         struct lquota_entry     *lqe = (struct lquota_entry *)arg;
 311         struct qsd_thread_info  *qti;
 312         int                      rc;
 313         bool                     adjust = false, cancel = false;
 314         ENTRY;
 315
 316         LASSERT(qqi != NULL && lqe != NULL);
 317
 318         /* environment passed by ptlrpcd is mostly used by CLIO and hasn't the
 319          * DT tags set. */
 320         rc = lu_env_refill_by_tags((struct lu_env *)env, LCT_DT_THREAD, 0);
 321         if (rc) {
 322                 LQUOTA_ERROR(lqe, "failed to refill environmnent %d", rc);
 323                 lqe_write_lock(lqe);
 324                 /* can't afford to adjust quota space with no suitable lu_env */
 325                 GOTO(out_noadjust, rc);
 326         }
 327         qti = qsd_info(env);
 328
 329         lqe_write_lock(lqe);
 330         LQUOTA_DEBUG(lqe, "DQACQ returned %d, flags:0x%x", ret,
 331                      reqbody->qb_flags);
 332
 333         /* despite -EDQUOT & -EINPROGRESS errors, the master might still
 334          * grant us back quota space to adjust quota overrun */
 335         if (ret != 0 && ret != -EDQUOT && ret != -EINPROGRESS) {
 336                 if (ret != -ETIMEDOUT && ret != -ENOTCONN &&
 337                    ret != -ESHUTDOWN && ret != -EAGAIN)
 338                         /* print errors only if return code is unexpected */
 339                         LQUOTA_ERROR(lqe, "DQACQ failed with %d, flags:0x%x",
 340                                      ret, reqbody->qb_flags);
 341                 GOTO(out, ret);
 342         }
 343
 344         /* Set the lqe_lockh */
 345         if (lustre_handle_is_used(lockh) &&
 346             !lustre_handle_equal(lockh, &lqe->lqe_lockh))
 347                 lustre_handle_copy(&lqe->lqe_lockh, lockh);
 348
 349         /* If the replied qb_count is zero, it means master didn't process
 350          * the DQACQ since the limit for this ID has been removed, so we
 351          * should not update quota entry & slave index copy neither. */
 352         if (repbody != NULL && repbody->qb_count != 0) {
 353                 LQUOTA_DEBUG(lqe, "DQACQ qb_count:%llu", repbody->qb_count);
 354
 355                 if (lqe->lqe_is_reset) {
 356                         lqe->lqe_granted = 0;
 357                 } else if (req_is_rel(reqbody->qb_flags)) {
 358                         if (lqe->lqe_granted < repbody->qb_count) {
 359                                 LQUOTA_ERROR(lqe, "can't release more space "
 360                                              "than owned %llu<%llu",
 361                                              lqe->lqe_granted,
 362                                              repbody->qb_count);
 363                                 lqe->lqe_granted = 0;
 364                         } else {
 365                                 lqe->lqe_granted -= repbody->qb_count;
 366                         }
 367                         /* Cancel the per-ID lock initiatively when there
 368                          * isn't any usage & grant, which can avoid master
 369                          * sending glimpse unnecessarily to this slave on
 370                          * quota revoking */
 371                         if (!lqe->lqe_pending_write && !lqe->lqe_granted &&
 372                             !lqe->lqe_waiting_write && !lqe->lqe_usage)
 373                                 cancel = true;
 374                 } else {
 375                         lqe->lqe_granted += repbody->qb_count;
 376                 }
 377                 qti->qti_rec.lqr_slv_rec.qsr_granted = lqe->lqe_granted;
 378                 lqe_write_unlock(lqe);
 379
 380                 /* Update the slave index file in the dedicated thread. So far,
 381                  * We don't update the version of slave index copy on DQACQ.
 382                  * No locking is necessary since nobody can change
 383                  * lqe->lqe_granted while lqe->lqe_pending_req > 0 */
 384                 if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_GRANT))
 385                         qti->qti_rec.lqr_slv_rec.qsr_granted =
 386                                                         0xFFFFFFFFFFDEC80CULL;
 387                 qsd_upd_schedule(qqi, lqe, &lqe->lqe_id, &qti->qti_rec, 0,
 388                                  false);
 389                 lqe_write_lock(lqe);
 390         }
 391
 392         /* extract information from lvb */
 393         if (ret == 0 && lvb != NULL) {
 394                 if (lvb->lvb_id_qunit != 0)
 395                         qsd_set_qunit(lqe, lvb->lvb_id_qunit);
 396                 qsd_set_edquot(lqe, !!(lvb->lvb_flags & LQUOTA_FL_EDQUOT));
 397         } else if (repbody != NULL && repbody->qb_qunit != 0) {
 398                 qsd_set_qunit(lqe, repbody->qb_qunit);
 399         }
 400
 401         /* turn off pre-acquire if it failed with -EDQUOT. This is done to avoid
 402          * flooding the master with acquire request. Pre-acquire will be turned
 403          * on again as soon as qunit is modified */
 404         if (req_is_preacq(reqbody->qb_flags) && ret == -EDQUOT)
 405                 lqe->lqe_nopreacq = true;
 406 out:
 407         adjust = qsd_adjust_needed(lqe);
 408         if (reqbody && req_is_acq(reqbody->qb_flags) && ret != -EDQUOT) {
 409                 lqe->lqe_acq_rc = ret;
 410                 lqe->lqe_acq_time = ktime_get_seconds();
 411         }
 412 out_noadjust:
 413         qsd_request_exit(lqe);
 414         lqe_write_unlock(lqe);
 415
 416         /* release reference on per-ID lock */
 417         if (lustre_handle_is_used(lockh))
 418                 ldlm_lock_decref(lockh, qsd_id_einfo.ei_mode);
 419
 420         if (cancel) {
 421                 qsd_adjust_schedule(lqe, false, true);
 422         } else if (adjust) {
 423                 if (!ret || ret == -EDQUOT)
 424                         qsd_adjust_schedule(lqe, false, false);
 425                 else
 426                         qsd_adjust_schedule(lqe, true, false);
 427         }
 428         lqe_putref(lqe);
 429
 430         if (lvb)
 431                 OBD_FREE_PTR(lvb);
 432         EXIT;
 433 }
 434
 435 /**
 436  * Try to consume local quota space.
 437  *
 438  * \param lqe   - is the qid entry to be processed
 439  * \param space - is the amount of quota space needed to complete the operation
 440  *
 441  * \retval 0       - success
 442  * \retval -EDQUOT - out of quota
 443  * \retval -EAGAIN - need to acquire space from master
 444  */
 445 static int qsd_acquire_local(struct lquota_entry *lqe, __u64 space)
 446 {
 447         __u64   usage;
 448         int     rc;
 449         ENTRY;
 450
 451         if (!lqe->lqe_enforced)
 452                 /* not enforced any more, we are good */
 453                 RETURN(-ESRCH);
 454
 455         lqe_write_lock(lqe);
 456         /* use latest usage */
 457         usage = lqe->lqe_usage;
 458         /* take pending write into account */
 459         usage += lqe->lqe_pending_write;
 460
 461         if (space + usage <= lqe->lqe_granted - lqe->lqe_pending_rel) {
 462                 /* Yay! we got enough space */
 463                 lqe->lqe_pending_write += space;
 464                 lqe->lqe_waiting_write -= space;
 465                 rc = 0;
 466         /* lqe_edquot flag is used to avoid flooding dqacq requests when
 467          * the user is over quota, however, the lqe_edquot could be stale
 468          * sometimes due to the race reply of dqacq vs. id lock glimpse
 469          * (see LU-4505), so we revalidate it every 5 seconds. */
 470         } else if (lqe->lqe_edquot &&
 471                    (lqe->lqe_edquot_time > ktime_get_seconds() - 5)) {
 472                 rc = -EDQUOT;
 473         }else {
 474                 rc = -EAGAIN;
 475         }
 476         lqe_write_unlock(lqe);
 477
 478         RETURN(rc);
 479 }
 480
 481 /**
 482  * Compute how much quota space should be acquire from the master based
 483  * on how much is currently granted to this slave and pending/waiting
 484  * operations.
 485  *
 486  * \param lqe - is the lquota entry for which we would like to adjust quota
 487  *              space.
 488  * \param qbody - is the quota body of the acquire request to fill
 489  *
 490  * \retval true  - space acquisition is needed and qbody is filled
 491  * \retval false - no space acquisition required
 492  */
 493 static inline bool qsd_calc_acquire(struct lquota_entry *lqe,
 494                                     struct quota_body *qbody)
 495 {
 496         __u64   usage, granted;
 497
 498         usage   = lqe->lqe_usage;
 499         usage  += lqe->lqe_pending_write + lqe->lqe_waiting_write;
 500         granted = lqe->lqe_granted;
 501
 502         qbody->qb_flags = 0;
 503
 504         /* if we overconsumed quota space, we report usage in request so that
 505          * master can adjust it unconditionally */
 506         if (lqe->lqe_usage > lqe->lqe_granted) {
 507                 qbody->qb_usage = lqe->lqe_usage;
 508                 qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
 509                 granted = lqe->lqe_usage;
 510         }
 511
 512         /* acquire as much as needed, but not more */
 513         if (usage > granted) {
 514                 qbody->qb_count  = usage - granted;
 515                 qbody->qb_flags |= QUOTA_DQACQ_FL_ACQ;
 516         }
 517
 518         return qbody->qb_flags != 0;
 519 }
 520
 521 /**
 522  * Acquire quota space from master.
 523  * There are at most 1 in-flight dqacq/dqrel.
 524  *
 525  * \param env    - the environment passed by the caller
 526  * \param lqe    - is the qid entry to be processed
 527  *
 528  * \retval 0            - success
 529  * \retval -EDQUOT      - out of quota
 530  * \retval -EINPROGRESS - inform client to retry write/create
 531  * \retval -EBUSY       - already a quota request in flight
 532  * \retval -ve          - other appropriate errors
 533  */
 534 static int qsd_acquire_remote(const struct lu_env *env,
 535                               struct lquota_entry *lqe)
 536 {
 537         struct qsd_thread_info  *qti = qsd_info(env);
 538         struct quota_body       *qbody = &qti->qti_body;
 539         struct qsd_instance     *qsd;
 540         struct qsd_qtype_info   *qqi;
 541         int                      rc;
 542         ENTRY;
 543
 544         memset(qbody, 0, sizeof(*qbody));
 545         rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
 546         if (rc)
 547                 RETURN(rc);
 548
 549         qqi = lqe2qqi(lqe);
 550         qsd = qqi->qqi_qsd;
 551
 552         lqe_write_lock(lqe);
 553
 554         /* is quota really enforced for this id? */
 555         if (!lqe->lqe_enforced) {
 556                 lqe_write_unlock(lqe);
 557                 LQUOTA_DEBUG(lqe, "quota not enforced any more");
 558                 RETURN(0);
 559         }
 560
 561         /* fill qb_count & qb_flags */
 562         if (!qsd_calc_acquire(lqe, qbody)) {
 563                 lqe_write_unlock(lqe);
 564                 LQUOTA_DEBUG(lqe, "No acquire required");
 565                 RETURN(0);
 566         }
 567
 568         /* check whether an acquire request completed recently */
 569         if (lqe->lqe_acq_rc != 0 &&
 570             lqe->lqe_acq_time > ktime_get_seconds() - 1) {
 571                 lqe_write_unlock(lqe);
 572                 LQUOTA_DEBUG(lqe, "using cached return code %d", lqe->lqe_acq_rc);
 573                 RETURN(lqe->lqe_acq_rc);
 574         }
 575
 576         /* only 1 quota request in flight for a given ID is allowed */
 577         rc = qsd_request_enter(lqe);
 578         if (rc) {
 579                 lqe_write_unlock(lqe);
 580                 RETURN(rc);
 581         }
 582
 583         lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
 584         lqe_write_unlock(lqe);
 585
 586         /* hold a refcount until completion */
 587         lqe_getref(lqe);
 588
 589         /* fill other quota body fields */
 590         qbody->qb_fid = qqi->qqi_fid;
 591         qbody->qb_id  = lqe->lqe_id;
 592
 593         /* check whether we already own a valid lock for this ID */
 594         rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
 595         if (rc) {
 596                 struct lquota_lvb *lvb;
 597
 598                 OBD_ALLOC_PTR(lvb);
 599                 if (lvb == NULL) {
 600                         rc = -ENOMEM;
 601                         qsd_req_completion(env, qqi, qbody, NULL,
 602                                            &qti->qti_lockh, NULL, lqe, rc);
 603                         RETURN(rc);
 604                 }
 605                 /* no lock found, should use intent */
 606                 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, true,
 607                                      IT_QUOTA_DQACQ, qsd_req_completion,
 608                                      qqi, lvb, (void *)lqe);
 609         } else {
 610                 /* lock found, should use regular dqacq */
 611                 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, true,
 612                                     qsd_req_completion, qqi, &qti->qti_lockh,
 613                                     lqe);
 614         }
 615
 616         /* the completion function will be called by qsd_send_dqacq or
 617          * qsd_intent_lock */
 618         RETURN(rc);
 619 }
 620
 621 /**
 622  * Acquire \a space of quota space in order to complete an operation.
 623  * Try to consume local quota space first and send acquire request to quota
 624  * master if required.
 625  *
 626  * \param env   - the environment passed by the caller
 627  * \param lqe   - is the qid entry to be processed
 628  * \param space - is the amount of quota required for the operation
 629  * \param ret   - is the return code (-EDQUOT, -EINPROGRESS, ...)
 630  *
 631  * \retval true  - stop waiting in wait_event_idle_timeout,
 632  *                 and real return value in \a ret
 633  * \retval false - continue waiting
 634  */
 635 static bool qsd_acquire(const struct lu_env *env, struct lquota_entry *lqe,
 636                         long long space, int *ret)
 637 {
 638         int rc = 0, count;
 639         int wait_pending = 0;
 640         struct qsd_qtype_info *qqi = lqe2qqi(lqe);
 641
 642         ENTRY;
 643
 644         for (count = 0; rc == 0; count++) {
 645                 LQUOTA_DEBUG(lqe, "acquiring:%lld count=%d", space, count);
 646 again:
 647                 if (lqe2qqi(lqe)->qqi_qsd->qsd_stopping) {
 648                         rc = -EINPROGRESS;
 649                         break;
 650                 }
 651
 652                 /* refresh disk usage */
 653                 rc = qsd_refresh_usage(env, lqe);
 654                 if (rc)
 655                         break;
 656
 657                 /* try to consume local quota space first */
 658                 rc = qsd_acquire_local(lqe, space);
 659                 if (rc != -EAGAIN)
 660                         /* rc == 0, Wouhou! enough local quota space
 661                          * rc < 0, something bad happened */
 662                          break;
 663                 /*
 664                  * There might be a window that commit transaction
 665                  * have updated usage but pending write doesn't change
 666                  * wait for it before acquiring remotely.
 667                  */
 668                 if (lqe->lqe_pending_write >= space && !wait_pending) {
 669                         wait_pending = 1;
 670                         dt_sync(env, qqi->qqi_qsd->qsd_dev);
 671                         goto again;
 672                 }
 673
 674                 /* if we have gotten some quota and stil wait more quota,
 675                  * it's better to give QMT some time to reclaim from clients */
 676                 if (count > 0)
 677                         schedule_timeout_interruptible(cfs_time_seconds(1));
 678
 679                 /* need to acquire more quota space from master */
 680                 rc = qsd_acquire_remote(env, lqe);
 681         }
 682
 683         if (rc == -EBUSY)
 684                 /* already a request in flight, continue waiting */
 685                 RETURN(false);
 686         *ret = rc;
 687         RETURN(true);
 688 }
 689
 690 /**
 691  * Quota enforcement handler. If local quota can satisfy this operation,
 692  * return success, otherwise, acquire more quota from master.
 693  * (for write operation, if master isn't available at this moment, return
 694  * -EINPROGRESS to inform client to retry the write)
 695  *
 696  * \param env   - the environment passed by the caller
 697  * \param qsd   - is the qsd instance associated with the device in charge
 698  *                of the operation.
 699  * \param qid   - is the qid information attached in the transaction handle
 700  * \param space - is the space required by the operation
 701  * \param flags - if the operation is write, return caller no user/group
 702  *                and sync commit flags
 703  *
 704  * \retval 0            - success
 705  * \retval -EDQUOT      - out of quota
 706  * \retval -EINPROGRESS - inform client to retry write
 707  * \retval -ve          - other appropriate errors
 708  */
 709 static int qsd_op_begin0(const struct lu_env *env, struct qsd_qtype_info *qqi,
 710                          struct lquota_id_info *qid, long long space,
 711                          enum osd_quota_local_flags *local_flags)
 712 {
 713         struct lquota_entry *lqe;
 714         enum osd_quota_local_flags qtype_flag = 0;
 715         int rc, ret = -EINPROGRESS;
 716         ENTRY;
 717
 718         if (qid->lqi_qentry != NULL) {
 719                 /* we already had to deal with this id for this transaction */
 720                 lqe = qid->lqi_qentry;
 721                 if (!lqe->lqe_enforced)
 722                         RETURN(0);
 723         } else {
 724                 /* look up lquota entry associated with qid */
 725                 lqe = lqe_locate(env, qqi->qqi_site, &qid->lqi_id);
 726                 if (IS_ERR(lqe))
 727                         RETURN(PTR_ERR(lqe));
 728                 if (!lqe->lqe_enforced) {
 729                         lqe_putref(lqe);
 730                         RETURN(0);
 731                 }
 732                 qid->lqi_qentry = lqe;
 733                 /* lqe will be released in qsd_op_end() */
 734         }
 735
 736         if (space <= 0) {
 737                 /* when space is negative or null, we don't need to consume
 738                  * quota space. That said, we still want to perform space
 739                  * adjustments in qsd_op_end, so we return here, but with
 740                  * a reference on the lqe */
 741                 if (local_flags != NULL) {
 742                         rc = qsd_refresh_usage(env, lqe);
 743                         GOTO(out_flags, rc);
 744                 }
 745                 RETURN(0);
 746         }
 747
 748         LQUOTA_DEBUG(lqe, "op_begin space:%lld", space);
 749
 750         lqe_write_lock(lqe);
 751         lqe->lqe_waiting_write += space;
 752         lqe_write_unlock(lqe);
 753
 754         /* acquire quota space for the operation, cap overall wait time to
 755          * prevent a service thread from being stuck for too long */
 756         rc = wait_event_idle_timeout(
 757                 lqe->lqe_waiters, qsd_acquire(env, lqe, space, &ret),
 758                 cfs_time_seconds(qsd_wait_timeout(qqi->qqi_qsd)));
 759
 760         if (rc > 0 && ret == 0) {
 761                 qid->lqi_space += space;
 762                 rc = 0;
 763         } else {
 764                 if (rc > 0)
 765                         rc = ret;
 766                 else if (rc == 0)
 767                         rc = -ETIMEDOUT;
 768
 769                 LQUOTA_DEBUG(lqe, "acquire quota failed:%d", rc);
 770
 771                 lqe_write_lock(lqe);
 772                 lqe->lqe_waiting_write -= space;
 773
 774                 if (local_flags && lqe->lqe_pending_write != 0)
 775                         /* Inform OSD layer that there are pending writes.
 776                          * It might want to retry after a sync if appropriate */
 777                          *local_flags |= QUOTA_FL_SYNC;
 778                 lqe_write_unlock(lqe);
 779
 780                 /* convert recoverable error into -EINPROGRESS, client will
 781                  * retry */
 782                 if (rc == -ETIMEDOUT || rc == -ENOTCONN || rc == -ENOLCK ||
 783                     rc == -EAGAIN || rc == -EINTR) {
 784                         rc = -EINPROGRESS;
 785                 } else if (rc == -ESRCH) {
 786                         rc = 0;
 787                         LQUOTA_ERROR(lqe, "ID isn't enforced on master, it "
 788                                      "probably due to a legeal race, if this "
 789                                      "message is showing up constantly, there "
 790                                      "could be some inconsistence between "
 791                                      "master & slave, and quota reintegration "
 792                                      "needs be re-triggered.");
 793                 }
 794         }
 795
 796         if (local_flags != NULL) {
 797 out_flags:
 798                 LASSERT(qid->lqi_is_blk);
 799                 if (rc != 0) {
 800                         *local_flags |= lquota_over_fl(qqi->qqi_qtype);
 801                 } else {
 802                         __u64   usage;
 803
 804                         lqe_read_lock(lqe);
 805                         usage = lqe->lqe_pending_write;
 806                         usage += lqe->lqe_waiting_write;
 807                         /* There is a chance to successfully grant more quota
 808                          * but get edquot flag through glimpse. */
 809                         if (lqe->lqe_edquot || (lqe->lqe_qunit != 0 &&
 810                            (usage % lqe->lqe_qunit >
 811                             qqi->qqi_qsd->qsd_sync_threshold)))
 812                                 usage += qqi->qqi_qsd->qsd_sync_threshold;
 813
 814                         usage += lqe->lqe_usage;
 815
 816                         qtype_flag = lquota_over_fl(qqi->qqi_qtype);
 817                         /* if we should notify client to start sync write */
 818                         if (usage >= lqe->lqe_granted - lqe->lqe_pending_rel)
 819                                 *local_flags |= qtype_flag;
 820                         else
 821                                 *local_flags &= ~qtype_flag;
 822                         lqe_read_unlock(lqe);
 823                 }
 824         }
 825         RETURN(rc);
 826 }
 827
 828 /**
 829  * helper function comparing two lquota_id_info structures
 830  */
 831 static inline bool qid_equal(struct lquota_id_info *q1,
 832                              struct lquota_id_info *q2)
 833 {
 834         if (q1->lqi_is_blk != q2->lqi_is_blk || q1->lqi_type != q2->lqi_type)
 835                 return false;
 836         return (q1->lqi_id.qid_uid == q2->lqi_id.qid_uid) ? true : false;
 837 }
 838
 839 /**
 840  * Enforce quota, it's called in the declaration of each operation.
 841  * qsd_op_end() will then be called later once all the operations have been
 842  * completed in order to release/adjust the quota space.
 843  *
 844  * \param env   - the environment passed by the caller
 845  * \param qsd   - is the qsd instance associated with the device in charge of
 846  *                the operation.
 847  * \param trans - is the quota transaction information
 848  * \param qi    - qid & space required by current operation
 849  * \param flags - if the operation is write, return caller no user/group and
 850  *                sync commit flags
 851  *
 852  * \retval 0            - success
 853  * \retval -EDQUOT      - out of quota
 854  * \retval -EINPROGRESS - inform client to retry write
 855  * \retval -ve          - other appropriate errors
 856  */
 857 int qsd_op_begin(const struct lu_env *env, struct qsd_instance *qsd,
 858                  struct lquota_trans *trans, struct lquota_id_info *qi,
 859                  enum osd_quota_local_flags *local_flags)
 860 {
 861         int     i, rc;
 862         bool    found = false;
 863         ENTRY;
 864
 865         /* fast path, ignore quota enforcement request for root owned files */
 866         if (qi->lqi_id.qid_uid == 0)
 867                 return 0;
 868
 869         if (unlikely(qsd == NULL))
 870                 RETURN(0);
 871
 872         if (qsd->qsd_dev->dd_rdonly)
 873                 RETURN(0);
 874
 875         /* We don't enforce quota until the qsd_instance is started */
 876         read_lock(&qsd->qsd_lock);
 877         if (!qsd->qsd_started) {
 878                 read_unlock(&qsd->qsd_lock);
 879                 RETURN(0);
 880         }
 881         read_unlock(&qsd->qsd_lock);
 882
 883         /* ignore block quota on MDTs, ignore inode quota on OSTs */
 884         if ((!qsd->qsd_is_md && !qi->lqi_is_blk) ||
 885             (qsd->qsd_is_md && qi->lqi_is_blk))
 886                 RETURN(0);
 887
 888         /* ignore quota enforcement request when:
 889          *    - quota isn't enforced for this quota type
 890          * or - the user/group is root
 891          * or - quota accounting isn't enabled */
 892         if (!qsd_type_enabled(qsd, qi->lqi_type) ||
 893             (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed)
 894                 RETURN(0);
 895
 896         if (local_flags && qi->lqi_id.qid_projid && qsd->qsd_root_prj_enable)
 897                 *local_flags |= QUOTA_FL_ROOT_PRJQUOTA;
 898
 899         LASSERTF(trans->lqt_id_cnt <= QUOTA_MAX_TRANSIDS, "id_cnt=%d\n",
 900                  trans->lqt_id_cnt);
 901         /* check whether we already allocated a slot for this id */
 902         for (i = 0; i < trans->lqt_id_cnt; i++) {
 903                 if (qid_equal(qi, &trans->lqt_ids[i])) {
 904                         found = true;
 905                         break;
 906                 }
 907         }
 908
 909         if (!found) {
 910                 if (unlikely(i >= QUOTA_MAX_TRANSIDS)) {
 911                         CERROR("%s: more than %d qids enforced for a "
 912                                "transaction?\n", qsd->qsd_svname, i);
 913                         RETURN(-EINVAL);
 914                 }
 915
 916                 /* fill new slot */
 917                 trans->lqt_ids[i].lqi_id     = qi->lqi_id;
 918                 trans->lqt_ids[i].lqi_type   = qi->lqi_type;
 919                 trans->lqt_ids[i].lqi_is_blk = qi->lqi_is_blk;
 920                 trans->lqt_id_cnt++;
 921         }
 922
 923         /* manage quota enforcement for this ID */
 924         rc = qsd_op_begin0(env, qsd->qsd_type_array[qi->lqi_type],
 925                            &trans->lqt_ids[i], qi->lqi_space, local_flags);
 926         RETURN(rc);
 927 }
 928 EXPORT_SYMBOL(qsd_op_begin);
 929
 930 /**
 931  * Adjust quota space (by acquiring or releasing) hold by the quota slave.
 932  * This function is called after each quota request completion and during
 933  * reintegration in order to report usage or re-acquire quota locks.
 934  * Space adjustment is aborted if there is already a quota request in flight
 935  * for this ID.
 936  *
 937  * \param env    - the environment passed by the caller
 938  * \param lqe    - is the qid entry to be processed
 939  *
 940  * \retval 0 on success, appropriate errors on failure
 941  */
 942 int qsd_adjust(const struct lu_env *env, struct lquota_entry *lqe)
 943 {
 944         struct qsd_thread_info  *qti = qsd_info(env);
 945         struct quota_body       *qbody = &qti->qti_body;
 946         struct qsd_instance     *qsd;
 947         struct qsd_qtype_info   *qqi;
 948         int                      rc;
 949         bool                     intent = false;
 950         ENTRY;
 951
 952         memset(qbody, 0, sizeof(*qbody));
 953         rc = qsd_ready(lqe, &qbody->qb_glb_lockh);
 954         if (rc) {
 955                 /* add to adjust list again to trigger adjustment later when
 956                  * slave is ready */
 957                 LQUOTA_DEBUG(lqe, "delaying adjustment since qsd isn't ready");
 958                 qsd_adjust_schedule(lqe, true, false);
 959                 RETURN(0);
 960         }
 961
 962         qqi = lqe2qqi(lqe);
 963         qsd = qqi->qqi_qsd;
 964
 965         if (qsd->qsd_dev->dd_rdonly)
 966                 RETURN(0);
 967
 968         lqe_write_lock(lqe);
 969
 970         /* fill qb_count & qb_flags */
 971         if (!qsd_calc_adjust(lqe, qbody)) {
 972                 lqe_write_unlock(lqe);
 973                 LQUOTA_DEBUG(lqe, "no adjustment required");
 974                 RETURN(0);
 975         }
 976
 977         /* only 1 quota request in flight for a given ID is allowed */
 978         rc = qsd_request_enter(lqe);
 979         if (rc) {
 980                 /* already a request in flight, space adjustment will be run
 981                  * again on request completion */
 982                 lqe_write_unlock(lqe);
 983                 RETURN(0);
 984         }
 985
 986         if (req_is_rel(qbody->qb_flags))
 987                 lqe->lqe_pending_rel = qbody->qb_count;
 988         lustre_handle_copy(&qti->qti_lockh, &lqe->lqe_lockh);
 989         lqe_write_unlock(lqe);
 990
 991         /* hold a refcount until completion */
 992         lqe_getref(lqe);
 993
 994         /* fill other quota body fields */
 995         qbody->qb_fid = qqi->qqi_fid;
 996         qbody->qb_id  = lqe->lqe_id;
 997
 998         if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
 999                 /* check whether we own a valid lock for this ID */
1000                 rc = qsd_id_lock_match(&qti->qti_lockh, &qbody->qb_lockh);
1001                 if (rc) {
1002                         memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
1003                         if (req_is_preacq(qbody->qb_flags)) {
1004                                 if (req_has_rep(qbody->qb_flags))
1005                                         /* still want to report usage */
1006                                         qbody->qb_flags = QUOTA_DQACQ_FL_REPORT;
1007                                 else
1008                                         /* no pre-acquire if no per-ID lock */
1009                                         GOTO(out, rc = -ENOLCK);
1010                         } else {
1011                                 /* no lock found, should use intent */
1012                                 intent = true;
1013                         }
1014                 } else if (req_is_acq(qbody->qb_flags) &&
1015                            qbody->qb_count == 0) {
1016                         /* found cached lock, no need to acquire */
1017                         GOTO(out, rc = 0);
1018                 }
1019         } else {
1020                 /* release and report don't need a per-ID lock */
1021                 memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
1022         }
1023
1024         if (!intent) {
1025                 rc = qsd_send_dqacq(env, qsd->qsd_exp, qbody, false,
1026                                     qsd_req_completion, qqi, &qti->qti_lockh,
1027                                     lqe);
1028         } else {
1029                 struct lquota_lvb *lvb;
1030
1031                 OBD_ALLOC_PTR(lvb);
1032                 if (lvb == NULL)
1033                         GOTO(out, rc = -ENOMEM);
1034
1035                 rc = qsd_intent_lock(env, qsd->qsd_exp, qbody, false,
1036                                      IT_QUOTA_DQACQ, qsd_req_completion,
1037                                      qqi, lvb, (void *)lqe);
1038         }
1039         /* the completion function will be called by qsd_send_dqacq or
1040          * qsd_intent_lock */
1041         RETURN(rc);
1042 out:
1043         qsd_req_completion(env, qqi, qbody, NULL, &qti->qti_lockh, NULL, lqe,
1044                            rc);
1045         return rc;
1046 }
1047
1048 /**
1049  * Post quota operation, pre-acquire/release quota from master.
1050  *
1051  * \param  env  - the environment passed by the caller
1052  * \param  qsd  - is the qsd instance attached to the OSD device which
1053  *                is handling the operation.
1054  * \param  qqi  - is the qsd_qtype_info structure associated with the quota ID
1055  *                subject to the operation
1056  * \param  qid  - stores information related to his ID for the operation
1057  *                which has just completed
1058  *
1059  * \retval 0    - success
1060  * \retval -ve  - failure
1061  */
1062 static void qsd_op_end0(const struct lu_env *env, struct qsd_qtype_info *qqi,
1063                         struct lquota_id_info *qid)
1064 {
1065         struct lquota_entry     *lqe;
1066         bool                     adjust;
1067         ENTRY;
1068
1069         lqe = qid->lqi_qentry;
1070         if (lqe == NULL)
1071                 RETURN_EXIT;
1072         qid->lqi_qentry = NULL;
1073
1074         /* refresh cached usage if a suitable environment is passed */
1075         if (env != NULL)
1076                 qsd_refresh_usage(env, lqe);
1077
1078         lqe_write_lock(lqe);
1079         if (qid->lqi_space > 0) {
1080                 if (lqe->lqe_pending_write < qid->lqi_space) {
1081                         LQUOTA_ERROR(lqe,
1082                                      "More pending write quota to reduce (pending %llu, space %lld)\n",
1083                                      lqe->lqe_pending_write, qid->lqi_space);
1084                         lqe->lqe_pending_write = 0;
1085                 } else {
1086                         lqe->lqe_pending_write -= qid->lqi_space;
1087                 }
1088         }
1089         if (env != NULL)
1090                 adjust = qsd_adjust_needed(lqe);
1091         else
1092                 adjust = true;
1093         lqe_write_unlock(lqe);
1094
1095         if (adjust) {
1096                 /* pre-acquire/release quota space is needed */
1097                 if (env != NULL)
1098                         qsd_adjust(env, lqe);
1099                 else
1100                         /* no suitable environment, handle adjustment in
1101                          * separate thread context */
1102                         qsd_adjust_schedule(lqe, false, false);
1103         }
1104         lqe_putref(lqe);
1105         EXIT;
1106 }
1107
1108 /**
1109  * Post quota operation. It's called after each operation transaction stopped.
1110  *
1111  * \param  env   - the environment passed by the caller
1112  * \param  qsd   - is the qsd instance associated with device which is handling
1113  *                 the operation.
1114  * \param  qids  - all qids information attached in the transaction handle
1115  * \param  count - is the number of qid entries in the qids array.
1116  *
1117  * \retval 0     - success
1118  * \retval -ve   - failure
1119  */
1120 void qsd_op_end(const struct lu_env *env, struct qsd_instance *qsd,
1121                 struct lquota_trans *trans)
1122 {
1123         int i;
1124         ENTRY;
1125
1126         if (unlikely(qsd == NULL))
1127                 RETURN_EXIT;
1128
1129         if (qsd->qsd_dev->dd_rdonly)
1130                 RETURN_EXIT;
1131
1132         /* We don't enforce quota until the qsd_instance is started */
1133         read_lock(&qsd->qsd_lock);
1134         if (!qsd->qsd_started) {
1135                 read_unlock(&qsd->qsd_lock);
1136                 RETURN_EXIT;
1137         }
1138         read_unlock(&qsd->qsd_lock);
1139
1140         LASSERT(trans != NULL);
1141
1142         for (i = 0; i < trans->lqt_id_cnt; i++) {
1143                 struct qsd_qtype_info *qqi;
1144
1145                 if (trans->lqt_ids[i].lqi_qentry == NULL)
1146                         continue;
1147
1148                 qqi = qsd->qsd_type_array[trans->lqt_ids[i].lqi_type];
1149                 qsd_op_end0(env, qqi, &trans->lqt_ids[i]);
1150         }
1151
1152         /* reset id_count to 0 so that a second accidental call to qsd_op_end()
1153          * does not result in failure */
1154         trans->lqt_id_cnt = 0;
1155         EXIT;
1156 }
1157 EXPORT_SYMBOL(qsd_op_end);
1158
1159 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1160  * setattr needs. As a reminder, only the root user can change ownership of
1161  * a file, that's why EDQUOT & EINPROGRESS errors are discarded
1162  */
1163 int qsd_transfer(const struct lu_env *env, struct qsd_instance *qsd,
1164                  struct lquota_trans *trans, unsigned int qtype,
1165                  u64 orig_id, u64 new_id, u64 bspace,
1166                  struct lquota_id_info *qi)
1167 {
1168         int rc;
1169
1170         if (unlikely(!qsd))
1171                 return 0;
1172
1173         LASSERT(qtype < LL_MAXQUOTAS);
1174         if (qtype == PRJQUOTA)
1175                 if (!projid_valid(make_kprojid(&init_user_ns, new_id)))
1176                         return -EINVAL;
1177
1178         qi->lqi_type = qtype;
1179
1180         /* inode accounting */
1181         qi->lqi_is_blk = false;
1182
1183         /* one more inode for the new owner ... */
1184         qi->lqi_id.qid_uid = new_id;
1185         qi->lqi_space = 1;
1186         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1187         if (rc == -EDQUOT || rc == -EINPROGRESS)
1188                 rc = 0;
1189         if (rc)
1190                 return rc;
1191
1192         /* and one less inode for the current id */
1193         qi->lqi_id.qid_uid = orig_id;
1194         qi->lqi_space = -1;
1195         /* can't get EDQUOT when reducing usage */
1196         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1197         if (rc == -EINPROGRESS)
1198                 rc = 0;
1199         if (rc)
1200                 return rc;
1201
1202         /* block accounting */
1203         qi->lqi_is_blk = true;
1204
1205         /* more blocks for the new owner ... */
1206         qi->lqi_id.qid_uid = new_id;
1207         qi->lqi_space = bspace;
1208         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1209         if (rc == -EDQUOT || rc == -EINPROGRESS)
1210                 rc = 0;
1211         if (rc)
1212                 return rc;
1213
1214         /* and finally less blocks for the current owner */
1215         qi->lqi_id.qid_uid = orig_id;
1216         qi->lqi_space = -bspace;
1217         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1218         /* can't get EDQUOT when reducing usage */
1219         if (rc == -EINPROGRESS)
1220                 rc = 0;
1221         return rc;
1222 }
1223 EXPORT_SYMBOL(qsd_transfer);
1224
1225 /**
1226  * Trigger pre-acquire/release if necessary.
1227  * It's only used by ldiskfs osd so far. When unlink a file in ldiskfs, the
1228  * quota accounting isn't updated when the transaction stopped. Instead, it'll
1229  * be updated on the final iput, so qsd_op_adjust() will be called then (in
1230  * osd_object_delete()) to trigger quota release if necessary.
1231  *
1232  * \param env - the environment passed by the caller
1233  * \param qsd - is the qsd instance associated with the device in charge
1234  *              of the operation.
1235  * \param qid - is the lquota ID of the user/group for which to trigger
1236  *              quota space adjustment
1237  * \param qtype - is the quota type (USRQUOTA or GRPQUOTA)
1238  */
1239 void qsd_op_adjust(const struct lu_env *env, struct qsd_instance *qsd,
1240                    union lquota_id *qid, int qtype)
1241 {
1242         struct lquota_entry    *lqe;
1243         struct qsd_qtype_info  *qqi;
1244         bool                    adjust;
1245         ENTRY;
1246
1247         if (unlikely(qsd == NULL))
1248                 RETURN_EXIT;
1249
1250         /* We don't enforce quota until the qsd_instance is started */
1251         read_lock(&qsd->qsd_lock);
1252         if (!qsd->qsd_started) {
1253                 read_unlock(&qsd->qsd_lock);
1254                 RETURN_EXIT;
1255         }
1256         read_unlock(&qsd->qsd_lock);
1257
1258         qqi = qsd->qsd_type_array[qtype];
1259         LASSERT(qqi);
1260
1261         if (!qsd_type_enabled(qsd, qtype) || qqi->qqi_acct_obj == NULL ||
1262             qid->qid_uid == 0 || qsd->qsd_dev->dd_rdonly)
1263                 RETURN_EXIT;
1264
1265         read_lock(&qsd->qsd_lock);
1266         if (!qsd->qsd_started) {
1267                 read_unlock(&qsd->qsd_lock);
1268                 RETURN_EXIT;
1269         }
1270         read_unlock(&qsd->qsd_lock);
1271
1272         lqe = lqe_locate(env, qqi->qqi_site, qid);
1273         if (IS_ERR(lqe)) {
1274                 CERROR("%s: fail to locate lqe for id:%llu, type:%d\n",
1275                        qsd->qsd_svname, qid->qid_uid, qtype);
1276                 RETURN_EXIT;
1277         }
1278
1279         qsd_refresh_usage(env, lqe);
1280
1281         lqe_read_lock(lqe);
1282         adjust = qsd_adjust_needed(lqe);
1283         lqe_read_unlock(lqe);
1284
1285         if (adjust)
1286                 qsd_adjust(env, lqe);
1287
1288         lqe_putref(lqe);
1289         EXIT;
1290 }
1291 EXPORT_SYMBOL(qsd_op_adjust);
1292
1293 /**
1294  * Reserve or free quota.
1295  *
1296  * Currently, It's used to reserve quota space before changing the file's group
1297  * for normal user and free the reserved quota after the group change.
1298  *
1299  * \param env     - the environment passed by the caller
1300  * \param qsd     - is the qsd instance associated with the device in charge of
1301  *                  the operation.
1302  * \param qi      - qid & space required by current operation
1303  *
1304  * \retval 0            - success
1305  * \retval -EDQUOT      - out of quota
1306  * \retval -EINPROGRESS - inform client to retry write
1307  * \retval -ve          - other appropriate errors
1308  */
1309 int qsd_reserve_or_free_quota(const struct lu_env *env,
1310                               struct qsd_instance *qsd,
1311                               struct lquota_id_info *qi)
1312 {
1313         struct qsd_qtype_info  *qqi;
1314         bool is_free = qi->lqi_space < 0;
1315         int rc = 0;
1316
1317         ENTRY;
1318
1319         if (unlikely(qsd == NULL))
1320                 RETURN(0);
1321
1322         if (qsd->qsd_dev->dd_rdonly)
1323                 RETURN(0);
1324
1325         if (is_free)
1326                 qi->lqi_space *= -1;
1327
1328         /* We don't enforce quota until the qsd_instance is started */
1329         read_lock(&qsd->qsd_lock);
1330         if (!qsd->qsd_started) {
1331                 read_unlock(&qsd->qsd_lock);
1332                 RETURN(0);
1333         }
1334         read_unlock(&qsd->qsd_lock);
1335
1336         qqi = qsd->qsd_type_array[qi->lqi_type];
1337         LASSERT(qqi);
1338
1339         CDEBUG(D_QUOTA, "type %s, acct %s, free %d, count %llu\n",
1340                qsd_type_enabled(qsd, qi->lqi_type) ? "enabled" : "disabled",
1341                (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed ? "failed" :
1342                "succeed", is_free, qi->lqi_space);
1343
1344         /* ignore quota enforcement request when:
1345          *    - quota isn't enforced for this quota type
1346          * or - the user/group is root
1347          * or - quota accounting isn't enabled
1348          */
1349         if (!is_free &&
1350             (!qsd_type_enabled(qsd, qi->lqi_type) || qi->lqi_id.qid_uid == 0 ||
1351               (qsd->qsd_type_array[qi->lqi_type])->qqi_acct_failed))
1352                 RETURN(0);
1353
1354         if (is_free) {
1355                 qsd_op_end0(env, qsd->qsd_type_array[qi->lqi_type], qi);
1356         } else {
1357                 long long qspace = qi->lqi_space;
1358
1359                 /* the acquired quota will add to lqi_space in qsd_op_begin0 */
1360                 qi->lqi_space = 0;
1361                 rc = qsd_op_begin0(env, qsd->qsd_type_array[qi->lqi_type], qi,
1362                                    qspace, NULL);
1363                 if (rc && qi->lqi_qentry) {
1364                         lqe_putref(qi->lqi_qentry);
1365                         qi->lqi_qentry = NULL;
1366                 }
1367         }
1368
1369         CDEBUG(D_QUOTA, "%s quota: type %i, uid %llu, gid %llu, space %llu\n",
1370                is_free ? "Free" : "Reserve", qi->lqi_type, qi->lqi_id.qid_uid,
1371                qi->lqi_id.qid_gid, qi->lqi_space);
1372
1373         RETURN(rc);
1374 }
1375 EXPORT_SYMBOL(qsd_reserve_or_free_quota);