lustre/quota/quota_interface.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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LQUOTA
  38
  39 #ifdef __KERNEL__
  40 # include <linux/version.h>
  41 # include <linux/module.h>
  42 # include <linux/init.h>
  43 # include <linux/fs.h>
  44 # include <linux/jbd.h>
  45 # include <linux/smp_lock.h>
  46 # include <linux/buffer_head.h>
  47 # include <linux/workqueue.h>
  48 # include <linux/mount.h>
  49 #else /* __KERNEL__ */
  50 # include <liblustre.h>
  51 #endif
  52
  53 #include <obd_class.h>
  54 #include <lustre_mds.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_cfg.h>
  57 #include <obd_ost.h>
  58 #include <lustre_fsfilt.h>
  59 #include <lustre_quota.h>
  60 #include <lprocfs_status.h>
  61 #include "quota_internal.h"
  62
  63 #ifdef __KERNEL__
  64
  65 static cfs_time_t last_print = 0;
  66 static cfs_spinlock_t last_print_lock = CFS_SPIN_LOCK_UNLOCKED;
  67
  68 static int filter_quota_setup(struct obd_device *obd)
  69 {
  70         int rc = 0;
  71         struct obd_device_target *obt = &obd->u.obt;
  72         ENTRY;
  73
  74         cfs_init_rwsem(&obt->obt_rwsem);
  75         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  76         cfs_sema_init(&obt->obt_quotachecking, 1);
  77         rc = qctxt_init(obd, NULL);
  78         if (rc)
  79                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  80
  81         RETURN(rc);
  82 }
  83
  84 static int filter_quota_cleanup(struct obd_device *obd)
  85 {
  86         ENTRY;
  87         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  88         RETURN(0);
  89 }
  90
  91 static int filter_quota_setinfo(struct obd_device *obd, void *data)
  92 {
  93         struct obd_export *exp = data;
  94         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
  95         struct obd_import *imp = exp->exp_imp_reverse;
  96         ENTRY;
  97
  98         LASSERT(imp != NULL);
  99
 100         /* setup the quota context import */
 101         cfs_spin_lock(&qctxt->lqc_lock);
 102         if (qctxt->lqc_import != NULL) {
 103                 cfs_spin_unlock(&qctxt->lqc_lock);
 104                 if (qctxt->lqc_import == imp)
 105                         CDEBUG(D_WARNING, "%s: lqc_import(%p) of obd(%p) was "
 106                                "activated already.\n", obd->obd_name, imp, obd);
 107                 else
 108                         CERROR("%s: lqc_import(%p:%p) of obd(%p) was "
 109                                "activated by others.\n", obd->obd_name,
 110                                qctxt->lqc_import, imp, obd);
 111         } else {
 112                 qctxt->lqc_import = imp;
 113                 /* make imp's connect flags equal relative exp's connect flags
 114                  * adding it to avoid the scan export list */
 115                 imp->imp_connect_data.ocd_connect_flags |=
 116                                 (exp->exp_connect_flags &
 117                                  (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 118                 cfs_spin_unlock(&qctxt->lqc_lock);
 119                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated "
 120                        "now.\n", obd->obd_name, imp, obd);
 121
 122                 cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 123                 /* start quota slave recovery thread. (release high limits) */
 124                 qslave_start_recovery(obd, qctxt);
 125         }
 126         RETURN(0);
 127 }
 128
 129 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 130 {
 131         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 132         struct obd_import *imp = exp->exp_imp_reverse;
 133         ENTRY;
 134
 135         /* lquota may be not set up before destroying export, b=14896 */
 136         if (!obd->obd_set_up)
 137                 RETURN(0);
 138
 139         if (unlikely(imp == NULL))
 140                 RETURN(0);
 141
 142         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 143          * should be invalid b=12374 */
 144         cfs_spin_lock(&qctxt->lqc_lock);
 145         if (qctxt->lqc_import == imp) {
 146                 qctxt->lqc_import = NULL;
 147                 cfs_spin_unlock(&qctxt->lqc_lock);
 148                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is invalid now.\n",
 149                        obd->obd_name, imp, obd);
 150                 ptlrpc_cleanup_imp(imp);
 151                 dqacq_interrupt(qctxt);
 152         } else {
 153                 cfs_spin_unlock(&qctxt->lqc_lock);
 154         }
 155         RETURN(0);
 156 }
 157
 158 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 159 {
 160         ENTRY;
 161
 162         if (!ll_sb_any_quota_active(obd->u.obt.obt_sb))
 163                 RETURN(0);
 164
 165         if (ignore) {
 166                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 167                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 168         } else {
 169                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 170         }
 171
 172         RETURN(0);
 173 }
 174
 175 #define GET_OA_ID(flag, oa) (flag == USRQUOTA ? oa->o_uid : oa->o_gid)
 176 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 177 {
 178         struct obd_device_target *obt = &obd->u.obt;
 179         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 180         int err, cnt, rc = 0;
 181         struct obd_quotactl *oqctl;
 182         ENTRY;
 183
 184         if (!ll_sb_any_quota_active(obt->obt_sb))
 185                 RETURN(0);
 186
 187         OBD_ALLOC_PTR(oqctl);
 188         if (!oqctl)
 189                 RETURN(-ENOMEM);
 190
 191         /* set over quota flags for a uid/gid */
 192         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 193         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 194
 195         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 196                 struct lustre_qunit_size *lqs = NULL;
 197
 198                 lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)),
 199                                        qctxt, 0);
 200                 if (IS_ERR(lqs)) {
 201                         rc = PTR_ERR(lqs);
 202                         CDEBUG(D_QUOTA, "search lqs for %s %d failed, "
 203                                "(rc = %d)\n",
 204                                cnt == USRQUOTA ? "user" : "group",
 205                                GET_OA_ID(cnt, oa), rc);
 206                         break;
 207                 } else if (lqs == NULL) {
 208                         /* continue to check group quota if the file's owner
 209                          * doesn't have quota limit. LU-530 */
 210                         continue;
 211                 } else {
 212                         cfs_spin_lock(&lqs->lqs_lock);
 213                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 214                                 oa->o_flags |= (cnt == USRQUOTA) ?
 215                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 216                                 cfs_spin_unlock(&lqs->lqs_lock);
 217                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 218                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 219                                        qctxt->lqc_sync_blk);
 220                                 /* this is for quota_search_lqs */
 221                                 lqs_putref(lqs);
 222                                 continue;
 223                         }
 224                         cfs_spin_unlock(&lqs->lqs_lock);
 225                         /* this is for quota_search_lqs */
 226                         lqs_putref(lqs);
 227                 }
 228
 229                 memset(oqctl, 0, sizeof(*oqctl));
 230
 231                 oqctl->qc_cmd = Q_GETQUOTA;
 232                 oqctl->qc_type = cnt;
 233                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 234                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 235                 if (err) {
 236                         if (!rc)
 237                                 rc = err;
 238                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 239                                                              OBD_MD_FLGRPQUOTA);
 240                         CDEBUG(D_QUOTA, "fsfilt getquota for %s %d failed, "
 241                                "(rc = %d)\n",
 242                                cnt == USRQUOTA ? "user" : "group",
 243                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid, err);
 244                         continue;
 245                 }
 246
 247                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 248                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 249                     oqctl->qc_dqblk.dqb_bhardlimit)) {
 250                         oa->o_flags |= (cnt == USRQUOTA) ?
 251                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 252                         CDEBUG(D_QUOTA, "out of quota for %s %d\n",
 253                                cnt == USRQUOTA ? "user" : "group",
 254                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid);
 255                 }
 256         }
 257         OBD_FREE_PTR(oqctl);
 258         RETURN(rc);
 259 }
 260
 261 /**
 262  * check whether the left quota of certain uid and gid can satisfy a block_write
 263  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 264  */
 265 static int quota_check_common(struct obd_device *obd, const unsigned int id[],
 266                               int pending[], int count, int cycle, int isblk,
 267                               struct inode *inode, int frags)
 268 {
 269         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 270         int i;
 271         struct qunit_data qdata[MAXQUOTAS];
 272         int mb = 0;
 273         int rc = 0, rc2[2] = { 0, 0 };
 274         ENTRY;
 275
 276         cfs_spin_lock(&qctxt->lqc_lock);
 277         if (!qctxt->lqc_valid){
 278                 cfs_spin_unlock(&qctxt->lqc_lock);
 279                 RETURN(rc);
 280         }
 281         cfs_spin_unlock(&qctxt->lqc_lock);
 282
 283         for (i = 0; i < MAXQUOTAS; i++) {
 284                 struct lustre_qunit_size *lqs = NULL;
 285
 286                 qdata[i].qd_id = id[i];
 287                 qdata[i].qd_flags = i;
 288                 if (isblk)
 289                         QDATA_SET_BLK(&qdata[i]);
 290                 qdata[i].qd_count = 0;
 291
 292                 /* ignore root user */
 293                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 294                         continue;
 295
 296                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 297                 if (lqs == NULL || IS_ERR(lqs))
 298                         continue;
 299
 300                 if (IS_ERR(lqs)) {
 301                         CERROR("can not find lqs for check_common: "
 302                                "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n",
 303                                id[i], i % 2 ? 'g': 'u', isblk, count,
 304                                PTR_ERR(lqs));
 305                         RETURN(PTR_ERR(lqs));
 306                 }
 307
 308                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 309                 cfs_spin_lock(&lqs->lqs_lock);
 310                 if (!cycle) {
 311                         if (isblk) {
 312                                 pending[i] = count * CFS_PAGE_SIZE;
 313                                 /* in order to complete this write, we need extra
 314                                  * meta blocks. This function can get it through
 315                                  * data needed to be written b=16542 */
 316                                 if (inode) {
 317                                         mb = pending[i];
 318                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 319                                                              &mb, inode,
 320                                                              frags);
 321                                         if (rc)
 322                                                 CERROR("%s: can't get extra "
 323                                                        "meta blocks\n",
 324                                                        obd->obd_name);
 325                                         else
 326                                                 pending[i] += mb;
 327                                 }
 328                                 LASSERTF(pending[i] >= 0, "pending is not valid"
 329                                          ", count=%d, mb=%d\n", count, mb);
 330                                 lqs->lqs_bwrite_pending += pending[i];
 331                         } else {
 332                                 pending[i] = count;
 333                                 lqs->lqs_iwrite_pending += pending[i];
 334                         }
 335                 }
 336
 337                 /* if xx_rec < 0, that means quota are releasing,
 338                  * and it may return before we use quota. So if
 339                  * we find this situation, we assuming it has
 340                  * returned b=18491 */
 341                 if (isblk && lqs->lqs_blk_rec < 0) {
 342                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 343                                 qdata[i].qd_count = 0;
 344                         else
 345                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 346                 }
 347                 if (!isblk && lqs->lqs_ino_rec < 0) {
 348                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 349                                 qdata[i].qd_count = 0;
 350                         else
 351                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 352                 }
 353
 354                 CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]"
 355                        " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]"
 356                        " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count,
 357                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 358                        qdata[i].qd_count, mb, pending[i]);
 359                 if (rc2[i] == QUOTA_RET_OK) {
 360                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 361                                 rc2[i] = QUOTA_RET_ACQUOTA;
 362                         if (!isblk && qdata[i].qd_count <
 363                             lqs->lqs_iwrite_pending)
 364                                 rc2[i] = QUOTA_RET_ACQUOTA;
 365                 }
 366
 367                 cfs_spin_unlock(&lqs->lqs_lock);
 368
 369                 if (lqs->lqs_blk_rec  < 0 &&
 370                     qdata[i].qd_count <
 371                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 372                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 373
 374                 /* When cycle is zero, lqs_*_pending will be changed. We will
 375                  * get reference of the lqs here and put reference of lqs in
 376                  * quota_pending_commit b=14784 */
 377                 if (!cycle)
 378                         lqs_getref(lqs);
 379
 380                 /* this is for quota_search_lqs */
 381                 lqs_putref(lqs);
 382         }
 383
 384         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 385                 RETURN(QUOTA_RET_ACQUOTA);
 386         else
 387                 RETURN(rc);
 388 }
 389
 390 int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag)
 391 {
 392         struct lustre_qunit_size *lqs;
 393         int i, q_set = 0;
 394
 395         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 396                 RETURN(0);
 397
 398         for (i = 0; i < MAXQUOTAS; i++) {
 399                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 400                                        &obd->u.obt.obt_qctxt, 0);
 401                 if (lqs && !IS_ERR(lqs)) {
 402                         if (lqs->lqs_flags & flag)
 403                                 q_set = 1;
 404                         lqs_putref(lqs);
 405                 }
 406         }
 407
 408         return q_set;
 409 }
 410
 411 static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
 412                                 const unsigned int id[], int pending[],
 413                                 int count, quota_acquire acquire,
 414                                 struct obd_trans_info *oti, int isblk,
 415                                 struct inode *inode, int frags)
 416 {
 417         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 418         struct timeval work_start;
 419         struct timeval work_end;
 420         long timediff;
 421         struct l_wait_info lwi = { 0 };
 422         int rc = 0, cycle = 0, count_err = 1;
 423         ENTRY;
 424
 425         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 426                 RETURN(0);
 427
 428         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 429                 /* If the client has been evicted or if it
 430                  * timed out and tried to reconnect already,
 431                  * abort the request immediately */
 432                 RETURN(-ENOTCONN);
 433
 434         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 435         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 436         /* Unfortunately, if quota master is too busy to handle the
 437          * pre-dqacq in time and quota hash on ost is used up, we
 438          * have to wait for the completion of in flight dqacq/dqrel,
 439          * in order to get enough quota for write b=12588 */
 440         cfs_gettimeofday(&work_start);
 441         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 442                                         inode, frags)) &
 443                QUOTA_RET_ACQUOTA) {
 444
 445                 cfs_spin_lock(&qctxt->lqc_lock);
 446                 if (!qctxt->lqc_import && oti) {
 447                         cfs_spin_unlock(&qctxt->lqc_lock);
 448                         LASSERT(oti->oti_thread);
 449                         /* The recovery thread doesn't have watchdog
 450                          * attached. LU-369 */
 451                         if (oti->oti_thread->t_watchdog)
 452                                 lc_watchdog_disable(oti->oti_thread->\
 453                                                 t_watchdog);
 454                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 455                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 456                                      &lwi);
 457                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 458                         if (oti->oti_thread->t_watchdog)
 459                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 460                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 461                 } else {
 462                         cfs_spin_unlock(&qctxt->lqc_lock);
 463                 }
 464
 465                 cycle++;
 466                 if (isblk)
 467                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 468                 /* after acquire(), we should run quota_check_common again
 469                  * so that we confirm there are enough quota to finish write */
 470                 rc = acquire(obd, id, oti, isblk);
 471
 472                 /* please reference to dqacq_completion for the below */
 473                 /* a new request is finished, try again */
 474                 if (rc == QUOTA_REQ_RETURNED) {
 475                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 476                         continue;
 477                 }
 478
 479                 /* it is out of quota already */
 480                 if (rc == -EDQUOT) {
 481                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 482                         break;
 483                 }
 484
 485                 /* Related quota has been disabled by master, but enabled by
 486                  * slave, do not try again. */
 487                 if (unlikely(rc == -ESRCH)) {
 488                         CERROR("mismatched quota configuration, stop try.\n");
 489                         break;
 490                 }
 491
 492                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 493                         /* The client has been evicted or tried to
 494                          * to reconnect already, abort the request */
 495                         RETURN(-ENOTCONN);
 496
 497                 /* -EBUSY and others, wait a second and try again */
 498                 if (rc < 0) {
 499                         cfs_waitq_t        waitq;
 500                         struct l_wait_info lwi;
 501
 502                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 503                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 504                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 505                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 506                                count_err++);
 507
 508                         cfs_waitq_init(&waitq);
 509                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 510                                           NULL);
 511                         l_wait_event(waitq, 0, &lwi);
 512                 }
 513
 514                 if (rc < 0 || cycle % 10 == 0) {
 515                         cfs_spin_lock(&last_print_lock);
 516                         if (last_print == 0 ||
 517                             cfs_time_before((last_print + cfs_time_seconds(30)),
 518                                             cfs_time_current())) {
 519                                 last_print = cfs_time_current();
 520                                 cfs_spin_unlock(&last_print_lock);
 521                                 CWARN("still haven't managed to acquire quota "
 522                                       "space from the quota master after %d "
 523                                       "retries (err=%d, rc=%d)\n",
 524                                       cycle, count_err - 1, rc);
 525                         } else {
 526                                 cfs_spin_unlock(&last_print_lock);
 527                         }
 528                 }
 529
 530                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 531                        cycle);
 532         }
 533         cfs_gettimeofday(&work_end);
 534         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 535         lprocfs_counter_add(qctxt->lqc_stats,
 536                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 537                                     LQUOTA_WAIT_FOR_CHK_INO,
 538                             timediff);
 539
 540         if (rc > 0)
 541                 rc = 0;
 542         RETURN(rc);
 543 }
 544
 545 /**
 546  * when a block_write or inode_create rpc is finished, adjust the record for
 547  * pending blocks and inodes
 548  */
 549 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 550                                 int pending[], int isblk)
 551 {
 552         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 553         struct timeval work_start;
 554         struct timeval work_end;
 555         long timediff;
 556         int i;
 557         struct qunit_data qdata[MAXQUOTAS];
 558         ENTRY;
 559
 560         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 561         CLASSERT(MAXQUOTAS < 4);
 562         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 563                 RETURN(0);
 564
 565         cfs_gettimeofday(&work_start);
 566         for (i = 0; i < MAXQUOTAS; i++) {
 567                 struct lustre_qunit_size *lqs = NULL;
 568
 569                 LASSERT(pending[i] >= 0);
 570                 if (pending[i] == 0)
 571                         continue;
 572
 573                 qdata[i].qd_id = id[i];
 574                 qdata[i].qd_flags = i;
 575                 if (isblk)
 576                         QDATA_SET_BLK(&qdata[i]);
 577                 qdata[i].qd_count = 0;
 578
 579                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 580                         continue;
 581
 582                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 583                 if (lqs == NULL || IS_ERR(lqs)) {
 584                         CERROR("can not find lqs for pending_commit: "
 585                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 586                                "maybe cause unexpected lqs refcount error!\n",
 587                                id[i], i ? 'g': 'u', pending[i], isblk,
 588                                lqs ? PTR_ERR(lqs) : -1);
 589                         continue;
 590                 }
 591
 592                 cfs_spin_lock(&lqs->lqs_lock);
 593                 if (isblk) {
 594                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 595                                  "there are too many blocks! [id %u] [%c] "
 596                                  "[bwrite_pending %lu] [pending %u]\n",
 597                                  id[i], i % 2 ? 'g' : 'u',
 598                                  lqs->lqs_bwrite_pending, pending[i]);
 599
 600                         lqs->lqs_bwrite_pending -= pending[i];
 601                 } else {
 602                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 603                                 "there are too many files! [id %u] [%c] "
 604                                 "[iwrite_pending %lu] [pending %u]\n",
 605                                 id[i], i % 2 ? 'g' : 'u',
 606                                 lqs->lqs_iwrite_pending, pending[i]);
 607
 608                         lqs->lqs_iwrite_pending -= pending[i];
 609                 }
 610                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 611                        obd->obd_name,
 612                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 613                        i, pending[i], isblk);
 614                 cfs_spin_unlock(&lqs->lqs_lock);
 615
 616                 /* for quota_search_lqs in pending_commit */
 617                 lqs_putref(lqs);
 618                 /* for quota_search_lqs in quota_check */
 619                 lqs_putref(lqs);
 620         }
 621         cfs_gettimeofday(&work_end);
 622         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 623         lprocfs_counter_add(qctxt->lqc_stats,
 624                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 625                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 626                             timediff);
 627
 628         RETURN(0);
 629 }
 630
 631 static int mds_quota_init(void)
 632 {
 633         return lustre_dquot_init();
 634 }
 635
 636 static int mds_quota_exit(void)
 637 {
 638         lustre_dquot_exit();
 639         return 0;
 640 }
 641
 642 static int mds_quota_setup(struct obd_device *obd)
 643 {
 644         struct obd_device_target *obt = &obd->u.obt;
 645         struct mds_obd *mds = &obd->u.mds;
 646         int rc;
 647         ENTRY;
 648
 649         if (unlikely(mds->mds_quota)) {
 650                 CWARN("try to reinitialize quota context!\n");
 651                 RETURN(0);
 652         }
 653
 654         cfs_init_rwsem(&obt->obt_rwsem);
 655         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 656         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 657         cfs_sema_init(&obt->obt_quotachecking, 1);
 658         /* initialize quota master and quota context */
 659         cfs_init_rwsem(&mds->mds_qonoff_sem);
 660         rc = qctxt_init(obd, dqacq_handler);
 661         if (rc) {
 662                 CERROR("%s: initialize quota context failed! (rc:%d)\n",
 663                        obd->obd_name, rc);
 664                 RETURN(rc);
 665         }
 666         mds->mds_quota = 1;
 667         RETURN(rc);
 668 }
 669
 670 static int mds_quota_cleanup(struct obd_device *obd)
 671 {
 672         ENTRY;
 673         if (unlikely(!obd->u.mds.mds_quota))
 674                 RETURN(0);
 675
 676         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 677         RETURN(0);
 678 }
 679
 680 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 681 {
 682         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 683         ENTRY;
 684
 685         if (unlikely(!obd->u.mds.mds_quota))
 686                 RETURN(0);
 687
 688         if (data != NULL)
 689                 QUOTA_MASTER_READY(qctxt);
 690         else
 691                 QUOTA_MASTER_UNREADY(qctxt);
 692         RETURN(0);
 693 }
 694
 695 static int mds_quota_fs_cleanup(struct obd_device *obd)
 696 {
 697         struct mds_obd *mds = &obd->u.mds;
 698         struct obd_quotactl oqctl;
 699         ENTRY;
 700
 701         if (unlikely(!mds->mds_quota))
 702                 RETURN(0);
 703
 704         mds->mds_quota = 0;
 705         memset(&oqctl, 0, sizeof(oqctl));
 706         oqctl.qc_type = UGQUOTA;
 707
 708         cfs_down_write(&mds->mds_qonoff_sem);
 709         mds_admin_quota_off(obd, &oqctl);
 710         cfs_up_write(&mds->mds_qonoff_sem);
 711         RETURN(0);
 712 }
 713
 714 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 715                                 struct obd_trans_info *oti, int isblk)
 716 {
 717         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 718         int rc;
 719         ENTRY;
 720
 721         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 722         RETURN(rc);
 723 }
 724
 725 quota_interface_t mds_quota_interface = {
 726         .quota_init     = mds_quota_init,
 727         .quota_exit     = mds_quota_exit,
 728         .quota_setup    = mds_quota_setup,
 729         .quota_cleanup  = mds_quota_cleanup,
 730         .quota_check    = target_quota_check,
 731         .quota_ctl      = mds_quota_ctl,
 732         .quota_setinfo  = mds_quota_setinfo,
 733         .quota_fs_cleanup = mds_quota_fs_cleanup,
 734         .quota_recovery = mds_quota_recovery,
 735         .quota_adjust   = mds_quota_adjust,
 736         .quota_chkquota = quota_chk_acq_common,
 737         .quota_acquire  = quota_acquire_common,
 738         .quota_pending_commit = quota_pending_commit,
 739 };
 740
 741 quota_interface_t filter_quota_interface = {
 742         .quota_setup    = filter_quota_setup,
 743         .quota_cleanup  = filter_quota_cleanup,
 744         .quota_check    = target_quota_check,
 745         .quota_ctl      = filter_quota_ctl,
 746         .quota_setinfo  = filter_quota_setinfo,
 747         .quota_clearinfo = filter_quota_clearinfo,
 748         .quota_enforce  = filter_quota_enforce,
 749         .quota_getflag  = filter_quota_getflag,
 750         .quota_acquire  = quota_acquire_common,
 751         .quota_adjust   = filter_quota_adjust,
 752         .quota_chkquota = quota_chk_acq_common,
 753         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 754         .quota_pending_commit = quota_pending_commit,
 755 };
 756
 757 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 758
 759 static int __init init_lustre_quota(void)
 760 {
 761         int rc = 0;
 762
 763         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 764                                                 proc_lustre_root,
 765                                                 NULL, NULL);
 766         if (IS_ERR(lquota_type_proc_dir)) {
 767                 CERROR("LProcFS failed in lquota-init\n");
 768                 rc = PTR_ERR(lquota_type_proc_dir);
 769                 return rc;
 770         }
 771
 772         rc = qunit_cache_init();
 773         if (rc)
 774                 return rc;
 775
 776         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 777         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 778
 779         return 0;
 780 }
 781
 782 static void /*__exit*/ exit_lustre_quota(void)
 783 {
 784         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 785         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 786
 787         qunit_cache_cleanup();
 788
 789         if (lquota_type_proc_dir)
 790                 lprocfs_remove(&lquota_type_proc_dir);
 791 }
 792
 793 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 794 MODULE_DESCRIPTION("Lustre Quota");
 795 MODULE_LICENSE("GPL");
 796
 797 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
 798
 799 EXPORT_SYMBOL(mds_quota_interface);
 800 EXPORT_SYMBOL(filter_quota_interface);
 801 #endif /* __KERNEL */