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/buffer_head.h>
  46 # include <linux/workqueue.h>
  47 # include <linux/mount.h>
  48 #else /* __KERNEL__ */
  49 # include <liblustre.h>
  50 #endif
  51
  52 #include <obd_class.h>
  53 #include <lustre_mds.h>
  54 #include <lustre_dlm.h>
  55 #include <lustre_cfg.h>
  56 #include <obd_ost.h>
  57 #include <lustre_fsfilt.h>
  58 #include <lustre_quota.h>
  59 #include <lprocfs_status.h>
  60 #include "quota_internal.h"
  61
  62 #ifdef __KERNEL__
  63
  64 static cfs_time_t last_print = 0;
  65 static DEFINE_SPINLOCK(last_print_lock);
  66
  67 static int filter_quota_setup(struct obd_device *obd)
  68 {
  69         int rc = 0;
  70         struct obd_device_target *obt = &obd->u.obt;
  71         ENTRY;
  72
  73         cfs_init_rwsem(&obt->obt_rwsem);
  74         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  75         cfs_sema_init(&obt->obt_quotachecking, 1);
  76         rc = qctxt_init(obd, NULL);
  77         if (rc)
  78                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  79
  80         RETURN(rc);
  81 }
  82
  83 static int filter_quota_cleanup(struct obd_device *obd)
  84 {
  85         ENTRY;
  86         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  87         RETURN(0);
  88 }
  89
  90 static int filter_quota_setinfo(struct obd_device *obd, void *data)
  91 {
  92         struct obd_export *exp = data;
  93         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
  94         struct obd_import *imp = exp->exp_imp_reverse;
  95         ENTRY;
  96
  97         LASSERT(imp != NULL);
  98
  99         /* setup the quota context import */
 100         cfs_spin_lock(&qctxt->lqc_lock);
 101         if (qctxt->lqc_import != NULL) {
 102                 cfs_spin_unlock(&qctxt->lqc_lock);
 103                 if (qctxt->lqc_import == imp)
 104                         CDEBUG(D_WARNING, "%s: lqc_import(%p) of obd(%p) was "
 105                                "activated already.\n", obd->obd_name, imp, obd);
 106                 else
 107                         CERROR("%s: lqc_import(%p:%p) of obd(%p) was "
 108                                "activated by others.\n", obd->obd_name,
 109                                qctxt->lqc_import, imp, obd);
 110         } else {
 111                 qctxt->lqc_import = imp;
 112                 /* make imp's connect flags equal relative exp's connect flags
 113                  * adding it to avoid the scan export list */
 114                 imp->imp_connect_data.ocd_connect_flags |=
 115                                 (exp->exp_connect_flags &
 116                                  (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 117                 cfs_spin_unlock(&qctxt->lqc_lock);
 118                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated "
 119                        "now.\n", obd->obd_name, imp, obd);
 120
 121                 cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 122                 /* start quota slave recovery thread. (release high limits) */
 123                 qslave_start_recovery(obd, qctxt);
 124         }
 125         RETURN(0);
 126 }
 127
 128 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 129 {
 130         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 131         struct obd_import *imp = exp->exp_imp_reverse;
 132         ENTRY;
 133
 134         /* lquota may be not set up before destroying export, b=14896 */
 135         if (!obd->obd_set_up)
 136                 RETURN(0);
 137
 138         if (unlikely(imp == NULL))
 139                 RETURN(0);
 140
 141         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 142          * should be invalid b=12374 */
 143         cfs_spin_lock(&qctxt->lqc_lock);
 144         if (qctxt->lqc_import == imp) {
 145                 qctxt->lqc_import = NULL;
 146                 cfs_spin_unlock(&qctxt->lqc_lock);
 147                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is invalid now.\n",
 148                        obd->obd_name, imp, obd);
 149                 ptlrpc_cleanup_imp(imp);
 150                 dqacq_interrupt(qctxt);
 151         } else {
 152                 cfs_spin_unlock(&qctxt->lqc_lock);
 153         }
 154         RETURN(0);
 155 }
 156
 157 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 158 {
 159         ENTRY;
 160
 161         if (!ll_sb_any_quota_active(obd->u.obt.obt_sb))
 162                 RETURN(0);
 163
 164         if (ignore) {
 165                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 166                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 167         } else {
 168                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 169         }
 170
 171         RETURN(0);
 172 }
 173
 174 #define GET_OA_ID(flag, oa) (flag == USRQUOTA ? oa->o_uid : oa->o_gid)
 175 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 176 {
 177         struct obd_device_target *obt = &obd->u.obt;
 178         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 179         int err, cnt, rc = 0;
 180         struct obd_quotactl *oqctl;
 181         ENTRY;
 182
 183         if (!ll_sb_any_quota_active(obt->obt_sb))
 184                 RETURN(0);
 185
 186         OBD_ALLOC_PTR(oqctl);
 187         if (!oqctl)
 188                 RETURN(-ENOMEM);
 189
 190         /* set over quota flags for a uid/gid */
 191         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 192         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 193
 194         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 195                 struct lustre_qunit_size *lqs = NULL;
 196
 197                 /* check if quota is enabled */
 198                 if (!ll_sb_has_quota_active(obt->obt_sb, cnt))
 199                         continue;
 200
 201                 lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)),
 202                                        qctxt, 0);
 203                 if (IS_ERR(lqs)) {
 204                         rc = PTR_ERR(lqs);
 205                         CDEBUG(D_QUOTA, "search lqs for %s %d failed, "
 206                                "(rc = %d)\n",
 207                                cnt == USRQUOTA ? "user" : "group",
 208                                GET_OA_ID(cnt, oa), rc);
 209                         break;
 210                 } else if (lqs == NULL) {
 211                         /* continue to check group quota if the file's owner
 212                          * doesn't have quota limit. LU-530 */
 213                         continue;
 214                 } else {
 215                         cfs_spin_lock(&lqs->lqs_lock);
 216                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 217                                 oa->o_flags |= (cnt == USRQUOTA) ?
 218                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 219                                 cfs_spin_unlock(&lqs->lqs_lock);
 220                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 221                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 222                                        qctxt->lqc_sync_blk);
 223                                 /* this is for quota_search_lqs */
 224                                 lqs_putref(lqs);
 225                                 continue;
 226                         }
 227                         cfs_spin_unlock(&lqs->lqs_lock);
 228                         /* this is for quota_search_lqs */
 229                         lqs_putref(lqs);
 230                 }
 231
 232                 memset(oqctl, 0, sizeof(*oqctl));
 233
 234                 oqctl->qc_cmd = Q_GETQUOTA;
 235                 oqctl->qc_type = cnt;
 236                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 237                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 238                 if (err) {
 239                         if (!rc)
 240                                 rc = err;
 241                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 242                                                              OBD_MD_FLGRPQUOTA);
 243                         CDEBUG(D_QUOTA, "fsfilt getquota for %s %d failed, "
 244                                "(rc = %d)\n",
 245                                cnt == USRQUOTA ? "user" : "group",
 246                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid, err);
 247                         continue;
 248                 }
 249
 250                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 251                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 252                     oqctl->qc_dqblk.dqb_bhardlimit)) {
 253                         oa->o_flags |= (cnt == USRQUOTA) ?
 254                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 255                         CDEBUG(D_QUOTA, "out of quota for %s %d\n",
 256                                cnt == USRQUOTA ? "user" : "group",
 257                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid);
 258                 }
 259         }
 260         OBD_FREE_PTR(oqctl);
 261         RETURN(rc);
 262 }
 263
 264 /**
 265  * check whether the left quota of certain uid and gid can satisfy a block_write
 266  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 267  */
 268 static int quota_check_common(struct obd_device *obd, const unsigned int id[],
 269                               int pending[], int count, int cycle, int isblk,
 270                               struct inode *inode, int frags)
 271 {
 272         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 273         int i;
 274         struct qunit_data qdata[MAXQUOTAS];
 275         int mb = 0;
 276         int rc = 0, rc2[2] = { 0, 0 };
 277         ENTRY;
 278
 279         cfs_spin_lock(&qctxt->lqc_lock);
 280         if (!qctxt->lqc_valid){
 281                 cfs_spin_unlock(&qctxt->lqc_lock);
 282                 RETURN(rc);
 283         }
 284         cfs_spin_unlock(&qctxt->lqc_lock);
 285
 286         for (i = 0; i < MAXQUOTAS; i++) {
 287                 struct lustre_qunit_size *lqs = NULL;
 288
 289                 qdata[i].qd_id = id[i];
 290                 qdata[i].qd_flags = i;
 291                 if (isblk)
 292                         QDATA_SET_BLK(&qdata[i]);
 293                 qdata[i].qd_count = 0;
 294
 295                 /* check if quota is enabled */
 296                 if (!ll_sb_has_quota_active(qctxt->lqc_sb, i))
 297                         continue;
 298
 299                 /* ignore root user */
 300                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 301                         continue;
 302
 303                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 304                 if (lqs == NULL || IS_ERR(lqs))
 305                         continue;
 306
 307                 if (IS_ERR(lqs)) {
 308                         CERROR("can not find lqs for check_common: "
 309                                "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n",
 310                                id[i], i % 2 ? 'g': 'u', isblk, count,
 311                                PTR_ERR(lqs));
 312                         RETURN(PTR_ERR(lqs));
 313                 }
 314
 315                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 316                 cfs_spin_lock(&lqs->lqs_lock);
 317                 if (!cycle) {
 318                         if (isblk) {
 319                                 pending[i] = count * CFS_PAGE_SIZE;
 320                                 /* in order to complete this write, we need extra
 321                                  * meta blocks. This function can get it through
 322                                  * data needed to be written b=16542 */
 323                                 if (inode) {
 324                                         mb = pending[i];
 325                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 326                                                              &mb, inode,
 327                                                              frags);
 328                                         if (rc)
 329                                                 CERROR("%s: can't get extra "
 330                                                        "meta blocks\n",
 331                                                        obd->obd_name);
 332                                         else
 333                                                 pending[i] += mb;
 334                                 }
 335                                 LASSERTF(pending[i] >= 0, "pending is not valid"
 336                                          ", count=%d, mb=%d\n", count, mb);
 337                                 lqs->lqs_bwrite_pending += pending[i];
 338                         } else {
 339                                 pending[i] = count;
 340                                 lqs->lqs_iwrite_pending += pending[i];
 341                         }
 342                 }
 343
 344                 /* if xx_rec < 0, that means quota are releasing,
 345                  * and it may return before we use quota. So if
 346                  * we find this situation, we assuming it has
 347                  * returned b=18491 */
 348                 if (isblk && lqs->lqs_blk_rec < 0) {
 349                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 350                                 qdata[i].qd_count = 0;
 351                         else
 352                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 353                 }
 354                 if (!isblk && lqs->lqs_ino_rec < 0) {
 355                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 356                                 qdata[i].qd_count = 0;
 357                         else
 358                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 359                 }
 360
 361                 CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]"
 362                        " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]"
 363                        " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count,
 364                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 365                        qdata[i].qd_count, mb, pending[i]);
 366                 if (rc2[i] == QUOTA_RET_OK) {
 367                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 368                                 rc2[i] = QUOTA_RET_ACQUOTA;
 369                         if (!isblk && qdata[i].qd_count <
 370                             lqs->lqs_iwrite_pending)
 371                                 rc2[i] = QUOTA_RET_ACQUOTA;
 372                 }
 373
 374                 cfs_spin_unlock(&lqs->lqs_lock);
 375
 376                 if (lqs->lqs_blk_rec  < 0 &&
 377                     qdata[i].qd_count <
 378                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 379                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 380
 381                 /* When cycle is zero, lqs_*_pending will be changed. We will
 382                  * get reference of the lqs here and put reference of lqs in
 383                  * quota_pending_commit b=14784 */
 384                 if (!cycle)
 385                         lqs_getref(lqs);
 386
 387                 /* this is for quota_search_lqs */
 388                 lqs_putref(lqs);
 389         }
 390
 391         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 392                 RETURN(QUOTA_RET_ACQUOTA);
 393         else
 394                 RETURN(rc);
 395 }
 396
 397 int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag)
 398 {
 399         struct lustre_qunit_size *lqs;
 400         int i, q_set = 0;
 401
 402         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 403                 RETURN(0);
 404
 405         for (i = 0; i < MAXQUOTAS; i++) {
 406                 /* check if quota is enabled */
 407                 if (!ll_sb_has_quota_active(obd->u.obt.obt_qctxt.lqc_sb, i))
 408                         continue;
 409                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 410                                        &obd->u.obt.obt_qctxt, 0);
 411                 if (lqs && !IS_ERR(lqs)) {
 412                         if (lqs->lqs_flags & flag)
 413                                 q_set = 1;
 414                         lqs_putref(lqs);
 415                 }
 416         }
 417
 418         return q_set;
 419 }
 420
 421 static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
 422                                 const unsigned int id[], int pending[],
 423                                 int count, quota_acquire acquire,
 424                                 struct obd_trans_info *oti, int isblk,
 425                                 struct inode *inode, int frags)
 426 {
 427         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 428         struct timeval work_start;
 429         struct timeval work_end;
 430         long timediff;
 431         struct l_wait_info lwi = { 0 };
 432         int rc = 0, cycle = 0, count_err = 1;
 433         ENTRY;
 434
 435         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 436                 RETURN(0);
 437
 438         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 439                 /* If the client has been evicted or if it
 440                  * timed out and tried to reconnect already,
 441                  * abort the request immediately */
 442                 RETURN(-ENOTCONN);
 443
 444         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 445         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 446         /* Unfortunately, if quota master is too busy to handle the
 447          * pre-dqacq in time and quota hash on ost is used up, we
 448          * have to wait for the completion of in flight dqacq/dqrel,
 449          * in order to get enough quota for write b=12588 */
 450         cfs_gettimeofday(&work_start);
 451         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 452                                         inode, frags)) &
 453                QUOTA_RET_ACQUOTA) {
 454                 struct ptlrpc_thread *thr = oti != NULL ?
 455                                             oti->oti_thread : NULL;
 456
 457                 cfs_spin_lock(&qctxt->lqc_lock);
 458                 if (!qctxt->lqc_import && oti != NULL) {
 459                         cfs_spin_unlock(&qctxt->lqc_lock);
 460
 461                         LASSERT(thr != NULL);
 462                         /* The recovery thread doesn't have watchdog
 463                          * attached. LU-369 */
 464                         if (thr->t_watchdog != NULL)
 465                                 lc_watchdog_disable(thr->t_watchdog);
 466                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 467                         l_wait_event(qctxt->lqc_wait_for_qmaster,
 468                                      check_qm(qctxt), &lwi);
 469
 470                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 471                         if (thr->t_watchdog != NULL) {
 472                                 lc_watchdog_touch(thr->t_watchdog,
 473                                    ptlrpc_server_get_timeout(thr->t_svcpt));
 474                         }
 475                 } else {
 476                         cfs_spin_unlock(&qctxt->lqc_lock);
 477                 }
 478
 479                 cycle++;
 480                 if (isblk)
 481                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 482                 /* after acquire(), we should run quota_check_common again
 483                  * so that we confirm there are enough quota to finish write */
 484                 rc = acquire(obd, id, oti, isblk);
 485
 486                 /* please reference to dqacq_completion for the below */
 487                 /* a new request is finished, try again */
 488                 if (rc == QUOTA_REQ_RETURNED) {
 489                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 490                         continue;
 491                 }
 492
 493                 /* it is out of quota already */
 494                 if (rc == -EDQUOT) {
 495                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 496                         break;
 497                 }
 498
 499                 /* Related quota has been disabled by master, but enabled by
 500                  * slave, do not try again. */
 501                 if (unlikely(rc == -ESRCH)) {
 502                         CERROR("mismatched quota configuration, stop try.\n");
 503                         break;
 504                 }
 505
 506                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 507                         /* The client has been evicted or tried to
 508                          * to reconnect already, abort the request */
 509                         RETURN(-ENOTCONN);
 510
 511                 /* -EBUSY and others, wait a second and try again */
 512                 if (rc < 0) {
 513                         cfs_waitq_t        waitq;
 514                         struct l_wait_info lwi;
 515
 516                         if (thr != NULL && thr->t_watchdog != NULL)
 517                                 lc_watchdog_touch(thr->t_watchdog,
 518                                    ptlrpc_server_get_timeout(thr->t_svcpt));
 519                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 520                                count_err++);
 521
 522                         cfs_waitq_init(&waitq);
 523                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 524                                           NULL);
 525                         l_wait_event(waitq, 0, &lwi);
 526                 }
 527
 528                 if (rc < 0 || cycle % 10 == 0) {
 529                         cfs_spin_lock(&last_print_lock);
 530                         if (last_print == 0 ||
 531                             cfs_time_before((last_print + cfs_time_seconds(30)),
 532                                             cfs_time_current())) {
 533                                 last_print = cfs_time_current();
 534                                 cfs_spin_unlock(&last_print_lock);
 535                                 CWARN("still haven't managed to acquire quota "
 536                                       "space from the quota master after %d "
 537                                       "retries (err=%d, rc=%d)\n",
 538                                       cycle, count_err - 1, rc);
 539                         } else {
 540                                 cfs_spin_unlock(&last_print_lock);
 541                         }
 542                 }
 543
 544                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 545                        cycle);
 546         }
 547         cfs_gettimeofday(&work_end);
 548         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 549         lprocfs_counter_add(qctxt->lqc_stats,
 550                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 551                                     LQUOTA_WAIT_FOR_CHK_INO,
 552                             timediff);
 553
 554         if (rc > 0)
 555                 rc = 0;
 556         RETURN(rc);
 557 }
 558
 559 /**
 560  * when a block_write or inode_create rpc is finished, adjust the record for
 561  * pending blocks and inodes
 562  */
 563 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 564                                 int pending[], int isblk)
 565 {
 566         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 567         struct timeval work_start;
 568         struct timeval work_end;
 569         long timediff;
 570         int i;
 571         struct qunit_data qdata[MAXQUOTAS];
 572         ENTRY;
 573
 574         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 575         CLASSERT(MAXQUOTAS < 4);
 576         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 577                 RETURN(0);
 578
 579         cfs_gettimeofday(&work_start);
 580         for (i = 0; i < MAXQUOTAS; i++) {
 581                 struct lustre_qunit_size *lqs = NULL;
 582
 583                 LASSERT(pending[i] >= 0);
 584                 if (pending[i] == 0)
 585                         continue;
 586
 587                 qdata[i].qd_id = id[i];
 588                 qdata[i].qd_flags = i;
 589                 if (isblk)
 590                         QDATA_SET_BLK(&qdata[i]);
 591                 qdata[i].qd_count = 0;
 592
 593                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 594                         continue;
 595
 596                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 597                 if (lqs == NULL || IS_ERR(lqs)) {
 598                         CERROR("can not find lqs for pending_commit: "
 599                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 600                                "maybe cause unexpected lqs refcount error!\n",
 601                                id[i], i ? 'g': 'u', pending[i], isblk,
 602                                lqs ? PTR_ERR(lqs) : -1);
 603                         continue;
 604                 }
 605
 606                 cfs_spin_lock(&lqs->lqs_lock);
 607                 if (isblk) {
 608                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 609                                  "there are too many blocks! [id %u] [%c] "
 610                                  "[bwrite_pending %lu] [pending %u]\n",
 611                                  id[i], i % 2 ? 'g' : 'u',
 612                                  lqs->lqs_bwrite_pending, pending[i]);
 613
 614                         lqs->lqs_bwrite_pending -= pending[i];
 615                 } else {
 616                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 617                                 "there are too many files! [id %u] [%c] "
 618                                 "[iwrite_pending %lu] [pending %u]\n",
 619                                 id[i], i % 2 ? 'g' : 'u',
 620                                 lqs->lqs_iwrite_pending, pending[i]);
 621
 622                         lqs->lqs_iwrite_pending -= pending[i];
 623                 }
 624                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 625                        obd->obd_name,
 626                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 627                        i, pending[i], isblk);
 628                 cfs_spin_unlock(&lqs->lqs_lock);
 629
 630                 /* for quota_search_lqs in pending_commit */
 631                 lqs_putref(lqs);
 632                 /* for quota_search_lqs in quota_check */
 633                 lqs_putref(lqs);
 634         }
 635         cfs_gettimeofday(&work_end);
 636         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 637         lprocfs_counter_add(qctxt->lqc_stats,
 638                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 639                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 640                             timediff);
 641
 642         RETURN(0);
 643 }
 644
 645 static int mds_quota_init(void)
 646 {
 647         return lustre_dquot_init();
 648 }
 649
 650 static int mds_quota_exit(void)
 651 {
 652         lustre_dquot_exit();
 653         return 0;
 654 }
 655
 656 static int mds_quota_setup(struct obd_device *obd)
 657 {
 658         struct obd_device_target *obt = &obd->u.obt;
 659         struct mds_obd *mds = &obd->u.mds;
 660         int rc;
 661         ENTRY;
 662
 663         if (unlikely(mds->mds_quota)) {
 664                 CWARN("try to reinitialize quota context!\n");
 665                 RETURN(0);
 666         }
 667
 668         cfs_init_rwsem(&obt->obt_rwsem);
 669         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 670         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 671         cfs_sema_init(&obt->obt_quotachecking, 1);
 672         /* initialize quota master and quota context */
 673         cfs_init_rwsem(&mds->mds_qonoff_sem);
 674         rc = qctxt_init(obd, dqacq_handler);
 675         if (rc) {
 676                 CERROR("%s: initialize quota context failed! (rc:%d)\n",
 677                        obd->obd_name, rc);
 678                 RETURN(rc);
 679         }
 680         mds->mds_quota = 1;
 681         RETURN(rc);
 682 }
 683
 684 static int mds_quota_cleanup(struct obd_device *obd)
 685 {
 686         ENTRY;
 687         if (unlikely(!obd->u.mds.mds_quota))
 688                 RETURN(0);
 689
 690         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 691         RETURN(0);
 692 }
 693
 694 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 695 {
 696         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 697         ENTRY;
 698
 699         if (unlikely(!obd->u.mds.mds_quota))
 700                 RETURN(0);
 701
 702         if (data != NULL)
 703                 QUOTA_MASTER_READY(qctxt);
 704         else
 705                 QUOTA_MASTER_UNREADY(qctxt);
 706         RETURN(0);
 707 }
 708
 709 static int mds_quota_fs_cleanup(struct obd_device *obd)
 710 {
 711         struct mds_obd *mds = &obd->u.mds;
 712         struct obd_quotactl oqctl;
 713         ENTRY;
 714
 715         if (unlikely(!mds->mds_quota))
 716                 RETURN(0);
 717
 718         mds->mds_quota = 0;
 719         memset(&oqctl, 0, sizeof(oqctl));
 720         oqctl.qc_type = UGQUOTA;
 721
 722         cfs_down_write(&mds->mds_qonoff_sem);
 723         mds_admin_quota_off(obd, &oqctl);
 724         cfs_up_write(&mds->mds_qonoff_sem);
 725         RETURN(0);
 726 }
 727
 728 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 729                                 struct obd_trans_info *oti, int isblk)
 730 {
 731         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 732         int rc;
 733         ENTRY;
 734
 735         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 736         RETURN(rc);
 737 }
 738
 739 quota_interface_t mds_quota_interface = {
 740         .quota_init     = mds_quota_init,
 741         .quota_exit     = mds_quota_exit,
 742         .quota_setup    = mds_quota_setup,
 743         .quota_cleanup  = mds_quota_cleanup,
 744         .quota_check    = target_quota_check,
 745         .quota_ctl      = mds_quota_ctl,
 746         .quota_setinfo  = mds_quota_setinfo,
 747         .quota_fs_cleanup = mds_quota_fs_cleanup,
 748         .quota_recovery = mds_quota_recovery,
 749         .quota_adjust   = mds_quota_adjust,
 750         .quota_chkquota = quota_chk_acq_common,
 751         .quota_acquire  = quota_acquire_common,
 752         .quota_pending_commit = quota_pending_commit,
 753 };
 754
 755 quota_interface_t filter_quota_interface = {
 756         .quota_setup    = filter_quota_setup,
 757         .quota_cleanup  = filter_quota_cleanup,
 758         .quota_check    = target_quota_check,
 759         .quota_ctl      = filter_quota_ctl,
 760         .quota_setinfo  = filter_quota_setinfo,
 761         .quota_clearinfo = filter_quota_clearinfo,
 762         .quota_enforce  = filter_quota_enforce,
 763         .quota_getflag  = filter_quota_getflag,
 764         .quota_acquire  = quota_acquire_common,
 765         .quota_adjust   = filter_quota_adjust,
 766         .quota_chkquota = quota_chk_acq_common,
 767         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 768         .quota_pending_commit = quota_pending_commit,
 769 };
 770
 771 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 772
 773 static int __init init_lustre_quota(void)
 774 {
 775         int rc = 0;
 776
 777         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 778                                                 proc_lustre_root,
 779                                                 NULL, NULL);
 780         if (IS_ERR(lquota_type_proc_dir)) {
 781                 CERROR("LProcFS failed in lquota-init\n");
 782                 rc = PTR_ERR(lquota_type_proc_dir);
 783                 return rc;
 784         }
 785
 786         rc = qunit_cache_init();
 787         if (rc)
 788                 return rc;
 789
 790         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 791         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 792
 793         return 0;
 794 }
 795
 796 static void /*__exit*/ exit_lustre_quota(void)
 797 {
 798         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 799         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 800
 801         qunit_cache_cleanup();
 802
 803         if (lquota_type_proc_dir)
 804                 lprocfs_remove(&lquota_type_proc_dir);
 805 }
 806
 807 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 808 MODULE_DESCRIPTION("Lustre Quota");
 809 MODULE_LICENSE("GPL");
 810
 811 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
 812
 813 EXPORT_SYMBOL(mds_quota_interface);
 814 EXPORT_SYMBOL(filter_quota_interface);
 815 #endif /* __KERNEL */