lustre/quota/quota_interface.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_LQUOTA
  41
  42 #ifdef __KERNEL__
  43 # include <linux/version.h>
  44 # include <linux/module.h>
  45 # include <linux/init.h>
  46 # include <linux/fs.h>
  47 # include <linux/jbd.h>
  48 # include <linux/smp_lock.h>
  49 # include <linux/buffer_head.h>
  50 # include <linux/workqueue.h>
  51 # include <linux/mount.h>
  52 #else /* __KERNEL__ */
  53 # include <liblustre.h>
  54 #endif
  55
  56 #include <obd_class.h>
  57 #include <lustre_mds.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_cfg.h>
  60 #include <obd_ost.h>
  61 #include <lustre_fsfilt.h>
  62 #include <lustre_quota.h>
  63 #include <lprocfs_status.h>
  64 #include "quota_internal.h"
  65
  66 #ifdef __KERNEL__
  67
  68 #ifdef HAVE_QUOTA_SUPPORT
  69
  70 static cfs_time_t last_print = 0;
  71 static spinlock_t last_print_lock = SPIN_LOCK_UNLOCKED;
  72
  73 static int filter_quota_setup(struct obd_device *obd)
  74 {
  75         int rc = 0;
  76         struct obd_device_target *obt = &obd->u.obt;
  77         ENTRY;
  78
  79         init_rwsem(&obt->obt_rwsem);
  80         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  81         atomic_set(&obt->obt_quotachecking, 1);
  82         rc = qctxt_init(obd, NULL);
  83         if (rc)
  84                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  85
  86         RETURN(rc);
  87 }
  88
  89 static int filter_quota_cleanup(struct obd_device *obd)
  90 {
  91         ENTRY;
  92         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  93         RETURN(0);
  94 }
  95
  96 static int filter_quota_setinfo(struct obd_device *obd, void *data)
  97 {
  98         struct obd_export *exp = data;
  99         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 100         struct obd_import *imp;
 101         ENTRY;
 102
 103         /* setup the quota context import */
 104         spin_lock(&qctxt->lqc_lock);
 105         qctxt->lqc_import = exp->exp_imp_reverse;
 106         spin_unlock(&qctxt->lqc_lock);
 107         CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated now, \n",
 108                obd->obd_name,exp->exp_imp_reverse, obd);
 109
 110         /* make imp's connect flags equal relative exp's connect flags
 111          * adding it to avoid the scan export list
 112          */
 113         imp = qctxt->lqc_import;
 114         if (likely(imp))
 115                 imp->imp_connect_data.ocd_connect_flags |=
 116                         (exp->exp_connect_flags &
 117                          (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 118
 119         cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 120         /* start quota slave recovery thread. (release high limits) */
 121         qslave_start_recovery(obd, qctxt);
 122         RETURN(0);
 123 }
 124
 125 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 126 {
 127         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 128         ENTRY;
 129
 130         /* lquota may be not set up before destroying export, b=14896 */
 131         if (!obd->obd_set_up)
 132                 RETURN(0);
 133
 134         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 135          * should be invalid b=12374 */
 136         if (qctxt->lqc_import && qctxt->lqc_import == exp->exp_imp_reverse) {
 137                 spin_lock(&qctxt->lqc_lock);
 138                 qctxt->lqc_import = NULL;
 139                 spin_unlock(&qctxt->lqc_lock);
 140                 ptlrpc_cleanup_imp(exp->exp_imp_reverse);
 141                 dqacq_interrupt(qctxt);
 142                 CDEBUG(D_QUOTA, "%s: lqc_import of obd(%p) is invalid now.\n",
 143                        obd->obd_name, obd);
 144         }
 145         RETURN(0);
 146 }
 147
 148 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 149 {
 150         ENTRY;
 151
 152         if (!sb_any_quota_enabled(obd->u.obt.obt_sb))
 153                 RETURN(0);
 154
 155         if (ignore) {
 156                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 157                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 158         } else {
 159                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 160         }
 161
 162         RETURN(0);
 163 }
 164
 165 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 166 {
 167         struct obd_device_target *obt = &obd->u.obt;
 168         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 169         int err, cnt, rc = 0;
 170         struct obd_quotactl *oqctl;
 171         ENTRY;
 172
 173         if (!sb_any_quota_enabled(obt->obt_sb))
 174                 RETURN(0);
 175
 176         OBD_ALLOC_PTR(oqctl);
 177         if (!oqctl) {
 178                 CERROR("Not enough memory!");
 179                 RETURN(-ENOMEM);
 180         }
 181
 182         /* set over quota flags for a uid/gid */
 183         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 184         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 185
 186         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 187                 struct quota_adjust_qunit oqaq_tmp;
 188                 struct lustre_qunit_size *lqs = NULL;
 189
 190                 oqaq_tmp.qaq_flags = cnt;
 191                 oqaq_tmp.qaq_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 192
 193                 quota_search_lqs(NULL, &oqaq_tmp, qctxt, &lqs);
 194                 if (lqs) {
 195                         spin_lock(&lqs->lqs_lock);
 196                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 197                                 oa->o_flags |= (cnt == USRQUOTA) ?
 198                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 199                                 spin_unlock(&lqs->lqs_lock);
 200                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 201                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 202                                        qctxt->lqc_sync_blk);
 203                                 /* this is for quota_search_lqs */
 204                                 lqs_putref(lqs);
 205                                 continue;
 206                         }
 207                         spin_unlock(&lqs->lqs_lock);
 208                         /* this is for quota_search_lqs */
 209                         lqs_putref(lqs);
 210                 }
 211
 212                 memset(oqctl, 0, sizeof(*oqctl));
 213
 214                 oqctl->qc_cmd = Q_GETQUOTA;
 215                 oqctl->qc_type = cnt;
 216                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 217                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 218                 if (err) {
 219                         if (!rc)
 220                                 rc = err;
 221                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 222                                                              OBD_MD_FLGRPQUOTA);
 223                         continue;
 224                 }
 225
 226                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 227                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 228                     oqctl->qc_dqblk.dqb_bhardlimit))
 229                         oa->o_flags |= (cnt == USRQUOTA) ?
 230                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 231         }
 232         OBD_FREE_PTR(oqctl);
 233         RETURN(rc);
 234 }
 235
 236 /**
 237  * check whether the left quota of certain uid and gid can satisfy a block_write
 238  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 239  */
 240 static int quota_check_common(struct obd_device *obd, unsigned int uid,
 241                               unsigned int gid, int count, int cycle, int isblk,
 242                               struct inode *inode, int frags, int *pending)
 243 {
 244         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 245         int i;
 246         __u32 id[MAXQUOTAS] = { uid, gid };
 247         struct qunit_data qdata[MAXQUOTAS];
 248         int mb = 0;
 249         int rc = 0, rc2[2] = { 0, 0 };
 250         ENTRY;
 251
 252         CLASSERT(MAXQUOTAS < 4);
 253         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 254                 RETURN(rc);
 255
 256         spin_lock(&qctxt->lqc_lock);
 257         if (!qctxt->lqc_valid){
 258                 spin_unlock(&qctxt->lqc_lock);
 259                 RETURN(rc);
 260         }
 261         spin_unlock(&qctxt->lqc_lock);
 262
 263         for (i = 0; i < MAXQUOTAS; i++) {
 264                 struct lustre_qunit_size *lqs = NULL;
 265
 266                 qdata[i].qd_id = id[i];
 267                 qdata[i].qd_flags = i;
 268                 if (isblk)
 269                         QDATA_SET_BLK(&qdata[i]);
 270                 qdata[i].qd_count = 0;
 271
 272                 /* ignore root user */
 273                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 274                         continue;
 275
 276                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 277                 if (!lqs)
 278                         continue;
 279
 280                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 281                 spin_lock(&lqs->lqs_lock);
 282                 if (!cycle) {
 283                         if (isblk) {
 284                                 *pending = count * CFS_PAGE_SIZE;
 285                                 /* in order to complete this write, we need extra
 286                                  * meta blocks. This function can get it through
 287                                  * data needed to be written b=16542 */
 288                                 if (inode) {
 289                                         mb = *pending;
 290                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 291                                                              &mb, inode,frags);
 292                                         if (rc)
 293                                                 CDEBUG(D_ERROR,
 294                                                        "can't get extra "
 295                                                        "meta blocks.\n");
 296                                         else
 297                                                 *pending += mb;
 298                                 }
 299                                 lqs->lqs_bwrite_pending += *pending;
 300                         } else {
 301                                 *pending = count;
 302                                 lqs->lqs_iwrite_pending += *pending;
 303                         }
 304                 }
 305
 306                 /* if xx_rec < 0, that means quota are releasing,
 307                  * and it may return before we use quota. So if
 308                  * we find this situation, we assuming it has
 309                  * returned b=18491 */
 310                 if (isblk && lqs->lqs_blk_rec < 0) {
 311                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 312                                 qdata[i].qd_count = 0;
 313                         else
 314                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 315                 }
 316                 if (!isblk && lqs->lqs_ino_rec < 0) {
 317                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 318                                 qdata[i].qd_count = 0;
 319                         else
 320                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 321                 }
 322
 323
 324                 CDEBUG(D_QUOTA, "count: %d, lqs pending: %lu, qd_count: "LPU64
 325                        ", metablocks: %d, isblk: %d, pending: %d.\n", count,
 326                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 327                        qdata[i].qd_count, mb, isblk, *pending);
 328                 if (rc2[i] == QUOTA_RET_OK) {
 329                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 330                                 rc2[i] = QUOTA_RET_ACQUOTA;
 331                         if (!isblk && qdata[i].qd_count <
 332                             lqs->lqs_iwrite_pending)
 333                                 rc2[i] = QUOTA_RET_ACQUOTA;
 334                 }
 335
 336                 spin_unlock(&lqs->lqs_lock);
 337
 338                 if (lqs->lqs_blk_rec  < 0 &&
 339                     qdata[i].qd_count <
 340                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 341                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 342
 343                 /* When cycle is zero, lqs_*_pending will be changed. We will
 344                  * get reference of the lqs here and put reference of lqs in
 345                  * quota_pending_commit b=14784 */
 346                 if (!cycle)
 347                         lqs_getref(lqs);
 348
 349                 /* this is for quota_search_lqs */
 350                 lqs_putref(lqs);
 351         }
 352
 353         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 354                 RETURN(QUOTA_RET_ACQUOTA);
 355         else
 356                 RETURN(rc);
 357 }
 358
 359 static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid,
 360                                 unsigned int gid, int count, int *pending,
 361                                 quota_acquire acquire,
 362                                 struct obd_trans_info *oti, int isblk,
 363                                 struct inode *inode, int frags)
 364 {
 365         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 366         struct timeval work_start;
 367         struct timeval work_end;
 368         long timediff;
 369         struct l_wait_info lwi = { 0 };
 370         int rc = 0, cycle = 0, count_err = 1;
 371         ENTRY;
 372
 373         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 374         *pending = 0;
 375         /* Unfortunately, if quota master is too busy to handle the
 376          * pre-dqacq in time and quota hash on ost is used up, we
 377          * have to wait for the completion of in flight dqacq/dqrel,
 378          * in order to get enough quota for write b=12588 */
 379         do_gettimeofday(&work_start);
 380         while ((rc = quota_check_common(obd, uid, gid, count, cycle, isblk,
 381                                         inode, frags, pending)) &
 382                QUOTA_RET_ACQUOTA) {
 383
 384                 spin_lock(&qctxt->lqc_lock);
 385                 if (!qctxt->lqc_import && oti) {
 386                         spin_unlock(&qctxt->lqc_lock);
 387
 388                         LASSERT(oti && oti->oti_thread &&
 389                                 oti->oti_thread->t_watchdog);
 390
 391                         lc_watchdog_disable(oti->oti_thread->t_watchdog);
 392                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 393                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 394                                      &lwi);
 395                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 396                         lc_watchdog_touch(oti->oti_thread->t_watchdog);
 397                 } else {
 398                         spin_unlock(&qctxt->lqc_lock);
 399                 }
 400
 401                 cycle++;
 402                 if (isblk)
 403                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 404                 /* after acquire(), we should run quota_check_common again
 405                  * so that we confirm there are enough quota to finish write */
 406                 rc = acquire(obd, uid, gid, oti, isblk);
 407
 408                 /* please reference to dqacq_completion for the below */
 409                 /* a new request is finished, try again */
 410                 if (rc == QUOTA_REQ_RETURNED) {
 411                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 412                         continue;
 413                 }
 414
 415                 /* it is out of quota already */
 416                 if (rc == -EDQUOT) {
 417                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 418                         break;
 419                 }
 420
 421                 /* -EBUSY and others, wait a second and try again */
 422                 if (rc < 0) {
 423                         cfs_waitq_t        waitq;
 424                         struct l_wait_info lwi;
 425
 426                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 427                                 lc_watchdog_touch(oti->oti_thread->t_watchdog);
 428                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 429                                count_err++);
 430
 431                         init_waitqueue_head(&waitq);
 432                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 433                                           NULL);
 434                         l_wait_event(waitq, 0, &lwi);
 435                 }
 436
 437                 if (rc < 0 || cycle % 10 == 2) {
 438                         spin_lock(&last_print_lock);
 439                         if (last_print == 0 ||
 440                             cfs_time_before((last_print + cfs_time_seconds(30)),
 441                                             cfs_time_current())) {
 442                                 last_print = cfs_time_current();
 443                                 spin_unlock(&last_print_lock);
 444                                 CWARN("still haven't managed to acquire quota "
 445                                       "space from the quota master after %d "
 446                                       "retries (err=%d, rc=%d)\n",
 447                                       cycle, count_err - 1, rc);
 448                         } else {
 449                                 spin_unlock(&last_print_lock);
 450                         }
 451                 }
 452
 453                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 454                        cycle);
 455         }
 456         do_gettimeofday(&work_end);
 457         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 458         lprocfs_counter_add(qctxt->lqc_stats,
 459                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 460                                     LQUOTA_WAIT_FOR_CHK_INO,
 461                             timediff);
 462
 463         RETURN(rc);
 464 }
 465
 466 /**
 467  * when a block_write or inode_create rpc is finished, adjust the record for
 468  * pending blocks and inodes
 469  */
 470 static int quota_pending_commit(struct obd_device *obd, unsigned int uid,
 471                                 unsigned int gid, int pending, int isblk)
 472 {
 473         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 474         struct timeval work_start;
 475         struct timeval work_end;
 476         long timediff;
 477         int i;
 478         __u32 id[MAXQUOTAS] = { uid, gid };
 479         struct qunit_data qdata[MAXQUOTAS];
 480         ENTRY;
 481
 482         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 483         CLASSERT(MAXQUOTAS < 4);
 484         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 485                 RETURN(0);
 486
 487         do_gettimeofday(&work_start);
 488         for (i = 0; i < MAXQUOTAS; i++) {
 489                 struct lustre_qunit_size *lqs = NULL;
 490
 491                 qdata[i].qd_id = id[i];
 492                 qdata[i].qd_flags = i;
 493                 if (isblk)
 494                         QDATA_SET_BLK(&qdata[i]);
 495                 qdata[i].qd_count = 0;
 496
 497                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 498                         continue;
 499
 500                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 501                 if (lqs) {
 502                         int flag = 0;
 503                         spin_lock(&lqs->lqs_lock);
 504                         if (isblk) {
 505                                 if (lqs->lqs_bwrite_pending >= pending) {
 506                                         lqs->lqs_bwrite_pending -= pending;
 507                                         spin_unlock(&lqs->lqs_lock);
 508                                         flag = 1;
 509                                 } else {
 510                                         spin_unlock(&lqs->lqs_lock);
 511                                         CDEBUG(D_ERROR,
 512                                                "there are too many blocks!\n");
 513                                 }
 514                         } else {
 515                                 if (lqs->lqs_iwrite_pending >= pending) {
 516                                         lqs->lqs_iwrite_pending -= pending;
 517                                         spin_unlock(&lqs->lqs_lock);
 518                                         flag = 1;
 519                                 } else {
 520                                         spin_unlock(&lqs->lqs_lock);
 521                                         CDEBUG(D_ERROR,
 522                                                "there are too many files!\n");
 523                                 }
 524                         }
 525                         CDEBUG(D_QUOTA, "lqs pending: %lu, pending: %d, "
 526                                "isblk: %d.\n",
 527                                isblk ? lqs->lqs_bwrite_pending :
 528                                lqs->lqs_iwrite_pending, pending, isblk);
 529
 530                         lqs_putref(lqs);
 531                         /* When lqs_*_pening is changed back, we'll putref lqs
 532                          * here b=14784 */
 533                         if (flag)
 534                                 lqs_putref(lqs);
 535                 }
 536         }
 537         do_gettimeofday(&work_end);
 538         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 539         lprocfs_counter_add(qctxt->lqc_stats,
 540                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 541                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 542                             timediff);
 543
 544         RETURN(0);
 545 }
 546
 547 static int mds_quota_init(void)
 548 {
 549         return lustre_dquot_init();
 550 }
 551
 552 static int mds_quota_exit(void)
 553 {
 554         lustre_dquot_exit();
 555         return 0;
 556 }
 557
 558 static int mds_quota_setup(struct obd_device *obd)
 559 {
 560         struct obd_device_target *obt = &obd->u.obt;
 561         struct mds_obd *mds = &obd->u.mds;
 562         int rc;
 563         ENTRY;
 564
 565         if (unlikely(mds->mds_quota)) {
 566                 CWARN("try to reinitialize quota context!\n");
 567                 RETURN(0);
 568         }
 569
 570         init_rwsem(&obt->obt_rwsem);
 571         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 572         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 573         atomic_set(&obt->obt_quotachecking, 1);
 574         /* initialize quota master and quota context */
 575         sema_init(&mds->mds_qonoff_sem, 1);
 576         rc = qctxt_init(obd, dqacq_handler);
 577         if (rc) {
 578                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 579                 RETURN(rc);
 580         }
 581         mds->mds_quota = 1;
 582         RETURN(rc);
 583 }
 584
 585 static int mds_quota_cleanup(struct obd_device *obd)
 586 {
 587         ENTRY;
 588         if (unlikely(!obd->u.mds.mds_quota))
 589                 RETURN(0);
 590
 591         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 592         RETURN(0);
 593 }
 594
 595 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 596 {
 597         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 598         ENTRY;
 599
 600         if (unlikely(!obd->u.mds.mds_quota))
 601                 RETURN(0);
 602
 603         if (data != NULL)
 604                 QUOTA_MASTER_READY(qctxt);
 605         else
 606                 QUOTA_MASTER_UNREADY(qctxt);
 607         RETURN(0);
 608 }
 609
 610 static int mds_quota_fs_cleanup(struct obd_device *obd)
 611 {
 612         struct mds_obd *mds = &obd->u.mds;
 613         struct obd_quotactl oqctl;
 614         ENTRY;
 615
 616         if (unlikely(!mds->mds_quota))
 617                 RETURN(0);
 618
 619         mds->mds_quota = 0;
 620         memset(&oqctl, 0, sizeof(oqctl));
 621         oqctl.qc_type = UGQUOTA;
 622
 623         down(&mds->mds_qonoff_sem);
 624         mds_admin_quota_off(obd, &oqctl);
 625         up(&mds->mds_qonoff_sem);
 626         RETURN(0);
 627 }
 628
 629 static int quota_acquire_common(struct obd_device *obd, unsigned int uid,
 630                                 unsigned int gid, struct obd_trans_info *oti,
 631                                 int isblk)
 632 {
 633         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 634         int rc;
 635         ENTRY;
 636
 637         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, isblk, 1, oti);
 638         RETURN(rc);
 639 }
 640
 641 #endif /* HAVE_QUOTA_SUPPORT */
 642 #endif /* __KERNEL__ */
 643
 644 struct osc_quota_info {
 645         struct list_head        oqi_hash;       /* hash list */
 646         struct client_obd      *oqi_cli;        /* osc obd */
 647         unsigned int            oqi_id;         /* uid/gid of a file */
 648         short                   oqi_type;       /* quota type */
 649 };
 650
 651 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 652
 653 static struct list_head qinfo_hash[NR_DQHASH];
 654 /* SLAB cache for client quota context */
 655 cfs_mem_cache_t *qinfo_cachep = NULL;
 656
 657 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 658                          __attribute__((__const__));
 659
 660 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 661 {
 662         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 663         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 664         return tmp;
 665 }
 666
 667 /* caller must hold qinfo_list_lock */
 668 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 669 {
 670         struct list_head *head = qinfo_hash +
 671                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 672
 673         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 674         list_add(&oqi->oqi_hash, head);
 675 }
 676
 677 /* caller must hold qinfo_list_lock */
 678 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 679 {
 680         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 681         list_del_init(&oqi->oqi_hash);
 682 }
 683
 684 /* caller must hold qinfo_list_lock */
 685 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 686                                                 unsigned int id, int type)
 687 {
 688         unsigned int hashent = hashfn(cli, id, type);
 689         struct osc_quota_info *oqi;
 690         ENTRY;
 691
 692         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 693         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 694                 if (oqi->oqi_cli == cli &&
 695                     oqi->oqi_id == id && oqi->oqi_type == type)
 696                         return oqi;
 697         }
 698         RETURN(NULL);
 699 }
 700
 701 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 702                                           unsigned int id, int type)
 703 {
 704         struct osc_quota_info *oqi;
 705         ENTRY;
 706
 707         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 708         if(!oqi)
 709                 RETURN(NULL);
 710
 711         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 712         oqi->oqi_cli = cli;
 713         oqi->oqi_id = id;
 714         oqi->oqi_type = type;
 715
 716         RETURN(oqi);
 717 }
 718
 719 static void free_qinfo(struct osc_quota_info *oqi)
 720 {
 721         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 722 }
 723
 724 int osc_quota_chkdq(struct client_obd *cli, unsigned int uid, unsigned int gid)
 725 {
 726         unsigned int id;
 727         int cnt, rc = QUOTA_OK;
 728         ENTRY;
 729
 730         spin_lock(&qinfo_list_lock);
 731         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 732                 struct osc_quota_info *oqi = NULL;
 733
 734                 id = (cnt == USRQUOTA) ? uid : gid;
 735                 oqi = find_qinfo(cli, id, cnt);
 736                 if (oqi) {
 737                         rc = NO_QUOTA;
 738                         break;
 739                 }
 740         }
 741         spin_unlock(&qinfo_list_lock);
 742
 743         RETURN(rc);
 744 }
 745
 746 int osc_quota_setdq(struct client_obd *cli, unsigned int uid, unsigned int gid,
 747                     obd_flag valid, obd_flag flags)
 748 {
 749         unsigned int id;
 750         obd_flag noquota;
 751         int cnt, rc = 0;
 752         ENTRY;
 753
 754
 755         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 756                 struct osc_quota_info *oqi, *old;
 757
 758                 if (!(valid & ((cnt == USRQUOTA) ?
 759                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 760                         continue;
 761
 762                 id = (cnt == USRQUOTA) ? uid : gid;
 763                 noquota = (cnt == USRQUOTA) ?
 764                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 765
 766                 oqi = alloc_qinfo(cli, id, cnt);
 767                 if (oqi) {
 768                         spin_lock(&qinfo_list_lock);
 769
 770                         old = find_qinfo(cli, id, cnt);
 771                         if (old && !noquota)
 772                                 remove_qinfo_hash(old);
 773                         else if (!old && noquota)
 774                                 insert_qinfo_hash(oqi);
 775
 776                         spin_unlock(&qinfo_list_lock);
 777
 778                         if (old || !noquota)
 779                                 free_qinfo(oqi);
 780                         if (old && !noquota)
 781                                 free_qinfo(old);
 782                 } else {
 783                         CERROR("not enough mem!\n");
 784                         rc = -ENOMEM;
 785                         break;
 786                 }
 787         }
 788
 789         RETURN(rc);
 790 }
 791
 792 int osc_quota_cleanup(struct obd_device *obd)
 793 {
 794         struct client_obd *cli = &obd->u.cli;
 795         struct osc_quota_info *oqi, *n;
 796         int i;
 797         ENTRY;
 798
 799         spin_lock(&qinfo_list_lock);
 800         for (i = 0; i < NR_DQHASH; i++) {
 801                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 802                         if (oqi->oqi_cli != cli)
 803                                 continue;
 804                         remove_qinfo_hash(oqi);
 805                         free_qinfo(oqi);
 806                 }
 807         }
 808         spin_unlock(&qinfo_list_lock);
 809
 810         RETURN(0);
 811 }
 812
 813 int osc_quota_init(void)
 814 {
 815         int i;
 816         ENTRY;
 817
 818         LASSERT(qinfo_cachep == NULL);
 819         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 820                                             sizeof(struct osc_quota_info),
 821                                             0, 0);
 822         if (!qinfo_cachep)
 823                 RETURN(-ENOMEM);
 824
 825         for (i = 0; i < NR_DQHASH; i++)
 826                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 827
 828         RETURN(0);
 829 }
 830
 831 int osc_quota_exit(void)
 832 {
 833         struct osc_quota_info *oqi, *n;
 834         int i, rc;
 835         ENTRY;
 836
 837         spin_lock(&qinfo_list_lock);
 838         for (i = 0; i < NR_DQHASH; i++) {
 839                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 840                         remove_qinfo_hash(oqi);
 841                         free_qinfo(oqi);
 842                 }
 843         }
 844         spin_unlock(&qinfo_list_lock);
 845
 846         rc = cfs_mem_cache_destroy(qinfo_cachep);
 847         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 848         qinfo_cachep = NULL;
 849
 850         RETURN(0);
 851 }
 852
 853 #ifdef __KERNEL__
 854 #ifdef HAVE_QUOTA_SUPPORT
 855 quota_interface_t mds_quota_interface = {
 856         .quota_init     = mds_quota_init,
 857         .quota_exit     = mds_quota_exit,
 858         .quota_setup    = mds_quota_setup,
 859         .quota_cleanup  = mds_quota_cleanup,
 860         .quota_check    = target_quota_check,
 861         .quota_ctl      = mds_quota_ctl,
 862         .quota_setinfo  = mds_quota_setinfo,
 863         .quota_fs_cleanup = mds_quota_fs_cleanup,
 864         .quota_recovery = mds_quota_recovery,
 865         .quota_adjust   = mds_quota_adjust,
 866         .quota_chkquota = quota_chk_acq_common,
 867         .quota_acquire  = quota_acquire_common,
 868         .quota_pending_commit = quota_pending_commit,
 869 };
 870
 871 quota_interface_t filter_quota_interface = {
 872         .quota_setup    = filter_quota_setup,
 873         .quota_cleanup  = filter_quota_cleanup,
 874         .quota_check    = target_quota_check,
 875         .quota_ctl      = filter_quota_ctl,
 876         .quota_setinfo  = filter_quota_setinfo,
 877         .quota_clearinfo = filter_quota_clearinfo,
 878         .quota_enforce  = filter_quota_enforce,
 879         .quota_getflag  = filter_quota_getflag,
 880         .quota_acquire  = quota_acquire_common,
 881         .quota_adjust   = filter_quota_adjust,
 882         .quota_chkquota = quota_chk_acq_common,
 883         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 884         .quota_pending_commit = quota_pending_commit,
 885 };
 886 #endif
 887 #endif /* __KERNEL__ */
 888
 889 quota_interface_t mdc_quota_interface = {
 890         .quota_ctl      = client_quota_ctl,
 891         .quota_check    = client_quota_check,
 892         .quota_poll_check = client_quota_poll_check,
 893 };
 894
 895 quota_interface_t lmv_quota_interface = {
 896         .quota_ctl      = lmv_quota_ctl,
 897         .quota_check    = lmv_quota_check,
 898 };
 899
 900 quota_interface_t osc_quota_interface = {
 901         .quota_ctl      = client_quota_ctl,
 902         .quota_check    = client_quota_check,
 903         .quota_poll_check = client_quota_poll_check,
 904         .quota_init     = osc_quota_init,
 905         .quota_exit     = osc_quota_exit,
 906         .quota_chkdq    = osc_quota_chkdq,
 907         .quota_setdq    = osc_quota_setdq,
 908         .quota_cleanup  = osc_quota_cleanup,
 909         .quota_adjust_qunit = client_quota_adjust_qunit,
 910 };
 911
 912 quota_interface_t lov_quota_interface = {
 913         .quota_ctl      = lov_quota_ctl,
 914         .quota_check    = lov_quota_check,
 915         .quota_adjust_qunit = lov_quota_adjust_qunit,
 916 };
 917
 918 #ifdef __KERNEL__
 919
 920 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 921
 922 static int __init init_lustre_quota(void)
 923 {
 924 #ifdef HAVE_QUOTA_SUPPORT
 925         int rc = 0;
 926
 927         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 928                                                 proc_lustre_root,
 929                                                 NULL, NULL);
 930         if (IS_ERR(lquota_type_proc_dir)) {
 931                 CERROR("LProcFS failed in lquota-init\n");
 932                 rc = PTR_ERR(lquota_type_proc_dir);
 933                 return rc;
 934         }
 935
 936         rc = qunit_cache_init();
 937         if (rc)
 938                 return rc;
 939
 940         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 941         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 942 #endif
 943         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
 944         PORTAL_SYMBOL_REGISTER(lmv_quota_interface);
 945         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
 946         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
 947         return 0;
 948 }
 949
 950 static void /*__exit*/ exit_lustre_quota(void)
 951 {
 952         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
 953         PORTAL_SYMBOL_UNREGISTER(lmv_quota_interface);
 954         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
 955         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
 956 #ifdef HAVE_QUOTA_SUPPORT
 957         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 958         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 959
 960         qunit_cache_cleanup();
 961
 962         if (lquota_type_proc_dir)
 963                 lprocfs_remove(&lquota_type_proc_dir);
 964 #endif
 965 }
 966
 967 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 968 MODULE_DESCRIPTION("Lustre Quota");
 969 MODULE_LICENSE("GPL");
 970
 971 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
 972
 973 #ifdef HAVE_QUOTA_SUPPORT
 974 EXPORT_SYMBOL(mds_quota_interface);
 975 EXPORT_SYMBOL(filter_quota_interface);
 976 #endif
 977 EXPORT_SYMBOL(mdc_quota_interface);
 978 EXPORT_SYMBOL(lmv_quota_interface);
 979 EXPORT_SYMBOL(osc_quota_interface);
 980 EXPORT_SYMBOL(lov_quota_interface);
 981 #endif /* __KERNEL */