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