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