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