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