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 #ifdef HAVE_QUOTA64
  80         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  81 #else
  82         obt->obt_qfmt = LUSTRE_QUOTA_V1;
  83 #endif
  84         atomic_set(&obt->obt_quotachecking, 1);
  85         rc = qctxt_init(obd, NULL);
  86         if (rc)
  87                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  88
  89         RETURN(rc);
  90 }
  91
  92 static int filter_quota_cleanup(struct obd_device *obd)
  93 {
  94         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  95         return 0;
  96 }
  97
  98 static int filter_quota_setinfo(struct obd_export *exp, struct obd_device *obd)
  99 {
 100         struct obd_import *imp;
 101         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 102         ENTRY;
 103
 104         /* setup the quota context import */
 105         spin_lock(&obd->u.obt.obt_qctxt.lqc_lock);
 106         obd->u.obt.obt_qctxt.lqc_import = exp->exp_imp_reverse;
 107         spin_unlock(&obd->u.obt.obt_qctxt.lqc_lock);
 108         CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated now, \n",
 109                obd->obd_name,exp->exp_imp_reverse, obd);
 110
 111         /* make imp's connect flags equal relative exp's connect flags
 112          * adding it to avoid the scan export list
 113          */
 114         imp = exp->exp_imp_reverse;
 115         if (imp)
 116                 imp->imp_connect_data.ocd_connect_flags |=
 117                         (exp->exp_connect_flags &
 118                          (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 119
 120         cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 121         /* start quota slave recovery thread. (release high limits) */
 122         qslave_start_recovery(obd, &obd->u.obt.obt_qctxt);
 123         RETURN(0);
 124 }
 125
 126 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 127 {
 128         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 129         ENTRY;
 130
 131         /* lquota may be not set up before destroying export, b=14896 */
 132         if (!obd->obd_set_up)
 133                 RETURN(0);
 134
 135         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 136          * should be invalid b=12374 */
 137         if (qctxt->lqc_import && qctxt->lqc_import == exp->exp_imp_reverse) {
 138                 spin_lock(&qctxt->lqc_lock);
 139                 qctxt->lqc_import = NULL;
 140                 spin_unlock(&qctxt->lqc_lock);
 141                 ptlrpc_cleanup_imp(exp->exp_imp_reverse);
 142                 dqacq_interrupt(qctxt);
 143                 CDEBUG(D_QUOTA, "%s: lqc_import of obd(%p) is invalid now.\n",
 144                        obd->obd_name, obd);
 145         }
 146         RETURN(0);
 147 }
 148
 149 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 150 {
 151         ENTRY;
 152
 153         if (!sb_any_quota_enabled(obd->u.obt.obt_sb))
 154                 RETURN(0);
 155
 156         if (ignore) {
 157                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 158                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 159         } else {
 160                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 161         }
 162
 163         RETURN(0);
 164 }
 165
 166 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 167 {
 168         struct obd_device_target *obt = &obd->u.obt;
 169         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 170         int err, cnt, rc = 0;
 171         struct obd_quotactl *oqctl;
 172         ENTRY;
 173
 174         if (!sb_any_quota_enabled(obt->obt_sb))
 175                 RETURN(0);
 176
 177         OBD_ALLOC_PTR(oqctl);
 178         if (!oqctl) {
 179                 CERROR("Not enough memory!");
 180                 RETURN(-ENOMEM);
 181         }
 182
 183         /* set over quota flags for a uid/gid */
 184         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 185         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 186
 187         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 188                 struct quota_adjust_qunit oqaq_tmp;
 189                 struct lustre_qunit_size *lqs = NULL;
 190
 191                 oqaq_tmp.qaq_flags = cnt;
 192                 oqaq_tmp.qaq_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 193
 194                 quota_search_lqs(NULL, &oqaq_tmp, qctxt, &lqs);
 195                 if (lqs) {
 196                         spin_lock(&lqs->lqs_lock);
 197                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 198                                 oa->o_flags |= (cnt == USRQUOTA) ?
 199                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 200                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 201                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 202                                        qctxt->lqc_sync_blk);
 203                                 spin_unlock(&lqs->lqs_lock);
 204                                 /* this is for quota_search_lqs */
 205                                 lqs_putref(lqs);
 206                                 continue;
 207                         }
 208                         spin_unlock(&lqs->lqs_lock);
 209                         /* this is for quota_search_lqs */
 210                         lqs_putref(lqs);
 211                 }
 212
 213                 memset(oqctl, 0, sizeof(*oqctl));
 214
 215                 oqctl->qc_cmd = Q_GETQUOTA;
 216                 oqctl->qc_type = cnt;
 217                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 218                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 219                 if (err) {
 220                         if (!rc)
 221                                 rc = err;
 222                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 223                                                              OBD_MD_FLGRPQUOTA);
 224                         continue;
 225                 }
 226
 227                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 228                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 229                     oqctl->qc_dqblk.dqb_bhardlimit))
 230                         oa->o_flags |= (cnt == USRQUOTA) ?
 231                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 232         }
 233         OBD_FREE_PTR(oqctl);
 234         RETURN(rc);
 235 }
 236
 237 static int filter_quota_acquire(struct obd_device *obd, unsigned int uid,
 238                                 unsigned int gid, struct obd_trans_info *oti)
 239 {
 240         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 241         int rc;
 242         ENTRY;
 243
 244         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, LQUOTA_FLAGS_BLK, 1, oti);
 245         RETURN(rc);
 246 }
 247
 248 /* check whether the left quota of certain uid and gid can satisfy a block_write
 249  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA */
 250 static int quota_check_common(struct obd_device *obd, unsigned int uid,
 251                               unsigned int gid, int count, int cycle, int isblk,
 252                               struct inode *inode, int frags, int *pending)
 253 {
 254         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 255         int i;
 256         __u32 id[MAXQUOTAS] = { uid, gid };
 257         struct qunit_data qdata[MAXQUOTAS];
 258         int mb = 0;
 259         int rc = 0, rc2[2] = { 0, 0 };
 260         ENTRY;
 261
 262         CLASSERT(MAXQUOTAS < 4);
 263         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 264                 RETURN(rc);
 265
 266         spin_lock(&qctxt->lqc_lock);
 267         if (!qctxt->lqc_valid){
 268                 spin_unlock(&qctxt->lqc_lock);
 269                 RETURN(rc);
 270         }
 271         spin_unlock(&qctxt->lqc_lock);
 272
 273         for (i = 0; i < MAXQUOTAS; i++) {
 274                 struct lustre_qunit_size *lqs = NULL;
 275
 276                 qdata[i].qd_id = id[i];
 277                 qdata[i].qd_flags = i;
 278                 if (isblk)
 279                         QDATA_SET_BLK(&qdata[i]);
 280                 qdata[i].qd_count = 0;
 281
 282                 /* ignore root user */
 283                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 284                         continue;
 285
 286                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 287                 if (!lqs)
 288                         continue;
 289
 290                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 291                 spin_lock(&lqs->lqs_lock);
 292                 if (!cycle) {
 293                         if (isblk) {
 294                                 *pending = count * CFS_PAGE_SIZE;
 295                                 /* in order to complete this write, we need extra
 296                                  * meta blocks. This function can get it through
 297                                  * data needed to be written b=16542 */
 298                                 mb = *pending;
 299                                 LASSERT(inode && frags > 0);
 300                                 if (fsfilt_get_mblk(obd, qctxt->lqc_sb, &mb,
 301                                                     inode, frags) < 0)
 302                                         CDEBUG(D_ERROR,
 303                                                "can't get extra meta blocks.\n");
 304                                 else
 305                                         *pending += mb;
 306                                 lqs->lqs_bwrite_pending += *pending;
 307                         } else {
 308                                 *pending = count;
 309                                 lqs->lqs_iwrite_pending += *pending;
 310                         }
 311                 }
 312
 313                 /* if xx_rec < 0, that means quota are releasing,
 314                  * and it may return before we use quota. So if
 315                  * we find this situation, we assuming it has
 316                  * returned b=18491 */
 317                 if (isblk && lqs->lqs_blk_rec < 0) {
 318                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 319                                 qdata[i].qd_count = 0;
 320                         else
 321                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 322                 }
 323                 if (!isblk && lqs->lqs_ino_rec < 0) {
 324                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 325                                 qdata[i].qd_count = 0;
 326                         else
 327                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 328                 }
 329
 330                 CDEBUG(D_QUOTA, "count: %d, lqs pending: %lu, qd_count: "LPU64
 331                        ", metablocks: %d, isblk: %d, pending: %d.\n", count,
 332                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 333                        qdata[i].qd_count, mb, isblk, *pending);
 334                 if (rc2[i] == QUOTA_RET_OK) {
 335                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 336                                 rc2[i] = QUOTA_RET_ACQUOTA;
 337                         if (!isblk && qdata[i].qd_count <
 338                             lqs->lqs_iwrite_pending)
 339                                 rc2[i] = QUOTA_RET_ACQUOTA;
 340                 }
 341
 342                 spin_unlock(&lqs->lqs_lock);
 343
 344                 if (lqs->lqs_blk_rec  < 0 &&
 345                     qdata[i].qd_count <
 346                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 347                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 348
 349                 /* When cycle is zero, lqs_*_pending will be changed. We will
 350                  * get reference of the lqs here and put reference of lqs in
 351                  * quota_pending_commit b=14784 */
 352                 if (!cycle)
 353                         lqs_getref(lqs);
 354
 355                 /* this is for quota_search_lqs */
 356                 lqs_putref(lqs);
 357         }
 358
 359         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 360                 RETURN(QUOTA_RET_ACQUOTA);
 361         else
 362                 RETURN(rc);
 363 }
 364
 365 static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid,
 366                                 unsigned int gid, int count, int *pending,
 367                                 int isblk, quota_acquire acquire,
 368                                 struct obd_trans_info *oti, struct inode *inode,
 369                                 int frags)
 370 {
 371         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 372         struct timeval work_start;
 373         struct timeval work_end;
 374         long timediff;
 375         struct l_wait_info lwi = { 0 };
 376         int rc = 0, cycle = 0, count_err = 1;
 377         ENTRY;
 378
 379         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 380         /* Unfortunately, if quota master is too busy to handle the
 381          * pre-dqacq in time and quota hash on ost is used up, we
 382          * have to wait for the completion of in flight dqacq/dqrel,
 383          * in order to get enough quota for write b=12588 */
 384         do_gettimeofday(&work_start);
 385         while ((rc = quota_check_common(obd, uid, gid, count, cycle, isblk,
 386                                         inode, frags, pending)) & QUOTA_RET_ACQUOTA) {
 387
 388                 spin_lock(&qctxt->lqc_lock);
 389                 if (!qctxt->lqc_import && oti) {
 390                         spin_unlock(&qctxt->lqc_lock);
 391
 392                         LASSERT(oti && oti->oti_thread &&
 393                                 oti->oti_thread->t_watchdog);
 394
 395                         lc_watchdog_disable(oti->oti_thread->t_watchdog);
 396                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 397                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 398                                      &lwi);
 399                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 400                         lc_watchdog_touch(oti->oti_thread->t_watchdog);
 401                 } else {
 402                         spin_unlock(&qctxt->lqc_lock);
 403                 }
 404
 405                 cycle++;
 406                 if (isblk)
 407                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 408                 /* after acquire(), we should run quota_check_common again
 409                  * so that we confirm there are enough quota to finish write */
 410                 rc = acquire(obd, uid, gid, oti);
 411
 412                 /* please reference to dqacq_completion for the below */
 413                 /* a new request is finished, try again */
 414                 if (rc == QUOTA_REQ_RETURNED) {
 415                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 416                         continue;
 417                 }
 418
 419                 /* it is out of quota already */
 420                 if (rc == -EDQUOT) {
 421                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 422                         break;
 423                 }
 424
 425                 /* -EBUSY and others, wait a second and try again */
 426                 if (rc < 0) {
 427                         cfs_waitq_t        waitq;
 428                         struct l_wait_info lwi;
 429
 430                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 431                                 lc_watchdog_touch(oti->oti_thread->t_watchdog);
 432                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 433                                count_err++);
 434
 435                         init_waitqueue_head(&waitq);
 436                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 437                                           NULL);
 438                         l_wait_event(waitq, 0, &lwi);
 439                 }
 440
 441                 if (rc < 0 || cycle % 10 == 2) {
 442                         spin_lock(&last_print_lock);
 443                         if (last_print == 0 ||
 444                             cfs_time_before((last_print + cfs_time_seconds(30)),
 445                                             cfs_time_current())) {
 446                                 CWARN("still haven't managed to acquire quota "
 447                                       "space from the quota master after %d "
 448                                       "retries (err=%d, rc=%d)\n",
 449                                       cycle, count_err - 1, rc);
 450                                 last_print = cfs_time_current();
 451                         }
 452                         spin_unlock(&last_print_lock);
 453                 }
 454
 455                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 456                        cycle);
 457         }
 458
 459         do_gettimeofday(&work_end);
 460         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 461         lprocfs_counter_add(qctxt->lqc_stats,
 462                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 463                                     LQUOTA_WAIT_FOR_CHK_INO,
 464                             timediff);
 465
 466         RETURN(rc);
 467 }
 468
 469 static int filter_quota_check(struct obd_device *obd, unsigned int uid,
 470                               unsigned int gid, int npage, int *pending,
 471                               quota_acquire acquire, struct obd_trans_info *oti,
 472                               struct inode *inode, int frags)
 473 {
 474         return quota_chk_acq_common(obd, uid, gid, npage, pending, LQUOTA_FLAGS_BLK,
 475                                     acquire, oti, inode, frags);
 476 }
 477
 478 /* when a block_write or inode_create rpc is finished, adjust the record for
 479  * pending blocks and inodes*/
 480 static int quota_pending_commit(struct obd_device *obd, unsigned int uid,
 481                                 unsigned int gid, int pending, int isblk)
 482 {
 483         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 484         struct timeval work_start;
 485         struct timeval work_end;
 486         long timediff;
 487         int i;
 488         __u32 id[MAXQUOTAS] = { uid, gid };
 489         struct qunit_data qdata[MAXQUOTAS];
 490         ENTRY;
 491
 492         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 493         CLASSERT(MAXQUOTAS < 4);
 494         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 495                 RETURN(0);
 496
 497         do_gettimeofday(&work_start);
 498         for (i = 0; i < MAXQUOTAS; i++) {
 499                 struct lustre_qunit_size *lqs = NULL;
 500
 501                 qdata[i].qd_id = id[i];
 502                 qdata[i].qd_flags = i;
 503                 if (isblk)
 504                         QDATA_SET_BLK(&qdata[i]);
 505                 qdata[i].qd_count = 0;
 506
 507                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 508                         continue;
 509
 510                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 511                 if (lqs) {
 512                         int flag = 0;
 513                         spin_lock(&lqs->lqs_lock);
 514                         if (isblk) {
 515                                 if (lqs->lqs_bwrite_pending >= pending) {
 516                                         lqs->lqs_bwrite_pending -= pending;
 517                                         flag = 1;
 518                                 } else {
 519                                         CDEBUG(D_ERROR,
 520                                                "there are too many blocks!\n");
 521                                 }
 522                         } else {
 523                                 if (lqs->lqs_iwrite_pending >= pending) {
 524                                         lqs->lqs_iwrite_pending -= pending;
 525                                         flag = 1;
 526                                 } else {
 527                                         CDEBUG(D_ERROR,
 528                                                "there are too many files!\n");
 529                                 }
 530                         }
 531                         CDEBUG(D_QUOTA, "lqs pending: %lu, pending: %d, "
 532                                "isblk: %d.\n",
 533                                isblk ? lqs->lqs_bwrite_pending :
 534                                lqs->lqs_iwrite_pending, pending, isblk);
 535
 536                         spin_unlock(&lqs->lqs_lock);
 537                         lqs_putref(lqs);
 538                         /* When lqs_*_pening is changed back, we'll putref lqs
 539                          * here b=14784 */
 540                         if (flag)
 541                                 lqs_putref(lqs);
 542                 }
 543         }
 544         do_gettimeofday(&work_end);
 545         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 546         lprocfs_counter_add(qctxt->lqc_stats,
 547                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 548                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 549                             timediff);
 550
 551         RETURN(0);
 552 }
 553
 554 static int filter_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 555                                        unsigned int gid, int blocks)
 556 {
 557         return quota_pending_commit(obd, uid, gid, blocks, LQUOTA_FLAGS_BLK);
 558 }
 559
 560 static int mds_quota_init(void)
 561 {
 562         return lustre_dquot_init();
 563 }
 564
 565 static int mds_quota_exit(void)
 566 {
 567         lustre_dquot_exit();
 568         return 0;
 569 }
 570
 571 static int mds_quota_setup(struct obd_device *obd)
 572 {
 573         struct obd_device_target *obt = &obd->u.obt;
 574         struct mds_obd *mds = &obd->u.mds;
 575         int rc;
 576         ENTRY;
 577
 578 #ifdef HAVE_QUOTA64
 579         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 580 #else
 581         obt->obt_qfmt = LUSTRE_QUOTA_V1;
 582 #endif
 583         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 584         atomic_set(&obt->obt_quotachecking, 1);
 585         /* initialize quota master and quota context */
 586         sema_init(&mds->mds_qonoff_sem, 1);
 587         rc = qctxt_init(obd, dqacq_handler);
 588         if (rc) {
 589                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 590                 RETURN(rc);
 591         }
 592         RETURN(rc);
 593 }
 594
 595 static int mds_quota_cleanup(struct obd_device *obd)
 596 {
 597         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 598         RETURN(0);
 599 }
 600
 601 static int mds_quota_fs_cleanup(struct obd_device *obd)
 602 {
 603         struct mds_obd *mds = &obd->u.mds;
 604         struct obd_quotactl oqctl;
 605         ENTRY;
 606
 607         memset(&oqctl, 0, sizeof(oqctl));
 608         oqctl.qc_type = UGQUOTA;
 609
 610         down(&mds->mds_qonoff_sem);
 611         mds_admin_quota_off(obd, &oqctl);
 612         up(&mds->mds_qonoff_sem);
 613         RETURN(0);
 614 }
 615
 616 static int mds_quota_check(struct obd_device *obd, unsigned int uid,
 617                            unsigned int gid, int inodes, int *pending,
 618                            quota_acquire acquire, struct obd_trans_info *oti,
 619                            struct inode *inode, int frags)
 620 {
 621         return quota_chk_acq_common(obd, uid, gid, inodes, pending, 0,
 622                                     acquire, oti, inode, frags);
 623 }
 624
 625 static int mds_quota_acquire(struct obd_device *obd, unsigned int uid,
 626                              unsigned int gid, struct obd_trans_info *oti)
 627 {
 628         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 629         int rc;
 630         ENTRY;
 631
 632         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, 0, 1, oti);
 633         RETURN(rc);
 634 }
 635
 636 static int mds_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 637                                     unsigned int gid, int inodes)
 638 {
 639         return quota_pending_commit(obd, uid, gid, inodes, 0);
 640 }
 641 #endif /* HAVE_QUOTA_SUPPORT */
 642 #endif /* __KERNEL__ */
 643
 644 struct osc_quota_info {
 645         struct list_head        oqi_hash;       /* hash list */
 646         struct client_obd      *oqi_cli;        /* osc obd */
 647         unsigned int            oqi_id;         /* uid/gid of a file */
 648         short                   oqi_type;       /* quota type */
 649 };
 650
 651 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 652
 653 static struct list_head qinfo_hash[NR_DQHASH];
 654 /* SLAB cache for client quota context */
 655 cfs_mem_cache_t *qinfo_cachep = NULL;
 656
 657 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 658                          __attribute__((__const__));
 659
 660 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 661 {
 662         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 663         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 664         return tmp;
 665 }
 666
 667 /* caller must hold qinfo_list_lock */
 668 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 669 {
 670         struct list_head *head = qinfo_hash +
 671                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 672
 673         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 674         list_add(&oqi->oqi_hash, head);
 675 }
 676
 677 /* caller must hold qinfo_list_lock */
 678 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 679 {
 680         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 681         list_del_init(&oqi->oqi_hash);
 682 }
 683
 684 /* caller must hold qinfo_list_lock */
 685 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 686                                                 unsigned int id, int type)
 687 {
 688         unsigned int hashent = hashfn(cli, id, type);
 689         struct osc_quota_info *oqi;
 690
 691         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 692         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 693                 if (oqi->oqi_cli == cli &&
 694                     oqi->oqi_id == id && oqi->oqi_type == type)
 695                         return oqi;
 696         }
 697         return NULL;
 698 }
 699
 700 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 701                                           unsigned int id, int type)
 702 {
 703         struct osc_quota_info *oqi;
 704         ENTRY;
 705
 706         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 707         if(!oqi)
 708                 RETURN(NULL);
 709
 710         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 711         oqi->oqi_cli = cli;
 712         oqi->oqi_id = id;
 713         oqi->oqi_type = type;
 714
 715         RETURN(oqi);
 716 }
 717
 718 static void free_qinfo(struct osc_quota_info *oqi)
 719 {
 720         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 721 }
 722
 723 int osc_quota_chkdq(struct client_obd *cli, unsigned int uid, unsigned int gid)
 724 {
 725         unsigned int id;
 726         int cnt, rc = QUOTA_OK;
 727         ENTRY;
 728
 729         spin_lock(&qinfo_list_lock);
 730         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 731                 struct osc_quota_info *oqi = NULL;
 732
 733                 id = (cnt == USRQUOTA) ? uid : gid;
 734                 oqi = find_qinfo(cli, id, cnt);
 735                 if (oqi) {
 736                         rc = NO_QUOTA;
 737                         break;
 738                 }
 739         }
 740         spin_unlock(&qinfo_list_lock);
 741
 742         RETURN(rc);
 743 }
 744
 745 int osc_quota_setdq(struct client_obd *cli, unsigned int uid, unsigned int gid,
 746                     obd_flag valid, obd_flag flags)
 747 {
 748         unsigned int id;
 749         obd_flag noquota;
 750         int cnt, rc = 0;
 751         ENTRY;
 752
 753
 754         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 755                 struct osc_quota_info *oqi, *old;
 756
 757                 if (!(valid & ((cnt == USRQUOTA) ?
 758                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 759                         continue;
 760
 761                 id = (cnt == USRQUOTA) ? uid : gid;
 762                 noquota = (cnt == USRQUOTA) ?
 763                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 764
 765                 oqi = alloc_qinfo(cli, id, cnt);
 766                 if (oqi) {
 767                         spin_lock(&qinfo_list_lock);
 768
 769                         old = find_qinfo(cli, id, cnt);
 770                         if (old && !noquota)
 771                                 remove_qinfo_hash(old);
 772                         else if (!old && noquota)
 773                                 insert_qinfo_hash(oqi);
 774
 775                         spin_unlock(&qinfo_list_lock);
 776
 777                         if (old || !noquota)
 778                                 free_qinfo(oqi);
 779                         if (old && !noquota)
 780                                 free_qinfo(old);
 781                 } else {
 782                         CERROR("not enough mem!\n");
 783                         rc = -ENOMEM;
 784                         break;
 785                 }
 786         }
 787
 788         RETURN(rc);
 789 }
 790
 791 int osc_quota_cleanup(struct obd_device *obd)
 792 {
 793         struct client_obd *cli = &obd->u.cli;
 794         struct osc_quota_info *oqi, *n;
 795         int i;
 796         ENTRY;
 797
 798         spin_lock(&qinfo_list_lock);
 799         for (i = 0; i < NR_DQHASH; i++) {
 800                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 801                         if (oqi->oqi_cli != cli)
 802                                 continue;
 803                         remove_qinfo_hash(oqi);
 804                         free_qinfo(oqi);
 805                 }
 806         }
 807         spin_unlock(&qinfo_list_lock);
 808
 809         RETURN(0);
 810 }
 811
 812 int osc_quota_init(void)
 813 {
 814         int i;
 815         ENTRY;
 816
 817         LASSERT(qinfo_cachep == NULL);
 818         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 819                                             sizeof(struct osc_quota_info),
 820                                             0, 0);
 821         if (!qinfo_cachep)
 822                 RETURN(-ENOMEM);
 823
 824         for (i = 0; i < NR_DQHASH; i++)
 825                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 826
 827         RETURN(0);
 828 }
 829
 830 int osc_quota_exit(void)
 831 {
 832         struct osc_quota_info *oqi, *n;
 833         int i, rc;
 834         ENTRY;
 835
 836         spin_lock(&qinfo_list_lock);
 837         for (i = 0; i < NR_DQHASH; i++) {
 838                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 839                         remove_qinfo_hash(oqi);
 840                         free_qinfo(oqi);
 841                 }
 842         }
 843         spin_unlock(&qinfo_list_lock);
 844
 845         rc = cfs_mem_cache_destroy(qinfo_cachep);
 846         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 847         qinfo_cachep = NULL;
 848
 849         RETURN(0);
 850 }
 851
 852 #ifdef __KERNEL__
 853 #ifdef HAVE_QUOTA_SUPPORT
 854 quota_interface_t mds_quota_interface = {
 855         .quota_init     = mds_quota_init,
 856         .quota_exit     = mds_quota_exit,
 857         .quota_setup    = mds_quota_setup,
 858         .quota_cleanup  = mds_quota_cleanup,
 859         .quota_check    = target_quota_check,
 860         .quota_ctl      = mds_quota_ctl,
 861         .quota_fs_cleanup       =mds_quota_fs_cleanup,
 862         .quota_recovery = mds_quota_recovery,
 863         .quota_adjust   = mds_quota_adjust,
 864         .quota_chkquota = mds_quota_check,
 865         .quota_acquire  = mds_quota_acquire,
 866         .quota_pending_commit = mds_quota_pending_commit,
 867 };
 868
 869 quota_interface_t filter_quota_interface = {
 870         .quota_setup    = filter_quota_setup,
 871         .quota_cleanup  = filter_quota_cleanup,
 872         .quota_check    = target_quota_check,
 873         .quota_ctl      = filter_quota_ctl,
 874         .quota_setinfo  = filter_quota_setinfo,
 875         .quota_clearinfo = filter_quota_clearinfo,
 876         .quota_enforce  = filter_quota_enforce,
 877         .quota_getflag  = filter_quota_getflag,
 878         .quota_acquire  = filter_quota_acquire,
 879         .quota_adjust   = filter_quota_adjust,
 880         .quota_chkquota = filter_quota_check,
 881         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 882         .quota_pending_commit = filter_quota_pending_commit,
 883 };
 884 #endif
 885 #endif /* __KERNEL__ */
 886
 887 quota_interface_t mdc_quota_interface = {
 888         .quota_ctl      = client_quota_ctl,
 889         .quota_check    = client_quota_check,
 890         .quota_poll_check = client_quota_poll_check,
 891 };
 892
 893 quota_interface_t osc_quota_interface = {
 894         .quota_ctl      = client_quota_ctl,
 895         .quota_check    = client_quota_check,
 896         .quota_poll_check = client_quota_poll_check,
 897         .quota_init     = osc_quota_init,
 898         .quota_exit     = osc_quota_exit,
 899         .quota_chkdq    = osc_quota_chkdq,
 900         .quota_setdq    = osc_quota_setdq,
 901         .quota_cleanup  = osc_quota_cleanup,
 902         .quota_adjust_qunit = client_quota_adjust_qunit,
 903 };
 904
 905 quota_interface_t lov_quota_interface = {
 906         .quota_check    = lov_quota_check,
 907         .quota_ctl      = lov_quota_ctl,
 908         .quota_adjust_qunit = lov_quota_adjust_qunit,
 909 };
 910
 911 #ifdef __KERNEL__
 912
 913 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 914
 915 static int __init init_lustre_quota(void)
 916 {
 917 #ifdef HAVE_QUOTA_SUPPORT
 918         int rc = 0;
 919
 920         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 921                                                 proc_lustre_root,
 922                                                 NULL, NULL);
 923         if (IS_ERR(lquota_type_proc_dir)) {
 924                 CERROR("LProcFS failed in lquota-init\n");
 925                 rc = PTR_ERR(lquota_type_proc_dir);
 926                 return rc;
 927         }
 928
 929         rc = qunit_cache_init();
 930         if (rc)
 931                 return rc;
 932
 933         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 934         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 935 #endif
 936         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
 937         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
 938         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
 939         return 0;
 940 }
 941
 942 static void /*__exit*/ exit_lustre_quota(void)
 943 {
 944         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
 945         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
 946         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
 947 #ifdef HAVE_QUOTA_SUPPORT
 948         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 949         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 950
 951         qunit_cache_cleanup();
 952
 953         if (lquota_type_proc_dir)
 954                 lprocfs_remove(&lquota_type_proc_dir);
 955 #endif
 956 }
 957
 958 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 959 MODULE_DESCRIPTION("Lustre Quota");
 960 MODULE_LICENSE("GPL");
 961
 962 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
 963
 964 #ifdef HAVE_QUOTA_SUPPORT
 965 EXPORT_SYMBOL(mds_quota_interface);
 966 EXPORT_SYMBOL(filter_quota_interface);
 967 #endif
 968 EXPORT_SYMBOL(mdc_quota_interface);
 969 EXPORT_SYMBOL(osc_quota_interface);
 970 EXPORT_SYMBOL(lov_quota_interface);
 971 #endif /* __KERNEL */