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         sema_init(&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 = exp->exp_imp_reverse;
 101         ENTRY;
 102
 103         LASSERT(imp != NULL);
 104
 105         /* setup the quota context import */
 106         spin_lock(&qctxt->lqc_lock);
 107         if (qctxt->lqc_import != NULL) {
 108                 spin_unlock(&qctxt->lqc_lock);
 109                 if (qctxt->lqc_import == imp)
 110                         CDEBUG(D_WARNING, "%s: lqc_import(%p) of obd(%p) was "
 111                                "activated already.\n", obd->obd_name, imp, obd);
 112                 else
 113                         CDEBUG(D_ERROR, "%s: lqc_import(%p:%p) of obd(%p) was "
 114                                "activated by others.\n", obd->obd_name,
 115                                qctxt->lqc_import, imp, obd);
 116         } else {
 117                 qctxt->lqc_import = imp;
 118                 /* make imp's connect flags equal relative exp's connect flags
 119                  * adding it to avoid the scan export list */
 120                 imp->imp_connect_data.ocd_connect_flags |=
 121                                 (exp->exp_connect_flags &
 122                                  (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 123                 spin_unlock(&qctxt->lqc_lock);
 124                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated "
 125                        "now.\n", obd->obd_name, imp, obd);
 126
 127                 cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 128                 /* start quota slave recovery thread. (release high limits) */
 129                 qslave_start_recovery(obd, qctxt);
 130         }
 131         RETURN(0);
 132 }
 133
 134 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 135 {
 136         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 137         struct obd_import *imp = exp->exp_imp_reverse;
 138         ENTRY;
 139
 140         /* lquota may be not set up before destroying export, b=14896 */
 141         if (!obd->obd_set_up)
 142                 RETURN(0);
 143
 144         if (unlikely(imp == NULL))
 145                 RETURN(0);
 146
 147         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 148          * should be invalid b=12374 */
 149         spin_lock(&qctxt->lqc_lock);
 150         if (qctxt->lqc_import == imp) {
 151                 qctxt->lqc_import = NULL;
 152                 spin_unlock(&qctxt->lqc_lock);
 153                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is invalid now.\n",
 154                        obd->obd_name, imp, obd);
 155                 ptlrpc_cleanup_imp(imp);
 156                 dqacq_interrupt(qctxt);
 157         } else {
 158                 spin_unlock(&qctxt->lqc_lock);
 159         }
 160         RETURN(0);
 161 }
 162
 163 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 164 {
 165         ENTRY;
 166
 167         if (!ll_sb_any_quota_active(obd->u.obt.obt_sb))
 168                 RETURN(0);
 169
 170         if (ignore) {
 171                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 172                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 173         } else {
 174                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 175         }
 176
 177         RETURN(0);
 178 }
 179
 180 #define GET_OA_ID(flag, oa) (flag == USRQUOTA ? oa->o_uid : oa->o_gid)
 181 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 182 {
 183         struct obd_device_target *obt = &obd->u.obt;
 184         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 185         int err, cnt, rc = 0;
 186         struct obd_quotactl *oqctl;
 187         ENTRY;
 188
 189         if (!ll_sb_any_quota_active(obt->obt_sb))
 190                 RETURN(0);
 191
 192         OBD_ALLOC_PTR(oqctl);
 193         if (!oqctl) {
 194                 CERROR("Not enough memory!");
 195                 RETURN(-ENOMEM);
 196         }
 197
 198         /* set over quota flags for a uid/gid */
 199         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 200         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 201
 202         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 203                 struct lustre_qunit_size *lqs = NULL;
 204
 205                 lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)),
 206                                        qctxt, 0);
 207                 if (lqs == NULL || IS_ERR(lqs)) {
 208                         rc = PTR_ERR(lqs);
 209                         break;
 210                 } else {
 211                         spin_lock(&lqs->lqs_lock);
 212                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 213                                 oa->o_flags |= (cnt == USRQUOTA) ?
 214                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 215                                 spin_unlock(&lqs->lqs_lock);
 216                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 217                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 218                                        qctxt->lqc_sync_blk);
 219                                 /* this is for quota_search_lqs */
 220                                 lqs_putref(lqs);
 221                                 continue;
 222                         }
 223                         spin_unlock(&lqs->lqs_lock);
 224                         /* this is for quota_search_lqs */
 225                         lqs_putref(lqs);
 226                 }
 227
 228                 memset(oqctl, 0, sizeof(*oqctl));
 229
 230                 oqctl->qc_cmd = Q_GETQUOTA;
 231                 oqctl->qc_type = cnt;
 232                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 233                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 234                 if (err) {
 235                         if (!rc)
 236                                 rc = err;
 237                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 238                                                              OBD_MD_FLGRPQUOTA);
 239                         continue;
 240                 }
 241
 242                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 243                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 244                     oqctl->qc_dqblk.dqb_bhardlimit))
 245                         oa->o_flags |= (cnt == USRQUOTA) ?
 246                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 247         }
 248         OBD_FREE_PTR(oqctl);
 249         RETURN(rc);
 250 }
 251
 252 /**
 253  * check whether the left quota of certain uid and gid can satisfy a block_write
 254  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 255  */
 256 static int quota_check_common(struct obd_device *obd, const unsigned int id[],
 257                               int pending[], int count, int cycle, int isblk,
 258                               struct inode *inode, int frags)
 259 {
 260         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 261         int i;
 262         struct qunit_data qdata[MAXQUOTAS];
 263         int mb = 0;
 264         int rc = 0, rc2[2] = { 0, 0 };
 265         ENTRY;
 266
 267         spin_lock(&qctxt->lqc_lock);
 268         if (!qctxt->lqc_valid){
 269                 spin_unlock(&qctxt->lqc_lock);
 270                 RETURN(rc);
 271         }
 272         spin_unlock(&qctxt->lqc_lock);
 273
 274         for (i = 0; i < MAXQUOTAS; i++) {
 275                 struct lustre_qunit_size *lqs = NULL;
 276
 277                 qdata[i].qd_id = id[i];
 278                 qdata[i].qd_flags = i;
 279                 if (isblk)
 280                         QDATA_SET_BLK(&qdata[i]);
 281                 qdata[i].qd_count = 0;
 282
 283                 /* ignore root user */
 284                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 285                         continue;
 286
 287                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 288                 if (lqs == NULL || IS_ERR(lqs))
 289                         continue;
 290
 291                 if (IS_ERR(lqs)) {
 292                         CERROR("can not find lqs for check_common: "
 293                                "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n",
 294                                id[i], i % 2 ? 'g': 'u', isblk, count,
 295                                PTR_ERR(lqs));
 296                         RETURN(PTR_ERR(lqs));
 297                 }
 298
 299                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 300                 spin_lock(&lqs->lqs_lock);
 301                 if (!cycle) {
 302                         if (isblk) {
 303                                 pending[i] = count * CFS_PAGE_SIZE;
 304                                 /* in order to complete this write, we need extra
 305                                  * meta blocks. This function can get it through
 306                                  * data needed to be written b=16542 */
 307                                 if (inode) {
 308                                         mb = pending[i];
 309                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 310                                                              &mb, inode,frags);
 311                                         if (rc)
 312                                                 CDEBUG(D_ERROR,
 313                                                        "can't get extra "
 314                                                        "meta blocks.\n");
 315                                         else
 316                                                 pending[i] += mb;
 317                                 }
 318                                 lqs->lqs_bwrite_pending += pending[i];
 319                         } else {
 320                                 pending[i] = count;
 321                                 lqs->lqs_iwrite_pending += pending[i];
 322                         }
 323                 }
 324
 325                 /* if xx_rec < 0, that means quota are releasing,
 326                  * and it may return before we use quota. So if
 327                  * we find this situation, we assuming it has
 328                  * returned b=18491 */
 329                 if (isblk && lqs->lqs_blk_rec < 0) {
 330                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 331                                 qdata[i].qd_count = 0;
 332                         else
 333                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 334                 }
 335                 if (!isblk && lqs->lqs_ino_rec < 0) {
 336                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 337                                 qdata[i].qd_count = 0;
 338                         else
 339                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 340                 }
 341
 342                 CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]"
 343                        " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]"
 344                        " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count,
 345                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 346                        qdata[i].qd_count, mb, pending[i]);
 347                 if (rc2[i] == QUOTA_RET_OK) {
 348                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 349                                 rc2[i] = QUOTA_RET_ACQUOTA;
 350                         if (!isblk && qdata[i].qd_count <
 351                             lqs->lqs_iwrite_pending)
 352                                 rc2[i] = QUOTA_RET_ACQUOTA;
 353                 }
 354
 355                 spin_unlock(&lqs->lqs_lock);
 356
 357                 if (lqs->lqs_blk_rec  < 0 &&
 358                     qdata[i].qd_count <
 359                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 360                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 361
 362                 /* When cycle is zero, lqs_*_pending will be changed. We will
 363                  * get reference of the lqs here and put reference of lqs in
 364                  * quota_pending_commit b=14784 */
 365                 if (!cycle)
 366                         lqs_getref(lqs);
 367
 368                 /* this is for quota_search_lqs */
 369                 lqs_putref(lqs);
 370         }
 371
 372         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 373                 RETURN(QUOTA_RET_ACQUOTA);
 374         else
 375                 RETURN(rc);
 376 }
 377
 378 int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag)
 379 {
 380         struct lustre_qunit_size *lqs;
 381         int i, q_set = 0;
 382
 383         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 384                 RETURN(0);
 385
 386         for (i = 0; i < MAXQUOTAS; i++) {
 387                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 388                                        &obd->u.obt.obt_qctxt, 0);
 389                 if (lqs && !IS_ERR(lqs)) {
 390                         if (lqs->lqs_flags & flag)
 391                                 q_set = 1;
 392                         lqs_putref(lqs);
 393                 }
 394         }
 395
 396         return q_set;
 397 }
 398
 399 static int quota_chk_acq_common(struct obd_device *obd, const unsigned int id[],
 400                                 int pending[], int count, quota_acquire acquire,
 401                                 struct obd_trans_info *oti, int isblk,
 402                                 struct inode *inode, int frags)
 403 {
 404         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 405         struct timeval work_start;
 406         struct timeval work_end;
 407         long timediff;
 408         struct l_wait_info lwi = { 0 };
 409         int rc = 0, cycle = 0, count_err = 1;
 410         ENTRY;
 411
 412         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 413                 RETURN(0);
 414
 415         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 416         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 417         /* Unfortunately, if quota master is too busy to handle the
 418          * pre-dqacq in time and quota hash on ost is used up, we
 419          * have to wait for the completion of in flight dqacq/dqrel,
 420          * in order to get enough quota for write b=12588 */
 421         do_gettimeofday(&work_start);
 422         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 423                                         inode, frags)) &
 424                QUOTA_RET_ACQUOTA) {
 425
 426                 spin_lock(&qctxt->lqc_lock);
 427                 if (!qctxt->lqc_import && oti) {
 428                         spin_unlock(&qctxt->lqc_lock);
 429
 430                         LASSERT(oti && oti->oti_thread &&
 431                                 oti->oti_thread->t_watchdog);
 432
 433                         lc_watchdog_disable(oti->oti_thread->t_watchdog);
 434                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 435                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 436                                      &lwi);
 437                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 438                         lc_watchdog_touch(oti->oti_thread->t_watchdog,
 439                                  GET_TIMEOUT(oti->oti_thread->t_svc));
 440                 } else {
 441                         spin_unlock(&qctxt->lqc_lock);
 442                 }
 443
 444                 cycle++;
 445                 if (isblk)
 446                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 447                 /* after acquire(), we should run quota_check_common again
 448                  * so that we confirm there are enough quota to finish write */
 449                 rc = acquire(obd, id, oti, isblk);
 450
 451                 /* please reference to dqacq_completion for the below */
 452                 /* a new request is finished, try again */
 453                 if (rc == QUOTA_REQ_RETURNED) {
 454                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 455                         continue;
 456                 }
 457
 458                 /* it is out of quota already */
 459                 if (rc == -EDQUOT) {
 460                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 461                         break;
 462                 }
 463
 464                 /* Related quota has been disabled by master, but enabled by
 465                  * slave, do not try again. */
 466                 if (unlikely(rc == -ESRCH)) {
 467                         CERROR("mismatched quota configuration, stop try.\n");
 468                         break;
 469                 }
 470
 471                 /* -EBUSY and others, wait a second and try again */
 472                 if (rc < 0) {
 473                         cfs_waitq_t        waitq;
 474                         struct l_wait_info lwi;
 475
 476                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 477                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 478                                          GET_TIMEOUT(oti->oti_thread->t_svc));
 479                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 480                                count_err++);
 481
 482                         init_waitqueue_head(&waitq);
 483                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 484                                           NULL);
 485                         l_wait_event(waitq, 0, &lwi);
 486                 }
 487
 488                 if (rc < 0 || cycle % 10 == 0) {
 489                         spin_lock(&last_print_lock);
 490                         if (last_print == 0 ||
 491                             cfs_time_before((last_print + cfs_time_seconds(30)),
 492                                             cfs_time_current())) {
 493                                 last_print = cfs_time_current();
 494                                 spin_unlock(&last_print_lock);
 495                                 CWARN("still haven't managed to acquire quota "
 496                                       "space from the quota master after %d "
 497                                       "retries (err=%d, rc=%d)\n",
 498                                       cycle, count_err - 1, rc);
 499                         } else {
 500                                 spin_unlock(&last_print_lock);
 501                         }
 502                 }
 503
 504                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 505                        cycle);
 506         }
 507         do_gettimeofday(&work_end);
 508         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 509         lprocfs_counter_add(qctxt->lqc_stats,
 510                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 511                                     LQUOTA_WAIT_FOR_CHK_INO,
 512                             timediff);
 513
 514         RETURN(rc);
 515 }
 516
 517 /**
 518  * when a block_write or inode_create rpc is finished, adjust the record for
 519  * pending blocks and inodes
 520  */
 521 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 522                                 int pending[], int isblk)
 523 {
 524         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 525         struct timeval work_start;
 526         struct timeval work_end;
 527         long timediff;
 528         int i;
 529         struct qunit_data qdata[MAXQUOTAS];
 530         ENTRY;
 531
 532         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 533         CLASSERT(MAXQUOTAS < 4);
 534         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 535                 RETURN(0);
 536
 537         do_gettimeofday(&work_start);
 538         for (i = 0; i < MAXQUOTAS; i++) {
 539                 struct lustre_qunit_size *lqs = NULL;
 540
 541                 LASSERT(pending[i] >= 0);
 542                 if (pending[i] == 0)
 543                         continue;
 544
 545                 qdata[i].qd_id = id[i];
 546                 qdata[i].qd_flags = i;
 547                 if (isblk)
 548                         QDATA_SET_BLK(&qdata[i]);
 549                 qdata[i].qd_count = 0;
 550
 551                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 552                         continue;
 553
 554                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 555                 if (lqs == NULL || IS_ERR(lqs)) {
 556                         CERROR("can not find lqs for pending_commit: "
 557                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 558                                "maybe cause unexpected lqs refcount error!\n",
 559                                id[i], i ? 'g': 'u', pending[i], isblk,
 560                                lqs ? PTR_ERR(lqs) : -1);
 561                         continue;
 562                 }
 563
 564                 spin_lock(&lqs->lqs_lock);
 565                 if (isblk) {
 566                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 567                                  "there are too many blocks! [id %u] [%c] "
 568                                  "[bwrite_pending %lu] [pending %u]\n",
 569                                  id[i], i % 2 ? 'g' : 'u',
 570                                  lqs->lqs_bwrite_pending, pending[i]);
 571
 572                         lqs->lqs_bwrite_pending -= pending[i];
 573                 } else {
 574                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 575                                 "there are too many files! [id %u] [%c] "
 576                                 "[iwrite_pending %lu] [pending %u]\n",
 577                                 id[i], i % 2 ? 'g' : 'u',
 578                                 lqs->lqs_iwrite_pending, pending[i]);
 579
 580                         lqs->lqs_iwrite_pending -= pending[i];
 581                 }
 582                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 583                        obd->obd_name,
 584                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 585                        i, pending[i], isblk);
 586                 spin_unlock(&lqs->lqs_lock);
 587
 588                 /* for quota_search_lqs in pending_commit */
 589                 lqs_putref(lqs);
 590                 /* for quota_search_lqs in quota_check */
 591                 lqs_putref(lqs);
 592         }
 593         do_gettimeofday(&work_end);
 594         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 595         lprocfs_counter_add(qctxt->lqc_stats,
 596                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 597                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 598                             timediff);
 599
 600         RETURN(0);
 601 }
 602
 603 static int mds_quota_init(void)
 604 {
 605         return lustre_dquot_init();
 606 }
 607
 608 static int mds_quota_exit(void)
 609 {
 610         lustre_dquot_exit();
 611         return 0;
 612 }
 613
 614 static int mds_quota_setup(struct obd_device *obd)
 615 {
 616         struct obd_device_target *obt = &obd->u.obt;
 617         struct mds_obd *mds = &obd->u.mds;
 618         int rc;
 619         ENTRY;
 620
 621         if (unlikely(mds->mds_quota)) {
 622                 CWARN("try to reinitialize quota context!\n");
 623                 RETURN(0);
 624         }
 625
 626         init_rwsem(&obt->obt_rwsem);
 627         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 628         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 629         sema_init(&obt->obt_quotachecking, 1);
 630         /* initialize quota master and quota context */
 631         sema_init(&mds->mds_qonoff_sem, 1);
 632         rc = qctxt_init(obd, dqacq_handler);
 633         if (rc) {
 634                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 635                 RETURN(rc);
 636         }
 637         mds->mds_quota = 1;
 638         RETURN(rc);
 639 }
 640
 641 static int mds_quota_cleanup(struct obd_device *obd)
 642 {
 643         ENTRY;
 644         if (unlikely(!obd->u.mds.mds_quota))
 645                 RETURN(0);
 646
 647         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 648         RETURN(0);
 649 }
 650
 651 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 652 {
 653         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 654         ENTRY;
 655
 656         if (unlikely(!obd->u.mds.mds_quota))
 657                 RETURN(0);
 658
 659         if (data != NULL)
 660                 QUOTA_MASTER_READY(qctxt);
 661         else
 662                 QUOTA_MASTER_UNREADY(qctxt);
 663         RETURN(0);
 664 }
 665
 666 static int mds_quota_fs_cleanup(struct obd_device *obd)
 667 {
 668         struct mds_obd *mds = &obd->u.mds;
 669         struct obd_quotactl oqctl;
 670         ENTRY;
 671
 672         if (unlikely(!mds->mds_quota))
 673                 RETURN(0);
 674
 675         mds->mds_quota = 0;
 676         memset(&oqctl, 0, sizeof(oqctl));
 677         oqctl.qc_type = UGQUOTA;
 678
 679         down(&mds->mds_qonoff_sem);
 680         mds_admin_quota_off(obd, &oqctl);
 681         up(&mds->mds_qonoff_sem);
 682         RETURN(0);
 683 }
 684
 685 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 686                                 struct obd_trans_info *oti, int isblk)
 687 {
 688         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 689         int rc;
 690         ENTRY;
 691
 692         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 693         RETURN(rc);
 694 }
 695
 696 #endif /* HAVE_QUOTA_SUPPORT */
 697 #endif /* __KERNEL__ */
 698
 699 struct osc_quota_info {
 700         struct list_head        oqi_hash;       /* hash list */
 701         struct client_obd      *oqi_cli;        /* osc obd */
 702         unsigned int            oqi_id;         /* uid/gid of a file */
 703         short                   oqi_type;       /* quota type */
 704 };
 705
 706 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 707
 708 static struct list_head qinfo_hash[NR_DQHASH];
 709 /* SLAB cache for client quota context */
 710 cfs_mem_cache_t *qinfo_cachep = NULL;
 711
 712 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 713                          __attribute__((__const__));
 714
 715 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 716 {
 717         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 718         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 719         return tmp;
 720 }
 721
 722 /* caller must hold qinfo_list_lock */
 723 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 724 {
 725         struct list_head *head = qinfo_hash +
 726                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 727
 728         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 729         list_add(&oqi->oqi_hash, head);
 730 }
 731
 732 /* caller must hold qinfo_list_lock */
 733 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 734 {
 735         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 736         list_del_init(&oqi->oqi_hash);
 737 }
 738
 739 /* caller must hold qinfo_list_lock */
 740 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 741                                                 unsigned int id, int type)
 742 {
 743         unsigned int hashent = hashfn(cli, id, type);
 744         struct osc_quota_info *oqi;
 745         ENTRY;
 746
 747         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 748         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 749                 if (oqi->oqi_cli == cli &&
 750                     oqi->oqi_id == id && oqi->oqi_type == type)
 751                         return oqi;
 752         }
 753         RETURN(NULL);
 754 }
 755
 756 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 757                                           unsigned int id, int type)
 758 {
 759         struct osc_quota_info *oqi;
 760         ENTRY;
 761
 762         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 763         if(!oqi)
 764                 RETURN(NULL);
 765
 766         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 767         oqi->oqi_cli = cli;
 768         oqi->oqi_id = id;
 769         oqi->oqi_type = type;
 770
 771         RETURN(oqi);
 772 }
 773
 774 static void free_qinfo(struct osc_quota_info *oqi)
 775 {
 776         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 777 }
 778
 779 int osc_quota_chkdq(struct client_obd *cli, const unsigned int qid[])
 780 {
 781         unsigned int id;
 782         int cnt, rc = QUOTA_OK;
 783         ENTRY;
 784
 785         spin_lock(&qinfo_list_lock);
 786         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 787                 struct osc_quota_info *oqi = NULL;
 788
 789                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 790                 oqi = find_qinfo(cli, id, cnt);
 791                 if (oqi) {
 792                         rc = NO_QUOTA;
 793                         break;
 794                 }
 795         }
 796         spin_unlock(&qinfo_list_lock);
 797
 798         RETURN(rc);
 799 }
 800
 801 int osc_quota_setdq(struct client_obd *cli, const unsigned int qid[],
 802                     obd_flag valid, obd_flag flags)
 803 {
 804         unsigned int id;
 805         obd_flag noquota;
 806         int cnt, rc = 0;
 807         ENTRY;
 808
 809
 810         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 811                 struct osc_quota_info *oqi, *old;
 812
 813                 if (!(valid & ((cnt == USRQUOTA) ?
 814                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 815                         continue;
 816
 817                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 818                 noquota = (cnt == USRQUOTA) ?
 819                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 820
 821                 oqi = alloc_qinfo(cli, id, cnt);
 822                 if (oqi) {
 823                         spin_lock(&qinfo_list_lock);
 824
 825                         old = find_qinfo(cli, id, cnt);
 826                         if (old && !noquota)
 827                                 remove_qinfo_hash(old);
 828                         else if (!old && noquota)
 829                                 insert_qinfo_hash(oqi);
 830
 831                         spin_unlock(&qinfo_list_lock);
 832
 833                         if (old || !noquota)
 834                                 free_qinfo(oqi);
 835                         if (old && !noquota)
 836                                 free_qinfo(old);
 837                 } else {
 838                         CERROR("not enough mem!\n");
 839                         rc = -ENOMEM;
 840                         break;
 841                 }
 842         }
 843
 844         RETURN(rc);
 845 }
 846
 847 int osc_quota_cleanup(struct obd_device *obd)
 848 {
 849         struct client_obd *cli = &obd->u.cli;
 850         struct osc_quota_info *oqi, *n;
 851         int i;
 852         ENTRY;
 853
 854         spin_lock(&qinfo_list_lock);
 855         for (i = 0; i < NR_DQHASH; i++) {
 856                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 857                         if (oqi->oqi_cli != cli)
 858                                 continue;
 859                         remove_qinfo_hash(oqi);
 860                         free_qinfo(oqi);
 861                 }
 862         }
 863         spin_unlock(&qinfo_list_lock);
 864
 865         RETURN(0);
 866 }
 867
 868 int osc_quota_init(void)
 869 {
 870         int i;
 871         ENTRY;
 872
 873         LASSERT(qinfo_cachep == NULL);
 874         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 875                                             sizeof(struct osc_quota_info),
 876                                             0, 0);
 877         if (!qinfo_cachep)
 878                 RETURN(-ENOMEM);
 879
 880         for (i = 0; i < NR_DQHASH; i++)
 881                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 882
 883         RETURN(0);
 884 }
 885
 886 int osc_quota_exit(void)
 887 {
 888         struct osc_quota_info *oqi, *n;
 889         int i, rc;
 890         ENTRY;
 891
 892         spin_lock(&qinfo_list_lock);
 893         for (i = 0; i < NR_DQHASH; i++) {
 894                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 895                         remove_qinfo_hash(oqi);
 896                         free_qinfo(oqi);
 897                 }
 898         }
 899         spin_unlock(&qinfo_list_lock);
 900
 901         rc = cfs_mem_cache_destroy(qinfo_cachep);
 902         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 903         qinfo_cachep = NULL;
 904
 905         RETURN(0);
 906 }
 907
 908 #ifdef __KERNEL__
 909 #ifdef HAVE_QUOTA_SUPPORT
 910 quota_interface_t mds_quota_interface = {
 911         .quota_init     = mds_quota_init,
 912         .quota_exit     = mds_quota_exit,
 913         .quota_setup    = mds_quota_setup,
 914         .quota_cleanup  = mds_quota_cleanup,
 915         .quota_check    = target_quota_check,
 916         .quota_ctl      = mds_quota_ctl,
 917         .quota_setinfo  = mds_quota_setinfo,
 918         .quota_fs_cleanup = mds_quota_fs_cleanup,
 919         .quota_recovery = mds_quota_recovery,
 920         .quota_adjust   = mds_quota_adjust,
 921         .quota_chkquota = quota_chk_acq_common,
 922         .quota_acquire  = quota_acquire_common,
 923         .quota_pending_commit = quota_pending_commit,
 924 };
 925
 926 quota_interface_t filter_quota_interface = {
 927         .quota_setup    = filter_quota_setup,
 928         .quota_cleanup  = filter_quota_cleanup,
 929         .quota_check    = target_quota_check,
 930         .quota_ctl      = filter_quota_ctl,
 931         .quota_setinfo  = filter_quota_setinfo,
 932         .quota_clearinfo = filter_quota_clearinfo,
 933         .quota_enforce  = filter_quota_enforce,
 934         .quota_getflag  = filter_quota_getflag,
 935         .quota_acquire  = quota_acquire_common,
 936         .quota_adjust   = filter_quota_adjust,
 937         .quota_chkquota = quota_chk_acq_common,
 938         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 939         .quota_pending_commit = quota_pending_commit,
 940 };
 941 #endif
 942 #endif /* __KERNEL__ */
 943
 944 quota_interface_t mdc_quota_interface = {
 945         .quota_ctl      = client_quota_ctl,
 946         .quota_check    = client_quota_check,
 947         .quota_poll_check = client_quota_poll_check,
 948 };
 949
 950 quota_interface_t lmv_quota_interface = {
 951         .quota_ctl      = lmv_quota_ctl,
 952         .quota_check    = lmv_quota_check,
 953 };
 954
 955 quota_interface_t osc_quota_interface = {
 956         .quota_ctl      = client_quota_ctl,
 957         .quota_check    = client_quota_check,
 958         .quota_poll_check = client_quota_poll_check,
 959         .quota_init     = osc_quota_init,
 960         .quota_exit     = osc_quota_exit,
 961         .quota_chkdq    = osc_quota_chkdq,
 962         .quota_setdq    = osc_quota_setdq,
 963         .quota_cleanup  = osc_quota_cleanup,
 964         .quota_adjust_qunit = client_quota_adjust_qunit,
 965 };
 966
 967 quota_interface_t lov_quota_interface = {
 968         .quota_ctl      = lov_quota_ctl,
 969         .quota_check    = lov_quota_check,
 970         .quota_adjust_qunit = lov_quota_adjust_qunit,
 971 };
 972
 973 #ifdef __KERNEL__
 974
 975 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 976
 977 static int __init init_lustre_quota(void)
 978 {
 979 #ifdef HAVE_QUOTA_SUPPORT
 980         int rc = 0;
 981
 982         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 983                                                 proc_lustre_root,
 984                                                 NULL, NULL);
 985         if (IS_ERR(lquota_type_proc_dir)) {
 986                 CERROR("LProcFS failed in lquota-init\n");
 987                 rc = PTR_ERR(lquota_type_proc_dir);
 988                 return rc;
 989         }
 990
 991         rc = qunit_cache_init();
 992         if (rc)
 993                 return rc;
 994
 995         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 996         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 997 #endif
 998         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
 999         PORTAL_SYMBOL_REGISTER(lmv_quota_interface);
1000         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
1001         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
1002         return 0;
1003 }
1004
1005 static void /*__exit*/ exit_lustre_quota(void)
1006 {
1007         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
1008         PORTAL_SYMBOL_UNREGISTER(lmv_quota_interface);
1009         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
1010         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
1011 #ifdef HAVE_QUOTA_SUPPORT
1012         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
1013         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
1014
1015         qunit_cache_cleanup();
1016
1017         if (lquota_type_proc_dir)
1018                 lprocfs_remove(&lquota_type_proc_dir);
1019 #endif
1020 }
1021
1022 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1023 MODULE_DESCRIPTION("Lustre Quota");
1024 MODULE_LICENSE("GPL");
1025
1026 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
1027
1028 #ifdef HAVE_QUOTA_SUPPORT
1029 EXPORT_SYMBOL(mds_quota_interface);
1030 EXPORT_SYMBOL(filter_quota_interface);
1031 #endif
1032 EXPORT_SYMBOL(mdc_quota_interface);
1033 EXPORT_SYMBOL(lmv_quota_interface);
1034 EXPORT_SYMBOL(osc_quota_interface);
1035 EXPORT_SYMBOL(lov_quota_interface);
1036 #endif /* __KERNEL */