lustre/quota/quota_interface.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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 cfs_spinlock_t last_print_lock = CFS_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         cfs_init_rwsem(&obt->obt_rwsem);
  80         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  81         cfs_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         cfs_spin_lock(&qctxt->lqc_lock);
 107         if (qctxt->lqc_import != NULL) {
 108                 cfs_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                         CERROR("%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                 cfs_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         cfs_spin_lock(&qctxt->lqc_lock);
 150         if (qctxt->lqc_import == imp) {
 151                 qctxt->lqc_import = NULL;
 152                 cfs_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                 cfs_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                 RETURN(-ENOMEM);
 195
 196         /* set over quota flags for a uid/gid */
 197         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 198         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 199
 200         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 201                 struct lustre_qunit_size *lqs = NULL;
 202
 203                 lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)),
 204                                        qctxt, 0);
 205                 if (IS_ERR(lqs)) {
 206                         rc = PTR_ERR(lqs);
 207                         CDEBUG(D_QUOTA, "search lqs for %s %d failed, "
 208                                "(rc = %d)\n",
 209                                cnt == USRQUOTA ? "user" : "group",
 210                                GET_OA_ID(cnt, oa), rc);
 211                         break;
 212                 } else if (lqs == NULL) {
 213                         /* continue to check group quota if the file's owner
 214                          * doesn't have quota limit. LU-530 */
 215                         continue;
 216                 } else {
 217                         cfs_spin_lock(&lqs->lqs_lock);
 218                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 219                                 oa->o_flags |= (cnt == USRQUOTA) ?
 220                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 221                                 cfs_spin_unlock(&lqs->lqs_lock);
 222                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 223                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 224                                        qctxt->lqc_sync_blk);
 225                                 /* this is for quota_search_lqs */
 226                                 lqs_putref(lqs);
 227                                 continue;
 228                         }
 229                         cfs_spin_unlock(&lqs->lqs_lock);
 230                         /* this is for quota_search_lqs */
 231                         lqs_putref(lqs);
 232                 }
 233
 234                 memset(oqctl, 0, sizeof(*oqctl));
 235
 236                 oqctl->qc_cmd = Q_GETQUOTA;
 237                 oqctl->qc_type = cnt;
 238                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 239                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 240                 if (err) {
 241                         if (!rc)
 242                                 rc = err;
 243                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 244                                                              OBD_MD_FLGRPQUOTA);
 245                         CDEBUG(D_QUOTA, "fsfilt getquota for %s %d failed, "
 246                                "(rc = %d)\n",
 247                                cnt == USRQUOTA ? "user" : "group",
 248                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid, err);
 249                         continue;
 250                 }
 251
 252                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 253                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 254                     oqctl->qc_dqblk.dqb_bhardlimit)) {
 255                         oa->o_flags |= (cnt == USRQUOTA) ?
 256                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 257                         CDEBUG(D_QUOTA, "out of quota for %s %d\n",
 258                                cnt == USRQUOTA ? "user" : "group",
 259                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid);
 260                 }
 261         }
 262         OBD_FREE_PTR(oqctl);
 263         RETURN(rc);
 264 }
 265
 266 /**
 267  * check whether the left quota of certain uid and gid can satisfy a block_write
 268  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 269  */
 270 static int quota_check_common(struct obd_device *obd, const unsigned int id[],
 271                               int pending[], int count, int cycle, int isblk,
 272                               struct inode *inode, int frags)
 273 {
 274         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 275         int i;
 276         struct qunit_data qdata[MAXQUOTAS];
 277         int mb = 0;
 278         int rc = 0, rc2[2] = { 0, 0 };
 279         ENTRY;
 280
 281         cfs_spin_lock(&qctxt->lqc_lock);
 282         if (!qctxt->lqc_valid){
 283                 cfs_spin_unlock(&qctxt->lqc_lock);
 284                 RETURN(rc);
 285         }
 286         cfs_spin_unlock(&qctxt->lqc_lock);
 287
 288         for (i = 0; i < MAXQUOTAS; i++) {
 289                 struct lustre_qunit_size *lqs = NULL;
 290
 291                 qdata[i].qd_id = id[i];
 292                 qdata[i].qd_flags = i;
 293                 if (isblk)
 294                         QDATA_SET_BLK(&qdata[i]);
 295                 qdata[i].qd_count = 0;
 296
 297                 /* ignore root user */
 298                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 299                         continue;
 300
 301                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 302                 if (lqs == NULL || IS_ERR(lqs))
 303                         continue;
 304
 305                 if (IS_ERR(lqs)) {
 306                         CERROR("can not find lqs for check_common: "
 307                                "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n",
 308                                id[i], i % 2 ? 'g': 'u', isblk, count,
 309                                PTR_ERR(lqs));
 310                         RETURN(PTR_ERR(lqs));
 311                 }
 312
 313                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 314                 cfs_spin_lock(&lqs->lqs_lock);
 315                 if (!cycle) {
 316                         if (isblk) {
 317                                 pending[i] = count * CFS_PAGE_SIZE;
 318                                 /* in order to complete this write, we need extra
 319                                  * meta blocks. This function can get it through
 320                                  * data needed to be written b=16542 */
 321                                 if (inode) {
 322                                         mb = pending[i];
 323                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 324                                                              &mb, inode,
 325                                                              frags);
 326                                         if (rc)
 327                                                 CERROR("%s: can't get extra "
 328                                                        "meta blocks\n",
 329                                                        obd->obd_name);
 330                                         else
 331                                                 pending[i] += mb;
 332                                 }
 333                                 LASSERTF(pending[i] >= 0, "pending is not valid"
 334                                          ", count=%d, mb=%d\n", count, mb);
 335                                 lqs->lqs_bwrite_pending += pending[i];
 336                         } else {
 337                                 pending[i] = count;
 338                                 lqs->lqs_iwrite_pending += pending[i];
 339                         }
 340                 }
 341
 342                 /* if xx_rec < 0, that means quota are releasing,
 343                  * and it may return before we use quota. So if
 344                  * we find this situation, we assuming it has
 345                  * returned b=18491 */
 346                 if (isblk && lqs->lqs_blk_rec < 0) {
 347                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 348                                 qdata[i].qd_count = 0;
 349                         else
 350                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 351                 }
 352                 if (!isblk && lqs->lqs_ino_rec < 0) {
 353                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 354                                 qdata[i].qd_count = 0;
 355                         else
 356                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 357                 }
 358
 359                 CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]"
 360                        " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]"
 361                        " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count,
 362                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 363                        qdata[i].qd_count, mb, pending[i]);
 364                 if (rc2[i] == QUOTA_RET_OK) {
 365                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 366                                 rc2[i] = QUOTA_RET_ACQUOTA;
 367                         if (!isblk && qdata[i].qd_count <
 368                             lqs->lqs_iwrite_pending)
 369                                 rc2[i] = QUOTA_RET_ACQUOTA;
 370                 }
 371
 372                 cfs_spin_unlock(&lqs->lqs_lock);
 373
 374                 if (lqs->lqs_blk_rec  < 0 &&
 375                     qdata[i].qd_count <
 376                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 377                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 378
 379                 /* When cycle is zero, lqs_*_pending will be changed. We will
 380                  * get reference of the lqs here and put reference of lqs in
 381                  * quota_pending_commit b=14784 */
 382                 if (!cycle)
 383                         lqs_getref(lqs);
 384
 385                 /* this is for quota_search_lqs */
 386                 lqs_putref(lqs);
 387         }
 388
 389         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 390                 RETURN(QUOTA_RET_ACQUOTA);
 391         else
 392                 RETURN(rc);
 393 }
 394
 395 int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag)
 396 {
 397         struct lustre_qunit_size *lqs;
 398         int i, q_set = 0;
 399
 400         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 401                 RETURN(0);
 402
 403         for (i = 0; i < MAXQUOTAS; i++) {
 404                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 405                                        &obd->u.obt.obt_qctxt, 0);
 406                 if (lqs && !IS_ERR(lqs)) {
 407                         if (lqs->lqs_flags & flag)
 408                                 q_set = 1;
 409                         lqs_putref(lqs);
 410                 }
 411         }
 412
 413         return q_set;
 414 }
 415
 416 static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
 417                                 const unsigned int id[], int pending[],
 418                                 int count, quota_acquire acquire,
 419                                 struct obd_trans_info *oti, int isblk,
 420                                 struct inode *inode, int frags)
 421 {
 422         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 423         struct timeval work_start;
 424         struct timeval work_end;
 425         long timediff;
 426         struct l_wait_info lwi = { 0 };
 427         int rc = 0, cycle = 0, count_err = 1;
 428         ENTRY;
 429
 430         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 431                 RETURN(0);
 432
 433         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 434                 /* If the client has been evicted or if it
 435                  * timed out and tried to reconnect already,
 436                  * abort the request immediately */
 437                 RETURN(-ENOTCONN);
 438
 439         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 440         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 441         /* Unfortunately, if quota master is too busy to handle the
 442          * pre-dqacq in time and quota hash on ost is used up, we
 443          * have to wait for the completion of in flight dqacq/dqrel,
 444          * in order to get enough quota for write b=12588 */
 445         cfs_gettimeofday(&work_start);
 446         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 447                                         inode, frags)) &
 448                QUOTA_RET_ACQUOTA) {
 449
 450                 cfs_spin_lock(&qctxt->lqc_lock);
 451                 if (!qctxt->lqc_import && oti) {
 452                         cfs_spin_unlock(&qctxt->lqc_lock);
 453                         LASSERT(oti->oti_thread);
 454                         /* The recovery thread doesn't have watchdog
 455                          * attached. LU-369 */
 456                         if (oti->oti_thread->t_watchdog)
 457                                 lc_watchdog_disable(oti->oti_thread->\
 458                                                 t_watchdog);
 459                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 460                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 461                                      &lwi);
 462                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 463                         if (oti->oti_thread->t_watchdog)
 464                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 465                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 466                 } else {
 467                         cfs_spin_unlock(&qctxt->lqc_lock);
 468                 }
 469
 470                 cycle++;
 471                 if (isblk)
 472                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 473                 /* after acquire(), we should run quota_check_common again
 474                  * so that we confirm there are enough quota to finish write */
 475                 rc = acquire(obd, id, oti, isblk);
 476
 477                 /* please reference to dqacq_completion for the below */
 478                 /* a new request is finished, try again */
 479                 if (rc == QUOTA_REQ_RETURNED) {
 480                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 481                         continue;
 482                 }
 483
 484                 /* it is out of quota already */
 485                 if (rc == -EDQUOT) {
 486                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 487                         break;
 488                 }
 489
 490                 /* Related quota has been disabled by master, but enabled by
 491                  * slave, do not try again. */
 492                 if (unlikely(rc == -ESRCH)) {
 493                         CERROR("mismatched quota configuration, stop try.\n");
 494                         break;
 495                 }
 496
 497                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 498                         /* The client has been evicted or tried to
 499                          * to reconnect already, abort the request */
 500                         RETURN(-ENOTCONN);
 501
 502                 /* -EBUSY and others, wait a second and try again */
 503                 if (rc < 0) {
 504                         cfs_waitq_t        waitq;
 505                         struct l_wait_info lwi;
 506
 507                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 508                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 509                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 510                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 511                                count_err++);
 512
 513                         cfs_waitq_init(&waitq);
 514                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 515                                           NULL);
 516                         l_wait_event(waitq, 0, &lwi);
 517                 }
 518
 519                 if (rc < 0 || cycle % 10 == 0) {
 520                         cfs_spin_lock(&last_print_lock);
 521                         if (last_print == 0 ||
 522                             cfs_time_before((last_print + cfs_time_seconds(30)),
 523                                             cfs_time_current())) {
 524                                 last_print = cfs_time_current();
 525                                 cfs_spin_unlock(&last_print_lock);
 526                                 CWARN("still haven't managed to acquire quota "
 527                                       "space from the quota master after %d "
 528                                       "retries (err=%d, rc=%d)\n",
 529                                       cycle, count_err - 1, rc);
 530                         } else {
 531                                 cfs_spin_unlock(&last_print_lock);
 532                         }
 533                 }
 534
 535                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 536                        cycle);
 537         }
 538         cfs_gettimeofday(&work_end);
 539         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 540         lprocfs_counter_add(qctxt->lqc_stats,
 541                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 542                                     LQUOTA_WAIT_FOR_CHK_INO,
 543                             timediff);
 544
 545         if (rc > 0)
 546                 rc = 0;
 547         RETURN(rc);
 548 }
 549
 550 /**
 551  * when a block_write or inode_create rpc is finished, adjust the record for
 552  * pending blocks and inodes
 553  */
 554 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 555                                 int pending[], int isblk)
 556 {
 557         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 558         struct timeval work_start;
 559         struct timeval work_end;
 560         long timediff;
 561         int i;
 562         struct qunit_data qdata[MAXQUOTAS];
 563         ENTRY;
 564
 565         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 566         CLASSERT(MAXQUOTAS < 4);
 567         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 568                 RETURN(0);
 569
 570         cfs_gettimeofday(&work_start);
 571         for (i = 0; i < MAXQUOTAS; i++) {
 572                 struct lustre_qunit_size *lqs = NULL;
 573
 574                 LASSERT(pending[i] >= 0);
 575                 if (pending[i] == 0)
 576                         continue;
 577
 578                 qdata[i].qd_id = id[i];
 579                 qdata[i].qd_flags = i;
 580                 if (isblk)
 581                         QDATA_SET_BLK(&qdata[i]);
 582                 qdata[i].qd_count = 0;
 583
 584                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 585                         continue;
 586
 587                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 588                 if (lqs == NULL || IS_ERR(lqs)) {
 589                         CERROR("can not find lqs for pending_commit: "
 590                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 591                                "maybe cause unexpected lqs refcount error!\n",
 592                                id[i], i ? 'g': 'u', pending[i], isblk,
 593                                lqs ? PTR_ERR(lqs) : -1);
 594                         continue;
 595                 }
 596
 597                 cfs_spin_lock(&lqs->lqs_lock);
 598                 if (isblk) {
 599                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 600                                  "there are too many blocks! [id %u] [%c] "
 601                                  "[bwrite_pending %lu] [pending %u]\n",
 602                                  id[i], i % 2 ? 'g' : 'u',
 603                                  lqs->lqs_bwrite_pending, pending[i]);
 604
 605                         lqs->lqs_bwrite_pending -= pending[i];
 606                 } else {
 607                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 608                                 "there are too many files! [id %u] [%c] "
 609                                 "[iwrite_pending %lu] [pending %u]\n",
 610                                 id[i], i % 2 ? 'g' : 'u',
 611                                 lqs->lqs_iwrite_pending, pending[i]);
 612
 613                         lqs->lqs_iwrite_pending -= pending[i];
 614                 }
 615                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 616                        obd->obd_name,
 617                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 618                        i, pending[i], isblk);
 619                 cfs_spin_unlock(&lqs->lqs_lock);
 620
 621                 /* for quota_search_lqs in pending_commit */
 622                 lqs_putref(lqs);
 623                 /* for quota_search_lqs in quota_check */
 624                 lqs_putref(lqs);
 625         }
 626         cfs_gettimeofday(&work_end);
 627         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 628         lprocfs_counter_add(qctxt->lqc_stats,
 629                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 630                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 631                             timediff);
 632
 633         RETURN(0);
 634 }
 635
 636 static int mds_quota_init(void)
 637 {
 638         return lustre_dquot_init();
 639 }
 640
 641 static int mds_quota_exit(void)
 642 {
 643         lustre_dquot_exit();
 644         return 0;
 645 }
 646
 647 static int mds_quota_setup(struct obd_device *obd)
 648 {
 649         struct obd_device_target *obt = &obd->u.obt;
 650         struct mds_obd *mds = &obd->u.mds;
 651         int rc;
 652         ENTRY;
 653
 654         if (unlikely(mds->mds_quota)) {
 655                 CWARN("try to reinitialize quota context!\n");
 656                 RETURN(0);
 657         }
 658
 659         cfs_init_rwsem(&obt->obt_rwsem);
 660         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 661         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 662         cfs_sema_init(&obt->obt_quotachecking, 1);
 663         /* initialize quota master and quota context */
 664         cfs_init_rwsem(&mds->mds_qonoff_sem);
 665         rc = qctxt_init(obd, dqacq_handler);
 666         if (rc) {
 667                 CERROR("%s: initialize quota context failed! (rc:%d)\n",
 668                        obd->obd_name, rc);
 669                 RETURN(rc);
 670         }
 671         mds->mds_quota = 1;
 672         RETURN(rc);
 673 }
 674
 675 static int mds_quota_cleanup(struct obd_device *obd)
 676 {
 677         ENTRY;
 678         if (unlikely(!obd->u.mds.mds_quota))
 679                 RETURN(0);
 680
 681         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 682         RETURN(0);
 683 }
 684
 685 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 686 {
 687         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 688         ENTRY;
 689
 690         if (unlikely(!obd->u.mds.mds_quota))
 691                 RETURN(0);
 692
 693         if (data != NULL)
 694                 QUOTA_MASTER_READY(qctxt);
 695         else
 696                 QUOTA_MASTER_UNREADY(qctxt);
 697         RETURN(0);
 698 }
 699
 700 static int mds_quota_fs_cleanup(struct obd_device *obd)
 701 {
 702         struct mds_obd *mds = &obd->u.mds;
 703         struct obd_quotactl oqctl;
 704         ENTRY;
 705
 706         if (unlikely(!mds->mds_quota))
 707                 RETURN(0);
 708
 709         mds->mds_quota = 0;
 710         memset(&oqctl, 0, sizeof(oqctl));
 711         oqctl.qc_type = UGQUOTA;
 712
 713         cfs_down_write(&mds->mds_qonoff_sem);
 714         mds_admin_quota_off(obd, &oqctl);
 715         cfs_up_write(&mds->mds_qonoff_sem);
 716         RETURN(0);
 717 }
 718
 719 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 720                                 struct obd_trans_info *oti, int isblk)
 721 {
 722         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 723         int rc;
 724         ENTRY;
 725
 726         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 727         RETURN(rc);
 728 }
 729
 730 #endif /* HAVE_QUOTA_SUPPORT */
 731 #endif /* __KERNEL__ */
 732
 733 #ifdef __KERNEL__
 734 #ifdef HAVE_QUOTA_SUPPORT
 735 quota_interface_t mds_quota_interface = {
 736         .quota_init     = mds_quota_init,
 737         .quota_exit     = mds_quota_exit,
 738         .quota_setup    = mds_quota_setup,
 739         .quota_cleanup  = mds_quota_cleanup,
 740         .quota_check    = target_quota_check,
 741         .quota_ctl      = mds_quota_ctl,
 742         .quota_setinfo  = mds_quota_setinfo,
 743         .quota_fs_cleanup = mds_quota_fs_cleanup,
 744         .quota_recovery = mds_quota_recovery,
 745         .quota_adjust   = mds_quota_adjust,
 746         .quota_chkquota = quota_chk_acq_common,
 747         .quota_acquire  = quota_acquire_common,
 748         .quota_pending_commit = quota_pending_commit,
 749 };
 750
 751 quota_interface_t filter_quota_interface = {
 752         .quota_setup    = filter_quota_setup,
 753         .quota_cleanup  = filter_quota_cleanup,
 754         .quota_check    = target_quota_check,
 755         .quota_ctl      = filter_quota_ctl,
 756         .quota_setinfo  = filter_quota_setinfo,
 757         .quota_clearinfo = filter_quota_clearinfo,
 758         .quota_enforce  = filter_quota_enforce,
 759         .quota_getflag  = filter_quota_getflag,
 760         .quota_acquire  = quota_acquire_common,
 761         .quota_adjust   = filter_quota_adjust,
 762         .quota_chkquota = quota_chk_acq_common,
 763         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 764         .quota_pending_commit = quota_pending_commit,
 765 };
 766 #endif
 767 #endif /* __KERNEL__ */
 768
 769 #ifdef __KERNEL__
 770
 771 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 772
 773 static int __init init_lustre_quota(void)
 774 {
 775 #ifdef HAVE_QUOTA_SUPPORT
 776         int rc = 0;
 777
 778         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 779                                                 proc_lustre_root,
 780                                                 NULL, NULL);
 781         if (IS_ERR(lquota_type_proc_dir)) {
 782                 CERROR("LProcFS failed in lquota-init\n");
 783                 rc = PTR_ERR(lquota_type_proc_dir);
 784                 return rc;
 785         }
 786
 787         rc = qunit_cache_init();
 788         if (rc)
 789                 return rc;
 790
 791         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 792         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 793 #endif
 794         return 0;
 795 }
 796
 797 static void /*__exit*/ exit_lustre_quota(void)
 798 {
 799 #ifdef HAVE_QUOTA_SUPPORT
 800         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 801         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 802
 803         qunit_cache_cleanup();
 804
 805         if (lquota_type_proc_dir)
 806                 lprocfs_remove(&lquota_type_proc_dir);
 807 #endif
 808 }
 809
 810 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 811 MODULE_DESCRIPTION("Lustre Quota");
 812 MODULE_LICENSE("GPL");
 813
 814 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
 815
 816 #ifdef HAVE_QUOTA_SUPPORT
 817 EXPORT_SYMBOL(mds_quota_interface);
 818 EXPORT_SYMBOL(filter_quota_interface);
 819 #endif
 820 #endif /* __KERNEL */