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, struct obd_export *exp,
 400                                 const unsigned int id[], int pending[],
 401                                 int count, quota_acquire acquire,
 402                                 struct obd_trans_info *oti, int isblk,
 403                                 struct inode *inode, int frags)
 404 {
 405         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 406         struct timeval work_start;
 407         struct timeval work_end;
 408         long timediff;
 409         struct l_wait_info lwi = { 0 };
 410         int rc = 0, cycle = 0, count_err = 1;
 411         ENTRY;
 412
 413         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 414                 RETURN(0);
 415
 416         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 417                 /* If the client has been evicted or if it
 418                  * timed out and tried to reconnect already,
 419                  * abort the request immediately */
 420                 RETURN(-ENOTCONN);
 421
 422         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 423         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 424         /* Unfortunately, if quota master is too busy to handle the
 425          * pre-dqacq in time and quota hash on ost is used up, we
 426          * have to wait for the completion of in flight dqacq/dqrel,
 427          * in order to get enough quota for write b=12588 */
 428         do_gettimeofday(&work_start);
 429         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 430                                         inode, frags)) &
 431                QUOTA_RET_ACQUOTA) {
 432
 433                 spin_lock(&qctxt->lqc_lock);
 434                 if (!qctxt->lqc_import && oti) {
 435                         spin_unlock(&qctxt->lqc_lock);
 436
 437                         LASSERT(oti && oti->oti_thread &&
 438                                 oti->oti_thread->t_watchdog);
 439
 440                         lc_watchdog_disable(oti->oti_thread->t_watchdog);
 441                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 442                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 443                                      &lwi);
 444                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 445                         lc_watchdog_touch(oti->oti_thread->t_watchdog,
 446                                  GET_TIMEOUT(oti->oti_thread->t_svc));
 447                 } else {
 448                         spin_unlock(&qctxt->lqc_lock);
 449                 }
 450
 451                 cycle++;
 452                 if (isblk)
 453                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 454                 /* after acquire(), we should run quota_check_common again
 455                  * so that we confirm there are enough quota to finish write */
 456                 rc = acquire(obd, id, oti, isblk);
 457
 458                 /* please reference to dqacq_completion for the below */
 459                 /* a new request is finished, try again */
 460                 if (rc == QUOTA_REQ_RETURNED) {
 461                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 462                         continue;
 463                 }
 464
 465                 /* it is out of quota already */
 466                 if (rc == -EDQUOT) {
 467                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 468                         break;
 469                 }
 470
 471                 /* Related quota has been disabled by master, but enabled by
 472                  * slave, do not try again. */
 473                 if (unlikely(rc == -ESRCH)) {
 474                         CERROR("mismatched quota configuration, stop try.\n");
 475                         break;
 476                 }
 477
 478                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 479                         /* The client has been evicted or tried to
 480                          * to reconnect already, abort the request */
 481                         RETURN(-ENOTCONN);
 482
 483                 /* -EBUSY and others, wait a second and try again */
 484                 if (rc < 0) {
 485                         cfs_waitq_t        waitq;
 486                         struct l_wait_info lwi;
 487
 488                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 489                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 490                                          GET_TIMEOUT(oti->oti_thread->t_svc));
 491                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 492                                count_err++);
 493
 494                         init_waitqueue_head(&waitq);
 495                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 496                                           NULL);
 497                         l_wait_event(waitq, 0, &lwi);
 498                 }
 499
 500                 if (rc < 0 || cycle % 10 == 0) {
 501                         spin_lock(&last_print_lock);
 502                         if (last_print == 0 ||
 503                             cfs_time_before((last_print + cfs_time_seconds(30)),
 504                                             cfs_time_current())) {
 505                                 last_print = cfs_time_current();
 506                                 spin_unlock(&last_print_lock);
 507                                 CWARN("still haven't managed to acquire quota "
 508                                       "space from the quota master after %d "
 509                                       "retries (err=%d, rc=%d)\n",
 510                                       cycle, count_err - 1, rc);
 511                         } else {
 512                                 spin_unlock(&last_print_lock);
 513                         }
 514                 }
 515
 516                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 517                        cycle);
 518         }
 519         do_gettimeofday(&work_end);
 520         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 521         lprocfs_counter_add(qctxt->lqc_stats,
 522                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 523                                     LQUOTA_WAIT_FOR_CHK_INO,
 524                             timediff);
 525
 526         if (rc > 0)
 527                 rc = 0;
 528         RETURN(rc);
 529 }
 530
 531 /**
 532  * when a block_write or inode_create rpc is finished, adjust the record for
 533  * pending blocks and inodes
 534  */
 535 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 536                                 int pending[], int isblk)
 537 {
 538         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 539         struct timeval work_start;
 540         struct timeval work_end;
 541         long timediff;
 542         int i;
 543         struct qunit_data qdata[MAXQUOTAS];
 544         ENTRY;
 545
 546         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 547         CLASSERT(MAXQUOTAS < 4);
 548         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 549                 RETURN(0);
 550
 551         do_gettimeofday(&work_start);
 552         for (i = 0; i < MAXQUOTAS; i++) {
 553                 struct lustre_qunit_size *lqs = NULL;
 554
 555                 LASSERT(pending[i] >= 0);
 556                 if (pending[i] == 0)
 557                         continue;
 558
 559                 qdata[i].qd_id = id[i];
 560                 qdata[i].qd_flags = i;
 561                 if (isblk)
 562                         QDATA_SET_BLK(&qdata[i]);
 563                 qdata[i].qd_count = 0;
 564
 565                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 566                         continue;
 567
 568                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 569                 if (lqs == NULL || IS_ERR(lqs)) {
 570                         CERROR("can not find lqs for pending_commit: "
 571                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 572                                "maybe cause unexpected lqs refcount error!\n",
 573                                id[i], i ? 'g': 'u', pending[i], isblk,
 574                                lqs ? PTR_ERR(lqs) : -1);
 575                         continue;
 576                 }
 577
 578                 spin_lock(&lqs->lqs_lock);
 579                 if (isblk) {
 580                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 581                                  "there are too many blocks! [id %u] [%c] "
 582                                  "[bwrite_pending %lu] [pending %u]\n",
 583                                  id[i], i % 2 ? 'g' : 'u',
 584                                  lqs->lqs_bwrite_pending, pending[i]);
 585
 586                         lqs->lqs_bwrite_pending -= pending[i];
 587                 } else {
 588                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 589                                 "there are too many files! [id %u] [%c] "
 590                                 "[iwrite_pending %lu] [pending %u]\n",
 591                                 id[i], i % 2 ? 'g' : 'u',
 592                                 lqs->lqs_iwrite_pending, pending[i]);
 593
 594                         lqs->lqs_iwrite_pending -= pending[i];
 595                 }
 596                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 597                        obd->obd_name,
 598                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 599                        i, pending[i], isblk);
 600                 spin_unlock(&lqs->lqs_lock);
 601
 602                 /* for quota_search_lqs in pending_commit */
 603                 lqs_putref(lqs);
 604                 /* for quota_search_lqs in quota_check */
 605                 lqs_putref(lqs);
 606         }
 607         do_gettimeofday(&work_end);
 608         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 609         lprocfs_counter_add(qctxt->lqc_stats,
 610                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 611                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 612                             timediff);
 613
 614         RETURN(0);
 615 }
 616
 617 static int mds_quota_init(void)
 618 {
 619         return lustre_dquot_init();
 620 }
 621
 622 static int mds_quota_exit(void)
 623 {
 624         lustre_dquot_exit();
 625         return 0;
 626 }
 627
 628 static int mds_quota_setup(struct obd_device *obd)
 629 {
 630         struct obd_device_target *obt = &obd->u.obt;
 631         struct mds_obd *mds = &obd->u.mds;
 632         int rc;
 633         ENTRY;
 634
 635         if (unlikely(mds->mds_quota)) {
 636                 CWARN("try to reinitialize quota context!\n");
 637                 RETURN(0);
 638         }
 639
 640         init_rwsem(&obt->obt_rwsem);
 641         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 642         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 643         sema_init(&obt->obt_quotachecking, 1);
 644         /* initialize quota master and quota context */
 645         sema_init(&mds->mds_qonoff_sem, 1);
 646         rc = qctxt_init(obd, dqacq_handler);
 647         if (rc) {
 648                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 649                 RETURN(rc);
 650         }
 651         mds->mds_quota = 1;
 652         RETURN(rc);
 653 }
 654
 655 static int mds_quota_cleanup(struct obd_device *obd)
 656 {
 657         ENTRY;
 658         if (unlikely(!obd->u.mds.mds_quota))
 659                 RETURN(0);
 660
 661         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 662         RETURN(0);
 663 }
 664
 665 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 666 {
 667         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 668         ENTRY;
 669
 670         if (unlikely(!obd->u.mds.mds_quota))
 671                 RETURN(0);
 672
 673         if (data != NULL)
 674                 QUOTA_MASTER_READY(qctxt);
 675         else
 676                 QUOTA_MASTER_UNREADY(qctxt);
 677         RETURN(0);
 678 }
 679
 680 static int mds_quota_fs_cleanup(struct obd_device *obd)
 681 {
 682         struct mds_obd *mds = &obd->u.mds;
 683         struct obd_quotactl oqctl;
 684         ENTRY;
 685
 686         if (unlikely(!mds->mds_quota))
 687                 RETURN(0);
 688
 689         mds->mds_quota = 0;
 690         memset(&oqctl, 0, sizeof(oqctl));
 691         oqctl.qc_type = UGQUOTA;
 692
 693         down(&mds->mds_qonoff_sem);
 694         mds_admin_quota_off(obd, &oqctl);
 695         up(&mds->mds_qonoff_sem);
 696         RETURN(0);
 697 }
 698
 699 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 700                                 struct obd_trans_info *oti, int isblk)
 701 {
 702         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 703         int rc;
 704         ENTRY;
 705
 706         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 707         RETURN(rc);
 708 }
 709
 710 #endif /* HAVE_QUOTA_SUPPORT */
 711 #endif /* __KERNEL__ */
 712
 713 struct osc_quota_info {
 714         struct list_head        oqi_hash;       /* hash list */
 715         struct client_obd      *oqi_cli;        /* osc obd */
 716         unsigned int            oqi_id;         /* uid/gid of a file */
 717         short                   oqi_type;       /* quota type */
 718 };
 719
 720 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 721
 722 static struct list_head qinfo_hash[NR_DQHASH];
 723 /* SLAB cache for client quota context */
 724 cfs_mem_cache_t *qinfo_cachep = NULL;
 725
 726 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 727                          __attribute__((__const__));
 728
 729 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 730 {
 731         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 732         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 733         return tmp;
 734 }
 735
 736 /* caller must hold qinfo_list_lock */
 737 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 738 {
 739         struct list_head *head = qinfo_hash +
 740                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 741
 742         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 743         list_add(&oqi->oqi_hash, head);
 744 }
 745
 746 /* caller must hold qinfo_list_lock */
 747 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 748 {
 749         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 750         list_del_init(&oqi->oqi_hash);
 751 }
 752
 753 /* caller must hold qinfo_list_lock */
 754 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 755                                                 unsigned int id, int type)
 756 {
 757         unsigned int hashent = hashfn(cli, id, type);
 758         struct osc_quota_info *oqi;
 759         ENTRY;
 760
 761         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 762         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 763                 if (oqi->oqi_cli == cli &&
 764                     oqi->oqi_id == id && oqi->oqi_type == type)
 765                         return oqi;
 766         }
 767         RETURN(NULL);
 768 }
 769
 770 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 771                                           unsigned int id, int type)
 772 {
 773         struct osc_quota_info *oqi;
 774         ENTRY;
 775
 776         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 777         if(!oqi)
 778                 RETURN(NULL);
 779
 780         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 781         oqi->oqi_cli = cli;
 782         oqi->oqi_id = id;
 783         oqi->oqi_type = type;
 784
 785         RETURN(oqi);
 786 }
 787
 788 static void free_qinfo(struct osc_quota_info *oqi)
 789 {
 790         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 791 }
 792
 793 int osc_quota_chkdq(struct client_obd *cli, const unsigned int qid[])
 794 {
 795         unsigned int id;
 796         int cnt, rc = QUOTA_OK;
 797         ENTRY;
 798
 799         spin_lock(&qinfo_list_lock);
 800         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 801                 struct osc_quota_info *oqi = NULL;
 802
 803                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 804                 oqi = find_qinfo(cli, id, cnt);
 805                 if (oqi) {
 806                         rc = NO_QUOTA;
 807                         break;
 808                 }
 809         }
 810         spin_unlock(&qinfo_list_lock);
 811
 812         RETURN(rc);
 813 }
 814
 815 int osc_quota_setdq(struct client_obd *cli, const unsigned int qid[],
 816                     obd_flag valid, obd_flag flags)
 817 {
 818         unsigned int id;
 819         obd_flag noquota;
 820         int cnt, rc = 0;
 821         ENTRY;
 822
 823
 824         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 825                 struct osc_quota_info *oqi, *old;
 826
 827                 if (!(valid & ((cnt == USRQUOTA) ?
 828                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 829                         continue;
 830
 831                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 832                 noquota = (cnt == USRQUOTA) ?
 833                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 834
 835                 oqi = alloc_qinfo(cli, id, cnt);
 836                 if (oqi) {
 837                         spin_lock(&qinfo_list_lock);
 838
 839                         old = find_qinfo(cli, id, cnt);
 840                         if (old && !noquota)
 841                                 remove_qinfo_hash(old);
 842                         else if (!old && noquota)
 843                                 insert_qinfo_hash(oqi);
 844
 845                         spin_unlock(&qinfo_list_lock);
 846
 847                         if (old || !noquota)
 848                                 free_qinfo(oqi);
 849                         if (old && !noquota)
 850                                 free_qinfo(old);
 851                 } else {
 852                         CERROR("not enough mem!\n");
 853                         rc = -ENOMEM;
 854                         break;
 855                 }
 856         }
 857
 858         RETURN(rc);
 859 }
 860
 861 int osc_quota_cleanup(struct obd_device *obd)
 862 {
 863         struct client_obd *cli = &obd->u.cli;
 864         struct osc_quota_info *oqi, *n;
 865         int i;
 866         ENTRY;
 867
 868         spin_lock(&qinfo_list_lock);
 869         for (i = 0; i < NR_DQHASH; i++) {
 870                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 871                         if (oqi->oqi_cli != cli)
 872                                 continue;
 873                         remove_qinfo_hash(oqi);
 874                         free_qinfo(oqi);
 875                 }
 876         }
 877         spin_unlock(&qinfo_list_lock);
 878
 879         RETURN(0);
 880 }
 881
 882 int osc_quota_init(void)
 883 {
 884         int i;
 885         ENTRY;
 886
 887         LASSERT(qinfo_cachep == NULL);
 888         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 889                                             sizeof(struct osc_quota_info),
 890                                             0, 0);
 891         if (!qinfo_cachep)
 892                 RETURN(-ENOMEM);
 893
 894         for (i = 0; i < NR_DQHASH; i++)
 895                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 896
 897         RETURN(0);
 898 }
 899
 900 int osc_quota_exit(void)
 901 {
 902         struct osc_quota_info *oqi, *n;
 903         int i, rc;
 904         ENTRY;
 905
 906         spin_lock(&qinfo_list_lock);
 907         for (i = 0; i < NR_DQHASH; i++) {
 908                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 909                         remove_qinfo_hash(oqi);
 910                         free_qinfo(oqi);
 911                 }
 912         }
 913         spin_unlock(&qinfo_list_lock);
 914
 915         rc = cfs_mem_cache_destroy(qinfo_cachep);
 916         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 917         qinfo_cachep = NULL;
 918
 919         RETURN(0);
 920 }
 921
 922 #ifdef __KERNEL__
 923 #ifdef HAVE_QUOTA_SUPPORT
 924 quota_interface_t mds_quota_interface = {
 925         .quota_init     = mds_quota_init,
 926         .quota_exit     = mds_quota_exit,
 927         .quota_setup    = mds_quota_setup,
 928         .quota_cleanup  = mds_quota_cleanup,
 929         .quota_check    = target_quota_check,
 930         .quota_ctl      = mds_quota_ctl,
 931         .quota_setinfo  = mds_quota_setinfo,
 932         .quota_fs_cleanup = mds_quota_fs_cleanup,
 933         .quota_recovery = mds_quota_recovery,
 934         .quota_adjust   = mds_quota_adjust,
 935         .quota_chkquota = quota_chk_acq_common,
 936         .quota_acquire  = quota_acquire_common,
 937         .quota_pending_commit = quota_pending_commit,
 938 };
 939
 940 quota_interface_t filter_quota_interface = {
 941         .quota_setup    = filter_quota_setup,
 942         .quota_cleanup  = filter_quota_cleanup,
 943         .quota_check    = target_quota_check,
 944         .quota_ctl      = filter_quota_ctl,
 945         .quota_setinfo  = filter_quota_setinfo,
 946         .quota_clearinfo = filter_quota_clearinfo,
 947         .quota_enforce  = filter_quota_enforce,
 948         .quota_getflag  = filter_quota_getflag,
 949         .quota_acquire  = quota_acquire_common,
 950         .quota_adjust   = filter_quota_adjust,
 951         .quota_chkquota = quota_chk_acq_common,
 952         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 953         .quota_pending_commit = quota_pending_commit,
 954 };
 955 #endif
 956 #endif /* __KERNEL__ */
 957
 958 quota_interface_t mdc_quota_interface = {
 959         .quota_ctl      = client_quota_ctl,
 960         .quota_check    = client_quota_check,
 961         .quota_poll_check = client_quota_poll_check,
 962 };
 963
 964 quota_interface_t lmv_quota_interface = {
 965         .quota_ctl      = lmv_quota_ctl,
 966         .quota_check    = lmv_quota_check,
 967 };
 968
 969 quota_interface_t osc_quota_interface = {
 970         .quota_ctl      = client_quota_ctl,
 971         .quota_check    = client_quota_check,
 972         .quota_poll_check = client_quota_poll_check,
 973         .quota_init     = osc_quota_init,
 974         .quota_exit     = osc_quota_exit,
 975         .quota_chkdq    = osc_quota_chkdq,
 976         .quota_setdq    = osc_quota_setdq,
 977         .quota_cleanup  = osc_quota_cleanup,
 978         .quota_adjust_qunit = client_quota_adjust_qunit,
 979 };
 980
 981 quota_interface_t lov_quota_interface = {
 982         .quota_ctl      = lov_quota_ctl,
 983         .quota_check    = lov_quota_check,
 984         .quota_adjust_qunit = lov_quota_adjust_qunit,
 985 };
 986
 987 #ifdef __KERNEL__
 988
 989 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 990
 991 static int __init init_lustre_quota(void)
 992 {
 993 #ifdef HAVE_QUOTA_SUPPORT
 994         int rc = 0;
 995
 996         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 997                                                 proc_lustre_root,
 998                                                 NULL, NULL);
 999         if (IS_ERR(lquota_type_proc_dir)) {
1000                 CERROR("LProcFS failed in lquota-init\n");
1001                 rc = PTR_ERR(lquota_type_proc_dir);
1002                 return rc;
1003         }
1004
1005         rc = qunit_cache_init();
1006         if (rc)
1007                 return rc;
1008
1009         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
1010         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
1011 #endif
1012         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
1013         PORTAL_SYMBOL_REGISTER(lmv_quota_interface);
1014         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
1015         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
1016         return 0;
1017 }
1018
1019 static void /*__exit*/ exit_lustre_quota(void)
1020 {
1021         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
1022         PORTAL_SYMBOL_UNREGISTER(lmv_quota_interface);
1023         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
1024         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
1025 #ifdef HAVE_QUOTA_SUPPORT
1026         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
1027         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
1028
1029         qunit_cache_cleanup();
1030
1031         if (lquota_type_proc_dir)
1032                 lprocfs_remove(&lquota_type_proc_dir);
1033 #endif
1034 }
1035
1036 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1037 MODULE_DESCRIPTION("Lustre Quota");
1038 MODULE_LICENSE("GPL");
1039
1040 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
1041
1042 #ifdef HAVE_QUOTA_SUPPORT
1043 EXPORT_SYMBOL(mds_quota_interface);
1044 EXPORT_SYMBOL(filter_quota_interface);
1045 #endif
1046 EXPORT_SYMBOL(mdc_quota_interface);
1047 EXPORT_SYMBOL(lmv_quota_interface);
1048 EXPORT_SYMBOL(osc_quota_interface);
1049 EXPORT_SYMBOL(lov_quota_interface);
1050 #endif /* __KERNEL */