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                         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                 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                 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 (lqs == NULL || IS_ERR(lqs)) {
 206                         rc = PTR_ERR(lqs);
 207                         break;
 208                 } else {
 209                         spin_lock(&lqs->lqs_lock);
 210                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 211                                 oa->o_flags |= (cnt == USRQUOTA) ?
 212                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 213                                 spin_unlock(&lqs->lqs_lock);
 214                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 215                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 216                                        qctxt->lqc_sync_blk);
 217                                 /* this is for quota_search_lqs */
 218                                 lqs_putref(lqs);
 219                                 continue;
 220                         }
 221                         spin_unlock(&lqs->lqs_lock);
 222                         /* this is for quota_search_lqs */
 223                         lqs_putref(lqs);
 224                 }
 225
 226                 memset(oqctl, 0, sizeof(*oqctl));
 227
 228                 oqctl->qc_cmd = Q_GETQUOTA;
 229                 oqctl->qc_type = cnt;
 230                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 231                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 232                 if (err) {
 233                         if (!rc)
 234                                 rc = err;
 235                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 236                                                              OBD_MD_FLGRPQUOTA);
 237                         continue;
 238                 }
 239
 240                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 241                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 242                     oqctl->qc_dqblk.dqb_bhardlimit))
 243                         oa->o_flags |= (cnt == USRQUOTA) ?
 244                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 245         }
 246         OBD_FREE_PTR(oqctl);
 247         RETURN(rc);
 248 }
 249
 250 /**
 251  * check whether the left quota of certain uid and gid can satisfy a block_write
 252  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 253  */
 254 static int quota_check_common(struct obd_device *obd, const unsigned int id[],
 255                               int pending[], int count, int cycle, int isblk,
 256                               struct inode *inode, int frags)
 257 {
 258         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 259         int i;
 260         struct qunit_data qdata[MAXQUOTAS];
 261         int mb = 0;
 262         int rc = 0, rc2[2] = { 0, 0 };
 263         ENTRY;
 264
 265         spin_lock(&qctxt->lqc_lock);
 266         if (!qctxt->lqc_valid){
 267                 spin_unlock(&qctxt->lqc_lock);
 268                 RETURN(rc);
 269         }
 270         spin_unlock(&qctxt->lqc_lock);
 271
 272         for (i = 0; i < MAXQUOTAS; i++) {
 273                 struct lustre_qunit_size *lqs = NULL;
 274
 275                 qdata[i].qd_id = id[i];
 276                 qdata[i].qd_flags = i;
 277                 if (isblk)
 278                         QDATA_SET_BLK(&qdata[i]);
 279                 qdata[i].qd_count = 0;
 280
 281                 /* ignore root user */
 282                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 283                         continue;
 284
 285                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 286                 if (lqs == NULL || IS_ERR(lqs))
 287                         continue;
 288
 289                 if (IS_ERR(lqs)) {
 290                         CERROR("can not find lqs for check_common: "
 291                                "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n",
 292                                id[i], i % 2 ? 'g': 'u', isblk, count,
 293                                PTR_ERR(lqs));
 294                         RETURN(PTR_ERR(lqs));
 295                 }
 296
 297                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 298                 spin_lock(&lqs->lqs_lock);
 299                 if (!cycle) {
 300                         if (isblk) {
 301                                 pending[i] = count * CFS_PAGE_SIZE;
 302                                 /* in order to complete this write, we need extra
 303                                  * meta blocks. This function can get it through
 304                                  * data needed to be written b=16542 */
 305                                 if (inode) {
 306                                         mb = pending[i];
 307                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 308                                                              &mb, inode,frags);
 309                                         if (rc)
 310                                                 CERROR("%s: can't get extra "
 311                                                        "meta blocks\n",
 312                                                        obd->obd_name);
 313                                         else
 314                                                 pending[i] += mb;
 315                                 }
 316                                 lqs->lqs_bwrite_pending += pending[i];
 317                         } else {
 318                                 pending[i] = count;
 319                                 lqs->lqs_iwrite_pending += pending[i];
 320                         }
 321                 }
 322
 323                 /* if xx_rec < 0, that means quota are releasing,
 324                  * and it may return before we use quota. So if
 325                  * we find this situation, we assuming it has
 326                  * returned b=18491 */
 327                 if (isblk && lqs->lqs_blk_rec < 0) {
 328                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 329                                 qdata[i].qd_count = 0;
 330                         else
 331                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 332                 }
 333                 if (!isblk && lqs->lqs_ino_rec < 0) {
 334                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 335                                 qdata[i].qd_count = 0;
 336                         else
 337                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 338                 }
 339
 340                 CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]"
 341                        " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]"
 342                        " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count,
 343                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 344                        qdata[i].qd_count, mb, pending[i]);
 345                 if (rc2[i] == QUOTA_RET_OK) {
 346                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 347                                 rc2[i] = QUOTA_RET_ACQUOTA;
 348                         if (!isblk && qdata[i].qd_count <
 349                             lqs->lqs_iwrite_pending)
 350                                 rc2[i] = QUOTA_RET_ACQUOTA;
 351                 }
 352
 353                 spin_unlock(&lqs->lqs_lock);
 354
 355                 if (lqs->lqs_blk_rec  < 0 &&
 356                     qdata[i].qd_count <
 357                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 358                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 359
 360                 /* When cycle is zero, lqs_*_pending will be changed. We will
 361                  * get reference of the lqs here and put reference of lqs in
 362                  * quota_pending_commit b=14784 */
 363                 if (!cycle)
 364                         lqs_getref(lqs);
 365
 366                 /* this is for quota_search_lqs */
 367                 lqs_putref(lqs);
 368         }
 369
 370         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 371                 RETURN(QUOTA_RET_ACQUOTA);
 372         else
 373                 RETURN(rc);
 374 }
 375
 376 int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag)
 377 {
 378         struct lustre_qunit_size *lqs;
 379         int i, q_set = 0;
 380
 381         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 382                 RETURN(0);
 383
 384         for (i = 0; i < MAXQUOTAS; i++) {
 385                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 386                                        &obd->u.obt.obt_qctxt, 0);
 387                 if (lqs && !IS_ERR(lqs)) {
 388                         if (lqs->lqs_flags & flag)
 389                                 q_set = 1;
 390                         lqs_putref(lqs);
 391                 }
 392         }
 393
 394         return q_set;
 395 }
 396
 397 static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
 398                                 const unsigned int id[], int pending[],
 399                                 int count, quota_acquire acquire,
 400                                 struct obd_trans_info *oti, int isblk,
 401                                 struct inode *inode, int frags)
 402 {
 403         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 404         struct timeval work_start;
 405         struct timeval work_end;
 406         long timediff;
 407         struct l_wait_info lwi = { 0 };
 408         int rc = 0, cycle = 0, count_err = 1;
 409         ENTRY;
 410
 411         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 412                 RETURN(0);
 413
 414         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 415                 /* If the client has been evicted or if it
 416                  * timed out and tried to reconnect already,
 417                  * abort the request immediately */
 418                 RETURN(-ENOTCONN);
 419
 420         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 421         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 422         /* Unfortunately, if quota master is too busy to handle the
 423          * pre-dqacq in time and quota hash on ost is used up, we
 424          * have to wait for the completion of in flight dqacq/dqrel,
 425          * in order to get enough quota for write b=12588 */
 426         do_gettimeofday(&work_start);
 427         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 428                                         inode, frags)) &
 429                QUOTA_RET_ACQUOTA) {
 430
 431                 spin_lock(&qctxt->lqc_lock);
 432                 if (!qctxt->lqc_import && oti) {
 433                         spin_unlock(&qctxt->lqc_lock);
 434
 435                         LASSERT(oti && oti->oti_thread &&
 436                                 oti->oti_thread->t_watchdog);
 437
 438                         lc_watchdog_disable(oti->oti_thread->t_watchdog);
 439                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 440                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 441                                      &lwi);
 442                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 443                         lc_watchdog_touch(oti->oti_thread->t_watchdog,
 444                                  GET_TIMEOUT(oti->oti_thread->t_svc));
 445                 } else {
 446                         spin_unlock(&qctxt->lqc_lock);
 447                 }
 448
 449                 cycle++;
 450                 if (isblk)
 451                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 452                 /* after acquire(), we should run quota_check_common again
 453                  * so that we confirm there are enough quota to finish write */
 454                 rc = acquire(obd, id, oti, isblk);
 455
 456                 /* please reference to dqacq_completion for the below */
 457                 /* a new request is finished, try again */
 458                 if (rc == QUOTA_REQ_RETURNED) {
 459                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 460                         continue;
 461                 }
 462
 463                 /* it is out of quota already */
 464                 if (rc == -EDQUOT) {
 465                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 466                         break;
 467                 }
 468
 469                 /* Related quota has been disabled by master, but enabled by
 470                  * slave, do not try again. */
 471                 if (unlikely(rc == -ESRCH)) {
 472                         CERROR("mismatched quota configuration, stop try.\n");
 473                         break;
 474                 }
 475
 476                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 477                         /* The client has been evicted or tried to
 478                          * to reconnect already, abort the request */
 479                         RETURN(-ENOTCONN);
 480
 481                 /* -EBUSY and others, wait a second and try again */
 482                 if (rc < 0) {
 483                         cfs_waitq_t        waitq;
 484                         struct l_wait_info lwi;
 485
 486                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 487                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 488                                          GET_TIMEOUT(oti->oti_thread->t_svc));
 489                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 490                                count_err++);
 491
 492                         init_waitqueue_head(&waitq);
 493                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 494                                           NULL);
 495                         l_wait_event(waitq, 0, &lwi);
 496                 }
 497
 498                 if (rc < 0 || cycle % 10 == 0) {
 499                         spin_lock(&last_print_lock);
 500                         if (last_print == 0 ||
 501                             cfs_time_before((last_print + cfs_time_seconds(30)),
 502                                             cfs_time_current())) {
 503                                 last_print = cfs_time_current();
 504                                 spin_unlock(&last_print_lock);
 505                                 CWARN("still haven't managed to acquire quota "
 506                                       "space from the quota master after %d "
 507                                       "retries (err=%d, rc=%d)\n",
 508                                       cycle, count_err - 1, rc);
 509                         } else {
 510                                 spin_unlock(&last_print_lock);
 511                         }
 512                 }
 513
 514                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 515                        cycle);
 516         }
 517         do_gettimeofday(&work_end);
 518         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 519         lprocfs_counter_add(qctxt->lqc_stats,
 520                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 521                                     LQUOTA_WAIT_FOR_CHK_INO,
 522                             timediff);
 523
 524         if (rc > 0)
 525                 rc = 0;
 526         RETURN(rc);
 527 }
 528
 529 /**
 530  * when a block_write or inode_create rpc is finished, adjust the record for
 531  * pending blocks and inodes
 532  */
 533 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 534                                 int pending[], int isblk)
 535 {
 536         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 537         struct timeval work_start;
 538         struct timeval work_end;
 539         long timediff;
 540         int i;
 541         struct qunit_data qdata[MAXQUOTAS];
 542         ENTRY;
 543
 544         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 545         CLASSERT(MAXQUOTAS < 4);
 546         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 547                 RETURN(0);
 548
 549         do_gettimeofday(&work_start);
 550         for (i = 0; i < MAXQUOTAS; i++) {
 551                 struct lustre_qunit_size *lqs = NULL;
 552
 553                 LASSERT(pending[i] >= 0);
 554                 if (pending[i] == 0)
 555                         continue;
 556
 557                 qdata[i].qd_id = id[i];
 558                 qdata[i].qd_flags = i;
 559                 if (isblk)
 560                         QDATA_SET_BLK(&qdata[i]);
 561                 qdata[i].qd_count = 0;
 562
 563                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 564                         continue;
 565
 566                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 567                 if (lqs == NULL || IS_ERR(lqs)) {
 568                         CERROR("can not find lqs for pending_commit: "
 569                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 570                                "maybe cause unexpected lqs refcount error!\n",
 571                                id[i], i ? 'g': 'u', pending[i], isblk,
 572                                lqs ? PTR_ERR(lqs) : -1);
 573                         continue;
 574                 }
 575
 576                 spin_lock(&lqs->lqs_lock);
 577                 if (isblk) {
 578                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 579                                  "there are too many blocks! [id %u] [%c] "
 580                                  "[bwrite_pending %lu] [pending %u]\n",
 581                                  id[i], i % 2 ? 'g' : 'u',
 582                                  lqs->lqs_bwrite_pending, pending[i]);
 583
 584                         lqs->lqs_bwrite_pending -= pending[i];
 585                 } else {
 586                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 587                                 "there are too many files! [id %u] [%c] "
 588                                 "[iwrite_pending %lu] [pending %u]\n",
 589                                 id[i], i % 2 ? 'g' : 'u',
 590                                 lqs->lqs_iwrite_pending, pending[i]);
 591
 592                         lqs->lqs_iwrite_pending -= pending[i];
 593                 }
 594                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 595                        obd->obd_name,
 596                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 597                        i, pending[i], isblk);
 598                 spin_unlock(&lqs->lqs_lock);
 599
 600                 /* for quota_search_lqs in pending_commit */
 601                 lqs_putref(lqs);
 602                 /* for quota_search_lqs in quota_check */
 603                 lqs_putref(lqs);
 604         }
 605         do_gettimeofday(&work_end);
 606         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 607         lprocfs_counter_add(qctxt->lqc_stats,
 608                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 609                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 610                             timediff);
 611
 612         RETURN(0);
 613 }
 614
 615 static int mds_quota_init(void)
 616 {
 617         return lustre_dquot_init();
 618 }
 619
 620 static int mds_quota_exit(void)
 621 {
 622         lustre_dquot_exit();
 623         return 0;
 624 }
 625
 626 static int mds_quota_setup(struct obd_device *obd)
 627 {
 628         struct obd_device_target *obt = &obd->u.obt;
 629         struct mds_obd *mds = &obd->u.mds;
 630         int rc;
 631         ENTRY;
 632
 633         if (unlikely(mds->mds_quota)) {
 634                 CWARN("try to reinitialize quota context!\n");
 635                 RETURN(0);
 636         }
 637
 638         init_rwsem(&obt->obt_rwsem);
 639         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 640         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 641         sema_init(&obt->obt_quotachecking, 1);
 642         /* initialize quota master and quota context */
 643         sema_init(&mds->mds_qonoff_sem, 1);
 644         rc = qctxt_init(obd, dqacq_handler);
 645         if (rc) {
 646                 CERROR("%s: initialize quota context failed! (rc:%d)\n",
 647                        obd->obd_name, rc);
 648                 RETURN(rc);
 649         }
 650         mds->mds_quota = 1;
 651         RETURN(rc);
 652 }
 653
 654 static int mds_quota_cleanup(struct obd_device *obd)
 655 {
 656         ENTRY;
 657         if (unlikely(!obd->u.mds.mds_quota))
 658                 RETURN(0);
 659
 660         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 661         RETURN(0);
 662 }
 663
 664 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 665 {
 666         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 667         ENTRY;
 668
 669         if (unlikely(!obd->u.mds.mds_quota))
 670                 RETURN(0);
 671
 672         if (data != NULL)
 673                 QUOTA_MASTER_READY(qctxt);
 674         else
 675                 QUOTA_MASTER_UNREADY(qctxt);
 676         RETURN(0);
 677 }
 678
 679 static int mds_quota_fs_cleanup(struct obd_device *obd)
 680 {
 681         struct mds_obd *mds = &obd->u.mds;
 682         struct obd_quotactl oqctl;
 683         ENTRY;
 684
 685         if (unlikely(!mds->mds_quota))
 686                 RETURN(0);
 687
 688         mds->mds_quota = 0;
 689         memset(&oqctl, 0, sizeof(oqctl));
 690         oqctl.qc_type = UGQUOTA;
 691
 692         down(&mds->mds_qonoff_sem);
 693         mds_admin_quota_off(obd, &oqctl);
 694         up(&mds->mds_qonoff_sem);
 695         RETURN(0);
 696 }
 697
 698 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 699                                 struct obd_trans_info *oti, int isblk)
 700 {
 701         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 702         int rc;
 703         ENTRY;
 704
 705         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 706         RETURN(rc);
 707 }
 708
 709 #endif /* HAVE_QUOTA_SUPPORT */
 710 #endif /* __KERNEL__ */
 711
 712 struct osc_quota_info {
 713         struct list_head        oqi_hash;       /* hash list */
 714         struct client_obd      *oqi_cli;        /* osc obd */
 715         unsigned int            oqi_id;         /* uid/gid of a file */
 716         short                   oqi_type;       /* quota type */
 717 };
 718
 719 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 720
 721 static struct list_head qinfo_hash[NR_DQHASH];
 722 /* SLAB cache for client quota context */
 723 cfs_mem_cache_t *qinfo_cachep = NULL;
 724
 725 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 726                          __attribute__((__const__));
 727
 728 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 729 {
 730         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 731         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 732         return tmp;
 733 }
 734
 735 /* caller must hold qinfo_list_lock */
 736 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 737 {
 738         struct list_head *head = qinfo_hash +
 739                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 740
 741         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 742         list_add(&oqi->oqi_hash, head);
 743 }
 744
 745 /* caller must hold qinfo_list_lock */
 746 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 747 {
 748         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 749         list_del_init(&oqi->oqi_hash);
 750 }
 751
 752 /* caller must hold qinfo_list_lock */
 753 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 754                                                 unsigned int id, int type)
 755 {
 756         unsigned int hashent = hashfn(cli, id, type);
 757         struct osc_quota_info *oqi;
 758         ENTRY;
 759
 760         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 761         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 762                 if (oqi->oqi_cli == cli &&
 763                     oqi->oqi_id == id && oqi->oqi_type == type)
 764                         return oqi;
 765         }
 766         RETURN(NULL);
 767 }
 768
 769 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 770                                           unsigned int id, int type)
 771 {
 772         struct osc_quota_info *oqi;
 773         ENTRY;
 774
 775         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 776         if(!oqi)
 777                 RETURN(NULL);
 778
 779         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 780         oqi->oqi_cli = cli;
 781         oqi->oqi_id = id;
 782         oqi->oqi_type = type;
 783
 784         RETURN(oqi);
 785 }
 786
 787 static void free_qinfo(struct osc_quota_info *oqi)
 788 {
 789         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 790 }
 791
 792 int osc_quota_chkdq(struct client_obd *cli, const unsigned int qid[])
 793 {
 794         unsigned int id;
 795         int cnt, rc = QUOTA_OK;
 796         ENTRY;
 797
 798         spin_lock(&qinfo_list_lock);
 799         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 800                 struct osc_quota_info *oqi = NULL;
 801
 802                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 803                 oqi = find_qinfo(cli, id, cnt);
 804                 if (oqi) {
 805                         rc = NO_QUOTA;
 806                         break;
 807                 }
 808         }
 809         spin_unlock(&qinfo_list_lock);
 810
 811         RETURN(rc);
 812 }
 813
 814 int osc_quota_setdq(struct client_obd *cli, const unsigned int qid[],
 815                     obd_flag valid, obd_flag flags)
 816 {
 817         unsigned int id;
 818         obd_flag noquota;
 819         int cnt, rc = 0;
 820         ENTRY;
 821
 822
 823         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 824                 struct osc_quota_info *oqi, *old;
 825
 826                 if (!(valid & ((cnt == USRQUOTA) ?
 827                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 828                         continue;
 829
 830                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 831                 noquota = (cnt == USRQUOTA) ?
 832                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 833
 834                 oqi = alloc_qinfo(cli, id, cnt);
 835                 if (!oqi) {
 836                         rc = -ENOMEM;
 837                         break;
 838                 }
 839
 840                 spin_lock(&qinfo_list_lock);
 841                 old = find_qinfo(cli, id, cnt);
 842                 if (old && !noquota)
 843                         remove_qinfo_hash(old);
 844                 else if (!old && noquota)
 845                         insert_qinfo_hash(oqi);
 846                 spin_unlock(&qinfo_list_lock);
 847
 848                 if (old || !noquota)
 849                         free_qinfo(oqi);
 850                 if (old && !noquota)
 851                         free_qinfo(old);
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 int osc_quota_cleanup(struct obd_device *obd)
 858 {
 859         struct client_obd *cli = &obd->u.cli;
 860         struct osc_quota_info *oqi, *n;
 861         int i;
 862         ENTRY;
 863
 864         spin_lock(&qinfo_list_lock);
 865         for (i = 0; i < NR_DQHASH; i++) {
 866                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 867                         if (oqi->oqi_cli != cli)
 868                                 continue;
 869                         remove_qinfo_hash(oqi);
 870                         free_qinfo(oqi);
 871                 }
 872         }
 873         spin_unlock(&qinfo_list_lock);
 874
 875         RETURN(0);
 876 }
 877
 878 int osc_quota_init(void)
 879 {
 880         int i;
 881         ENTRY;
 882
 883         LASSERT(qinfo_cachep == NULL);
 884         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 885                                             sizeof(struct osc_quota_info),
 886                                             0, 0);
 887         if (!qinfo_cachep)
 888                 RETURN(-ENOMEM);
 889
 890         for (i = 0; i < NR_DQHASH; i++)
 891                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 892
 893         RETURN(0);
 894 }
 895
 896 int osc_quota_exit(void)
 897 {
 898         struct osc_quota_info *oqi, *n;
 899         int i, rc;
 900         ENTRY;
 901
 902         spin_lock(&qinfo_list_lock);
 903         for (i = 0; i < NR_DQHASH; i++) {
 904                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 905                         remove_qinfo_hash(oqi);
 906                         free_qinfo(oqi);
 907                 }
 908         }
 909         spin_unlock(&qinfo_list_lock);
 910
 911         rc = cfs_mem_cache_destroy(qinfo_cachep);
 912         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 913         qinfo_cachep = NULL;
 914
 915         RETURN(0);
 916 }
 917
 918 #ifdef __KERNEL__
 919 #ifdef HAVE_QUOTA_SUPPORT
 920 quota_interface_t mds_quota_interface = {
 921         .quota_init     = mds_quota_init,
 922         .quota_exit     = mds_quota_exit,
 923         .quota_setup    = mds_quota_setup,
 924         .quota_cleanup  = mds_quota_cleanup,
 925         .quota_check    = target_quota_check,
 926         .quota_ctl      = mds_quota_ctl,
 927         .quota_setinfo  = mds_quota_setinfo,
 928         .quota_fs_cleanup = mds_quota_fs_cleanup,
 929         .quota_recovery = mds_quota_recovery,
 930         .quota_adjust   = mds_quota_adjust,
 931         .quota_chkquota = quota_chk_acq_common,
 932         .quota_acquire  = quota_acquire_common,
 933         .quota_pending_commit = quota_pending_commit,
 934 };
 935
 936 quota_interface_t filter_quota_interface = {
 937         .quota_setup    = filter_quota_setup,
 938         .quota_cleanup  = filter_quota_cleanup,
 939         .quota_check    = target_quota_check,
 940         .quota_ctl      = filter_quota_ctl,
 941         .quota_setinfo  = filter_quota_setinfo,
 942         .quota_clearinfo = filter_quota_clearinfo,
 943         .quota_enforce  = filter_quota_enforce,
 944         .quota_getflag  = filter_quota_getflag,
 945         .quota_acquire  = quota_acquire_common,
 946         .quota_adjust   = filter_quota_adjust,
 947         .quota_chkquota = quota_chk_acq_common,
 948         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 949         .quota_pending_commit = quota_pending_commit,
 950 };
 951 #endif
 952 #endif /* __KERNEL__ */
 953
 954 quota_interface_t mdc_quota_interface = {
 955         .quota_ctl      = client_quota_ctl,
 956         .quota_check    = client_quota_check,
 957         .quota_poll_check = client_quota_poll_check,
 958 };
 959
 960 quota_interface_t lmv_quota_interface = {
 961         .quota_ctl      = lmv_quota_ctl,
 962         .quota_check    = lmv_quota_check,
 963 };
 964
 965 quota_interface_t osc_quota_interface = {
 966         .quota_ctl      = client_quota_ctl,
 967         .quota_check    = client_quota_check,
 968         .quota_poll_check = client_quota_poll_check,
 969         .quota_init     = osc_quota_init,
 970         .quota_exit     = osc_quota_exit,
 971         .quota_chkdq    = osc_quota_chkdq,
 972         .quota_setdq    = osc_quota_setdq,
 973         .quota_cleanup  = osc_quota_cleanup,
 974         .quota_adjust_qunit = client_quota_adjust_qunit,
 975 };
 976
 977 quota_interface_t lov_quota_interface = {
 978         .quota_ctl      = lov_quota_ctl,
 979         .quota_check    = lov_quota_check,
 980         .quota_adjust_qunit = lov_quota_adjust_qunit,
 981 };
 982
 983 #ifdef __KERNEL__
 984
 985 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 986
 987 static int __init init_lustre_quota(void)
 988 {
 989 #ifdef HAVE_QUOTA_SUPPORT
 990         int rc = 0;
 991
 992         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 993                                                 proc_lustre_root,
 994                                                 NULL, NULL);
 995         if (IS_ERR(lquota_type_proc_dir)) {
 996                 CERROR("LProcFS failed in lquota-init\n");
 997                 rc = PTR_ERR(lquota_type_proc_dir);
 998                 return rc;
 999         }
1000
1001         rc = qunit_cache_init();
1002         if (rc)
1003                 return rc;
1004
1005         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
1006         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
1007 #endif
1008         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
1009         PORTAL_SYMBOL_REGISTER(lmv_quota_interface);
1010         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
1011         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
1012         return 0;
1013 }
1014
1015 static void /*__exit*/ exit_lustre_quota(void)
1016 {
1017         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
1018         PORTAL_SYMBOL_UNREGISTER(lmv_quota_interface);
1019         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
1020         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
1021 #ifdef HAVE_QUOTA_SUPPORT
1022         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
1023         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
1024
1025         qunit_cache_cleanup();
1026
1027         if (lquota_type_proc_dir)
1028                 lprocfs_remove(&lquota_type_proc_dir);
1029 #endif
1030 }
1031
1032 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1033 MODULE_DESCRIPTION("Lustre Quota");
1034 MODULE_LICENSE("GPL");
1035
1036 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
1037
1038 #ifdef HAVE_QUOTA_SUPPORT
1039 EXPORT_SYMBOL(mds_quota_interface);
1040 EXPORT_SYMBOL(filter_quota_interface);
1041 #endif
1042 EXPORT_SYMBOL(mdc_quota_interface);
1043 EXPORT_SYMBOL(lmv_quota_interface);
1044 EXPORT_SYMBOL(osc_quota_interface);
1045 EXPORT_SYMBOL(lov_quota_interface);
1046 #endif /* __KERNEL */