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 (c) 2006, 2010, Oracle and/or its affiliates. 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/smp_lock.h>
  48 #  include <linux/buffer_head.h>
  49 #  include <linux/workqueue.h>
  50 #  include <linux/mount.h>
  51 #else /* __KERNEL__ */
  52 # include <liblustre.h>
  53 #endif
  54
  55 #include <obd_class.h>
  56 #include <lustre_mds.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_cfg.h>
  59 #include <obd_ost.h>
  60 #include <lustre_fsfilt.h>
  61 #include <lustre_quota.h>
  62 #include <lprocfs_status.h>
  63 #include "quota_internal.h"
  64
  65 #ifdef __KERNEL__
  66
  67 #ifdef HAVE_QUOTA_SUPPORT
  68
  69 static cfs_time_t last_print = 0;
  70 static spinlock_t last_print_lock = SPIN_LOCK_UNLOCKED;
  71
  72 static int filter_quota_setup(struct obd_device *obd)
  73 {
  74         int rc = 0;
  75         struct obd_device_target *obt = &obd->u.obt;
  76         ENTRY;
  77
  78 #ifdef HAVE_QUOTA64
  79         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  80 #else
  81         obt->obt_qfmt = LUSTRE_QUOTA_V1;
  82 #endif
  83         atomic_set(&obt->obt_quotachecking, 1);
  84         rc = qctxt_init(obd, NULL);
  85         if (rc)
  86                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  87
  88         RETURN(rc);
  89 }
  90
  91 static int filter_quota_cleanup(struct obd_device *obd)
  92 {
  93         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  94         return 0;
  95 }
  96
  97 static int filter_quota_setinfo(struct obd_export *exp, struct obd_device *obd)
  98 {
  99         struct obd_import *imp;
 100         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 101         ENTRY;
 102
 103         /* setup the quota context import */
 104         spin_lock(&obd->u.obt.obt_qctxt.lqc_lock);
 105         obd->u.obt.obt_qctxt.lqc_import = exp->exp_imp_reverse;
 106         spin_unlock(&obd->u.obt.obt_qctxt.lqc_lock);
 107         CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated now, \n",
 108                obd->obd_name,exp->exp_imp_reverse, obd);
 109
 110         /* make imp's connect flags equal relative exp's connect flags
 111          * adding it to avoid the scan export list
 112          */
 113         imp = exp->exp_imp_reverse;
 114         if (imp)
 115                 imp->imp_connect_data.ocd_connect_flags |=
 116                         (exp->exp_connect_flags &
 117                          (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 118
 119         cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 120         /* start quota slave recovery thread. (release high limits) */
 121         qslave_start_recovery(obd, &obd->u.obt.obt_qctxt);
 122         RETURN(0);
 123 }
 124
 125 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 126 {
 127         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 128         ENTRY;
 129
 130         /* lquota may be not set up before destroying export, b=14896 */
 131         if (!obd->obd_set_up)
 132                 RETURN(0);
 133
 134         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 135          * should be invalid b=12374 */
 136         if (qctxt->lqc_import && qctxt->lqc_import == exp->exp_imp_reverse) {
 137                 spin_lock(&qctxt->lqc_lock);
 138                 qctxt->lqc_import = NULL;
 139                 spin_unlock(&qctxt->lqc_lock);
 140                 ptlrpc_cleanup_imp(exp->exp_imp_reverse);
 141                 dqacq_interrupt(qctxt);
 142                 CDEBUG(D_QUOTA, "%s: lqc_import of obd(%p) is invalid now.\n",
 143                        obd->obd_name, obd);
 144         }
 145         RETURN(0);
 146 }
 147
 148 static int target_quota_enforce(struct obd_device *obd, unsigned int ignore)
 149 {
 150         ENTRY;
 151
 152         if (!ll_sb_any_quota_active(obd->u.obt.obt_sb))
 153                 RETURN(-EINVAL);
 154
 155         if (!!cfs_cap_raised(CFS_CAP_SYS_RESOURCE) == !!ignore)
 156                 RETURN(-EALREADY);
 157
 158         if (ignore) {
 159                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 160                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 161         } else {
 162                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 163         }
 164
 165         RETURN(0);
 166 }
 167
 168 #define GET_OA_ID(flag, oa) (flag == USRQUOTA ? oa->o_uid : oa->o_gid)
 169 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 170 {
 171         struct obd_device_target *obt = &obd->u.obt;
 172         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 173         int err, cnt, rc = 0;
 174         struct obd_quotactl *oqctl;
 175         ENTRY;
 176
 177         if (!ll_sb_any_quota_active(obt->obt_sb))
 178                 RETURN(0);
 179
 180         OBD_ALLOC_PTR(oqctl);
 181         if (!oqctl)
 182                 RETURN(-ENOMEM);
 183
 184         /* set over quota flags for a uid/gid */
 185         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 186         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 187
 188         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 189                 struct lustre_qunit_size *lqs = NULL;
 190
 191                 lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)),
 192                                        qctxt, 0);
 193                 if (lqs == NULL || IS_ERR(lqs)) {
 194                         rc = PTR_ERR(lqs);
 195                         break;
 196                 } else {
 197                         spin_lock(&lqs->lqs_lock);
 198                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 199                                 oa->o_flags |= (cnt == USRQUOTA) ?
 200                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 201                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 202                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 203                                        qctxt->lqc_sync_blk);
 204                                 spin_unlock(&lqs->lqs_lock);
 205                                 /* this is for quota_search_lqs */
 206                                 lqs_putref(lqs);
 207                                 continue;
 208                         }
 209                         spin_unlock(&lqs->lqs_lock);
 210                         /* this is for quota_search_lqs */
 211                         lqs_putref(lqs);
 212                 }
 213
 214                 memset(oqctl, 0, sizeof(*oqctl));
 215
 216                 oqctl->qc_cmd = Q_GETQUOTA;
 217                 oqctl->qc_type = cnt;
 218                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 219                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 220                 if (err) {
 221                         if (!rc)
 222                                 rc = err;
 223                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 224                                                              OBD_MD_FLGRPQUOTA);
 225                         continue;
 226                 }
 227
 228                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 229                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 230                     oqctl->qc_dqblk.dqb_bhardlimit))
 231                         oa->o_flags |= (cnt == USRQUOTA) ?
 232                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 233         }
 234         OBD_FREE_PTR(oqctl);
 235         RETURN(rc);
 236 }
 237
 238 static int filter_quota_acquire(struct obd_device *obd, unsigned int uid,
 239                                 unsigned int gid, struct obd_trans_info *oti)
 240 {
 241         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 242         int rc;
 243         ENTRY;
 244
 245         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, LQUOTA_FLAGS_BLK, 1, oti);
 246         RETURN(rc);
 247 }
 248
 249 /* check whether the left quota of certain uid and gid can satisfy a block_write
 250  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA */
 251 static int quota_check_common(struct obd_device *obd, unsigned int uid,
 252                               unsigned int gid, int count, int cycle, int isblk,
 253                               struct inode *inode, int frags, int pending[2])
 254 {
 255         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 256         int i;
 257         __u32 id[MAXQUOTAS] = { uid, gid };
 258         struct qunit_data qdata[MAXQUOTAS];
 259         int mb = 0;
 260         int rc = 0, rc2[2] = { 0, 0 };
 261         ENTRY;
 262
 263         spin_lock(&qctxt->lqc_lock);
 264         if (!qctxt->lqc_valid){
 265                 spin_unlock(&qctxt->lqc_lock);
 266                 RETURN(rc);
 267         }
 268         spin_unlock(&qctxt->lqc_lock);
 269
 270         for (i = 0; i < MAXQUOTAS; i++) {
 271                 struct lustre_qunit_size *lqs = NULL;
 272
 273                 qdata[i].qd_id = id[i];
 274                 qdata[i].qd_flags = i;
 275                 if (isblk)
 276                         QDATA_SET_BLK(&qdata[i]);
 277                 qdata[i].qd_count = 0;
 278
 279                 /* ignore root user */
 280                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 281                         continue;
 282
 283                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 284                 if (lqs == NULL || IS_ERR(lqs))
 285                         continue;
 286
 287                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 288                 spin_lock(&lqs->lqs_lock);
 289                 if (!cycle) {
 290                         if (isblk) {
 291                                 pending[i] = count * CFS_PAGE_SIZE;
 292                                 /* in order to complete this write, we need extra
 293                                  * meta blocks. This function can get it through
 294                                  * data needed to be written b=16542 */
 295                                 mb = pending[i];
 296                                 LASSERT(inode && frags > 0);
 297                                 if (fsfilt_get_mblk(obd, qctxt->lqc_sb, &mb,
 298                                                     inode, frags) < 0)
 299                                         CERROR("%s: can't get extra meta "
 300                                                "blocks\n", obd->obd_name);
 301                                 else
 302                                         pending[i] += mb;
 303                                 lqs->lqs_bwrite_pending += pending[i];
 304                         } else {
 305                                 pending[i] = count;
 306                                 lqs->lqs_iwrite_pending += pending[i];
 307                         }
 308                 }
 309
 310                 /* if xx_rec < 0, that means quota are releasing,
 311                  * and it may return before we use quota. So if
 312                  * we find this situation, we assuming it has
 313                  * returned b=18491 */
 314                 if (isblk && lqs->lqs_blk_rec < 0) {
 315                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 316                                 qdata[i].qd_count = 0;
 317                         else
 318                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 319                 }
 320                 if (!isblk && lqs->lqs_ino_rec < 0) {
 321                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 322                                 qdata[i].qd_count = 0;
 323                         else
 324                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 325                 }
 326
 327                 CDEBUG(D_QUOTA, "count=%d lqs_pending=%lu qd_count="LPU64
 328                        " isblk=%d mb=%d pending[%d]=%d\n", count,
 329                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 330                        qdata[i].qd_count, isblk, mb, i, pending[i]);
 331                 if (rc2[i] == QUOTA_RET_OK) {
 332                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 333                                 rc2[i] = QUOTA_RET_ACQUOTA;
 334                         if (!isblk && qdata[i].qd_count <
 335                             lqs->lqs_iwrite_pending)
 336                                 rc2[i] = QUOTA_RET_ACQUOTA;
 337                 }
 338
 339                 spin_unlock(&lqs->lqs_lock);
 340
 341                 if (lqs->lqs_blk_rec  < 0 &&
 342                     qdata[i].qd_count <
 343                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 344                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 345
 346                 /* When cycle is zero, lqs_*_pending will be changed. We will
 347                  * get reference of the lqs here and put reference of lqs in
 348                  * quota_pending_commit b=14784 */
 349                 if (!cycle)
 350                         lqs_getref(lqs);
 351
 352                 /* this is for quota_search_lqs */
 353                 lqs_putref(lqs);
 354         }
 355
 356         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 357                 RETURN(QUOTA_RET_ACQUOTA);
 358         else
 359                 RETURN(rc);
 360 }
 361
 362 static int quota_chk_acq_common(struct obd_export *exp, unsigned int uid,
 363                                 unsigned int gid, int count, int pending[2],
 364                                 int isblk, quota_acquire acquire,
 365                                 struct obd_trans_info *oti, struct inode *inode,
 366                                 int frags)
 367 {
 368         struct obd_device *obd = exp->exp_obd;
 369         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 370         struct timeval work_start;
 371         struct timeval work_end;
 372         long timediff;
 373         struct l_wait_info lwi = { 0 };
 374         int rc = 0, cycle = 0, count_err = 1;
 375         ENTRY;
 376
 377         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 378         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 379                 /* If the client has been evicted or if it
 380                  * timed out and tried to reconnect already,
 381                  * abort the request immediately */
 382                 RETURN(-ENOTCONN);
 383
 384         /* Unfortunately, if quota master is too busy to handle the
 385          * pre-dqacq in time and quota hash on ost is used up, we
 386          * have to wait for the completion of in flight dqacq/dqrel,
 387          * in order to get enough quota for write b=12588 */
 388         do_gettimeofday(&work_start);
 389         while ((rc = quota_check_common(obd, uid, gid, count, cycle, isblk,
 390                                         inode, frags, pending)) & QUOTA_RET_ACQUOTA) {
 391
 392                 spin_lock(&qctxt->lqc_lock);
 393                 if (!qctxt->lqc_import && oti) {
 394                         spin_unlock(&qctxt->lqc_lock);
 395
 396                         LASSERT(oti && oti->oti_thread &&
 397                                 oti->oti_thread->t_watchdog);
 398
 399                         lc_watchdog_disable(oti->oti_thread->t_watchdog);
 400                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 401                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 402                                      &lwi);
 403                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 404                         lc_watchdog_touch(oti->oti_thread->t_watchdog,
 405                                  GET_TIMEOUT(oti->oti_thread->t_svc));
 406                 } else {
 407                         spin_unlock(&qctxt->lqc_lock);
 408                 }
 409
 410                 cycle++;
 411                 if (isblk)
 412                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 413                 /* after acquire(), we should run quota_check_common again
 414                  * so that we confirm there are enough quota to finish write */
 415                 rc = acquire(obd, uid, gid, oti);
 416
 417                 /* please reference to dqacq_completion for the below */
 418                 /* a new request is finished, try again */
 419                 if (rc == QUOTA_REQ_RETURNED) {
 420                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 421                         continue;
 422                 }
 423
 424                 /* it is out of quota already */
 425                 if (rc == -EDQUOT) {
 426                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 427                         break;
 428                 }
 429
 430                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 431                         /* The client has been evicted or tried to
 432                          * to reconnect already, abort the request */
 433                         RETURN(-ENOTCONN);
 434
 435                 /* -EBUSY and others, wait a second and try again */
 436                 if (rc < 0) {
 437                         cfs_waitq_t        waitq;
 438                         struct l_wait_info lwi;
 439
 440                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 441                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 442                                          GET_TIMEOUT(oti->oti_thread->t_svc));
 443                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 444                                count_err++);
 445
 446                         init_waitqueue_head(&waitq);
 447                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 448                                           NULL);
 449                         l_wait_event(waitq, 0, &lwi);
 450                 }
 451
 452                 if (rc < 0 || cycle % 10 == 0) {
 453                         spin_lock(&last_print_lock);
 454                         if (last_print == 0 ||
 455                             cfs_time_before((last_print + cfs_time_seconds(30)),
 456                                             cfs_time_current())) {
 457                                 CWARN("still haven't managed to acquire quota "
 458                                       "space from the quota master after %d "
 459                                       "retries (err=%d, rc=%d)\n",
 460                                       cycle, count_err - 1, rc);
 461                                 last_print = cfs_time_current();
 462                         }
 463                         spin_unlock(&last_print_lock);
 464                 }
 465
 466                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 467                        cycle);
 468         }
 469
 470         do_gettimeofday(&work_end);
 471         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 472         lprocfs_counter_add(qctxt->lqc_stats,
 473                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 474                                     LQUOTA_WAIT_FOR_CHK_INO,
 475                             timediff);
 476
 477         if (rc > 0)
 478                 rc = 0;
 479         RETURN(rc);
 480 }
 481
 482 int quota_is_set(struct obd_device *obd, unsigned int uid,
 483                  unsigned int gid, int flag)
 484 {
 485         struct lustre_qunit_size *lqs;
 486         __u32 id[MAXQUOTAS] = { uid, gid };
 487         int i, q_set = 0;
 488
 489         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 490                 RETURN(0);
 491
 492         for (i = 0; i < MAXQUOTAS; i++) {
 493                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 494                                        &obd->u.obt.obt_qctxt, 0);
 495                 if (lqs && !IS_ERR(lqs)) {
 496                         if (lqs->lqs_flags & flag)
 497                                 q_set = 1;
 498                         lqs_putref(lqs);
 499                 }
 500         }
 501
 502         return q_set;
 503 }
 504
 505 static int filter_quota_check(struct obd_export *exp, unsigned int uid,
 506                               unsigned int gid, int npage, int pending[2],
 507                               quota_acquire acquire, struct obd_trans_info *oti,
 508                               struct inode *inode, int frags)
 509 {
 510         return quota_is_set(exp->exp_obd, uid, gid, QB_SET) ?
 511                 quota_chk_acq_common(exp, uid, gid, npage, pending,
 512                                      LQUOTA_FLAGS_BLK, acquire, oti, inode,
 513                                      frags) : 0;
 514 }
 515
 516 /* when a block_write or inode_create rpc is finished, adjust the record for
 517  * pending blocks and inodes*/
 518 static int quota_pending_commit(struct obd_device *obd, unsigned int uid,
 519                                 unsigned int gid, int pending[2], int isblk)
 520 {
 521         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 522         struct timeval work_start;
 523         struct timeval work_end;
 524         long timediff;
 525         int i;
 526         __u32 id[MAXQUOTAS] = { uid, gid };
 527         struct qunit_data qdata[MAXQUOTAS];
 528         ENTRY;
 529
 530         CDEBUG(D_QUOTA, "%s: commit pending quota\n", obd->obd_name);
 531         CLASSERT(MAXQUOTAS < 4);
 532
 533         do_gettimeofday(&work_start);
 534         for (i = 0; i < MAXQUOTAS; i++) {
 535                 struct lustre_qunit_size *lqs = NULL;
 536                 int flag = 0;
 537
 538                 qdata[i].qd_id = id[i];
 539                 qdata[i].qd_flags = i;
 540                 if (isblk)
 541                         QDATA_SET_BLK(&qdata[i]);
 542                 qdata[i].qd_count = 0;
 543
 544                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 545                         continue;
 546
 547                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 548                 if (lqs == NULL || IS_ERR(lqs))
 549                         continue;
 550
 551                 spin_lock(&lqs->lqs_lock);
 552                 if (isblk) {
 553                         if (lqs->lqs_bwrite_pending >= pending[i]) {
 554                                 lqs->lqs_bwrite_pending -= pending[i];
 555                                 flag = 1;
 556                         } else {
 557                                 CERROR("%s: there are too many blocks!\n",
 558                                        obd->obd_name);
 559                         }
 560                 } else {
 561                         if (lqs->lqs_iwrite_pending >= pending[i]) {
 562                                 lqs->lqs_iwrite_pending -= pending[i];
 563                                 flag = 1;
 564                         } else {
 565                                 CERROR("%s: there are too many files!\n",
 566                                        obd->obd_name);
 567                         }
 568                 }
 569                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 570                        obd->obd_name,
 571                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 572                        i, pending[i], isblk);
 573
 574                 spin_unlock(&lqs->lqs_lock);
 575                 lqs_putref(lqs);
 576                 /* When lqs_*_pening is changed back, we'll putref lqs
 577                  * here b=14784 */
 578                 if (flag)
 579                         lqs_putref(lqs);
 580         }
 581         do_gettimeofday(&work_end);
 582         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 583         lprocfs_counter_add(qctxt->lqc_stats,
 584                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 585                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 586                             timediff);
 587
 588         RETURN(0);
 589 }
 590
 591 static int filter_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 592                                        unsigned int gid, int pending[2])
 593 {
 594         return quota_pending_commit(obd, uid, gid, pending, LQUOTA_FLAGS_BLK);
 595 }
 596
 597 static int mds_quota_init(void)
 598 {
 599         return lustre_dquot_init();
 600 }
 601
 602 static int mds_quota_exit(void)
 603 {
 604         lustre_dquot_exit();
 605         return 0;
 606 }
 607
 608 static int mds_quota_setup(struct obd_device *obd)
 609 {
 610         struct obd_device_target *obt = &obd->u.obt;
 611         struct mds_obd *mds = &obd->u.mds;
 612         int rc;
 613         ENTRY;
 614
 615 #ifdef HAVE_QUOTA64
 616         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 617 #else
 618         obt->obt_qfmt = LUSTRE_QUOTA_V1;
 619 #endif
 620         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 621         atomic_set(&obt->obt_quotachecking, 1);
 622         /* initialize quota master and quota context */
 623         sema_init(&mds->mds_qonoff_sem, 1);
 624         rc = qctxt_init(obd, dqacq_handler);
 625         if (rc) {
 626                 CERROR("%s: initialize quota context failed! (rc:%d)\n",
 627                        obd->obd_name, rc);
 628                 RETURN(rc);
 629         }
 630         RETURN(rc);
 631 }
 632
 633 static int mds_quota_cleanup(struct obd_device *obd)
 634 {
 635         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 636         RETURN(0);
 637 }
 638
 639 static int mds_quota_fs_cleanup(struct obd_device *obd)
 640 {
 641         struct mds_obd *mds = &obd->u.mds;
 642         struct obd_quotactl oqctl;
 643         ENTRY;
 644
 645         memset(&oqctl, 0, sizeof(oqctl));
 646         oqctl.qc_type = UGQUOTA;
 647
 648         down(&mds->mds_qonoff_sem);
 649         mds_admin_quota_off(obd, &oqctl);
 650         up(&mds->mds_qonoff_sem);
 651         RETURN(0);
 652 }
 653
 654 static int mds_quota_check(struct obd_export *exp, unsigned int uid,
 655                            unsigned int gid, int inodes, int pending[2],
 656                            quota_acquire acquire, struct obd_trans_info *oti,
 657                            struct inode *inode, int frags)
 658 {
 659         return quota_is_set(exp->exp_obd, uid, gid, QI_SET) ?
 660                 quota_chk_acq_common(exp, uid, gid, inodes, pending, 0,
 661                                      acquire, oti, inode, frags) : 0;
 662 }
 663
 664 static int mds_quota_acquire(struct obd_device *obd, unsigned int uid,
 665                              unsigned int gid, struct obd_trans_info *oti)
 666 {
 667         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 668         int rc;
 669         ENTRY;
 670
 671         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, 0, 1, oti);
 672         RETURN(rc);
 673 }
 674
 675 static int mds_quota_pending_commit(struct obd_device *obd, unsigned int uid,
 676                                     unsigned int gid, int pending[2])
 677 {
 678         return quota_pending_commit(obd, uid, gid, pending, 0);
 679 }
 680 #endif /* HAVE_QUOTA_SUPPORT */
 681 #endif /* __KERNEL__ */
 682
 683 struct osc_quota_info {
 684         struct list_head        oqi_hash;       /* hash list */
 685         struct client_obd      *oqi_cli;        /* osc obd */
 686         unsigned int            oqi_id;         /* uid/gid of a file */
 687         short                   oqi_type;       /* quota type */
 688 };
 689
 690 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 691
 692 static struct list_head qinfo_hash[NR_DQHASH];
 693 /* SLAB cache for client quota context */
 694 cfs_mem_cache_t *qinfo_cachep = NULL;
 695
 696 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 697                          __attribute__((__const__));
 698
 699 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 700 {
 701         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 702         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 703         return tmp;
 704 }
 705
 706 /* caller must hold qinfo_list_lock */
 707 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 708 {
 709         struct list_head *head = qinfo_hash +
 710                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 711
 712         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 713         list_add(&oqi->oqi_hash, head);
 714 }
 715
 716 /* caller must hold qinfo_list_lock */
 717 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 718 {
 719         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 720         list_del_init(&oqi->oqi_hash);
 721 }
 722
 723 /* caller must hold qinfo_list_lock */
 724 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 725                                                 unsigned int id, int type)
 726 {
 727         unsigned int hashent = hashfn(cli, id, type);
 728         struct osc_quota_info *oqi;
 729
 730         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 731         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 732                 if (oqi->oqi_cli == cli &&
 733                     oqi->oqi_id == id && oqi->oqi_type == type)
 734                         return oqi;
 735         }
 736         return NULL;
 737 }
 738
 739 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 740                                           unsigned int id, int type)
 741 {
 742         struct osc_quota_info *oqi;
 743         ENTRY;
 744
 745         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 746         if(!oqi)
 747                 RETURN(NULL);
 748
 749         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 750         oqi->oqi_cli = cli;
 751         oqi->oqi_id = id;
 752         oqi->oqi_type = type;
 753
 754         RETURN(oqi);
 755 }
 756
 757 static void free_qinfo(struct osc_quota_info *oqi)
 758 {
 759         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 760 }
 761
 762 int osc_quota_chkdq(struct client_obd *cli, unsigned int uid, unsigned int gid)
 763 {
 764         unsigned int id;
 765         int cnt, rc = QUOTA_OK;
 766         ENTRY;
 767
 768         spin_lock(&qinfo_list_lock);
 769         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 770                 struct osc_quota_info *oqi = NULL;
 771
 772                 id = (cnt == USRQUOTA) ? uid : gid;
 773                 oqi = find_qinfo(cli, id, cnt);
 774                 if (oqi) {
 775                         rc = NO_QUOTA;
 776                         break;
 777                 }
 778         }
 779         spin_unlock(&qinfo_list_lock);
 780
 781         RETURN(rc);
 782 }
 783
 784 int osc_quota_setdq(struct client_obd *cli, unsigned int uid, unsigned int gid,
 785                     obd_flag valid, obd_flag flags)
 786 {
 787         unsigned int id;
 788         obd_flag noquota;
 789         int cnt, rc = 0;
 790         ENTRY;
 791
 792
 793         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 794                 struct osc_quota_info *oqi, *old;
 795
 796                 if (!(valid & ((cnt == USRQUOTA) ?
 797                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 798                         continue;
 799
 800                 id = (cnt == USRQUOTA) ? uid : gid;
 801                 noquota = (cnt == USRQUOTA) ?
 802                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 803
 804                 oqi = alloc_qinfo(cli, id, cnt);
 805                 if (!oqi) {
 806                         rc = -ENOMEM;
 807                         break;
 808                 }
 809
 810                 spin_lock(&qinfo_list_lock);
 811                 old = find_qinfo(cli, id, cnt);
 812                 if (old && !noquota)
 813                         remove_qinfo_hash(old);
 814                 else if (!old && noquota)
 815                         insert_qinfo_hash(oqi);
 816                 spin_unlock(&qinfo_list_lock);
 817
 818                 if (old || !noquota)
 819                         free_qinfo(oqi);
 820                 if (old && !noquota)
 821                         free_qinfo(old);
 822         }
 823
 824         RETURN(rc);
 825 }
 826
 827 int osc_quota_cleanup(struct obd_device *obd)
 828 {
 829         struct client_obd *cli = &obd->u.cli;
 830         struct osc_quota_info *oqi, *n;
 831         int i;
 832         ENTRY;
 833
 834         spin_lock(&qinfo_list_lock);
 835         for (i = 0; i < NR_DQHASH; i++) {
 836                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 837                         if (oqi->oqi_cli != cli)
 838                                 continue;
 839                         remove_qinfo_hash(oqi);
 840                         free_qinfo(oqi);
 841                 }
 842         }
 843         spin_unlock(&qinfo_list_lock);
 844
 845         RETURN(0);
 846 }
 847
 848 int osc_quota_init(void)
 849 {
 850         int i;
 851         ENTRY;
 852
 853         LASSERT(qinfo_cachep == NULL);
 854         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 855                                             sizeof(struct osc_quota_info),
 856                                             0, 0);
 857         if (!qinfo_cachep)
 858                 RETURN(-ENOMEM);
 859
 860         for (i = 0; i < NR_DQHASH; i++)
 861                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 862
 863         RETURN(0);
 864 }
 865
 866 int osc_quota_exit(void)
 867 {
 868         struct osc_quota_info *oqi, *n;
 869         int i, rc;
 870         ENTRY;
 871
 872         spin_lock(&qinfo_list_lock);
 873         for (i = 0; i < NR_DQHASH; i++) {
 874                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 875                         remove_qinfo_hash(oqi);
 876                         free_qinfo(oqi);
 877                 }
 878         }
 879         spin_unlock(&qinfo_list_lock);
 880
 881         rc = cfs_mem_cache_destroy(qinfo_cachep);
 882         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 883         qinfo_cachep = NULL;
 884
 885         RETURN(0);
 886 }
 887
 888 #ifdef __KERNEL__
 889 #ifdef HAVE_QUOTA_SUPPORT
 890 quota_interface_t mds_quota_interface = {
 891         .quota_init     = mds_quota_init,
 892         .quota_exit     = mds_quota_exit,
 893         .quota_setup    = mds_quota_setup,
 894         .quota_cleanup  = mds_quota_cleanup,
 895         .quota_check    = target_quota_check,
 896         .quota_ctl      = mds_quota_ctl,
 897         .quota_fs_cleanup       =mds_quota_fs_cleanup,
 898         .quota_recovery = mds_quota_recovery,
 899         .quota_adjust   = mds_quota_adjust,
 900         .quota_chkquota = mds_quota_check,
 901         .quota_enforce  = target_quota_enforce,
 902         .quota_acquire  = mds_quota_acquire,
 903         .quota_pending_commit = mds_quota_pending_commit,
 904 };
 905
 906 quota_interface_t filter_quota_interface = {
 907         .quota_setup    = filter_quota_setup,
 908         .quota_cleanup  = filter_quota_cleanup,
 909         .quota_check    = target_quota_check,
 910         .quota_ctl      = filter_quota_ctl,
 911         .quota_setinfo  = filter_quota_setinfo,
 912         .quota_clearinfo = filter_quota_clearinfo,
 913         .quota_enforce  = target_quota_enforce,
 914         .quota_getflag  = filter_quota_getflag,
 915         .quota_acquire  = filter_quota_acquire,
 916         .quota_adjust   = filter_quota_adjust,
 917         .quota_chkquota = filter_quota_check,
 918         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 919         .quota_pending_commit = filter_quota_pending_commit,
 920 };
 921 #endif
 922 #endif /* __KERNEL__ */
 923
 924 quota_interface_t mdc_quota_interface = {
 925         .quota_ctl      = client_quota_ctl,
 926         .quota_check    = client_quota_check,
 927         .quota_poll_check = client_quota_poll_check,
 928 };
 929
 930 quota_interface_t osc_quota_interface = {
 931         .quota_ctl      = client_quota_ctl,
 932         .quota_check    = client_quota_check,
 933         .quota_poll_check = client_quota_poll_check,
 934         .quota_init     = osc_quota_init,
 935         .quota_exit     = osc_quota_exit,
 936         .quota_chkdq    = osc_quota_chkdq,
 937         .quota_setdq    = osc_quota_setdq,
 938         .quota_cleanup  = osc_quota_cleanup,
 939         .quota_adjust_qunit = client_quota_adjust_qunit,
 940 };
 941
 942 quota_interface_t lov_quota_interface = {
 943         .quota_check    = lov_quota_check,
 944         .quota_ctl      = lov_quota_ctl,
 945         .quota_adjust_qunit = lov_quota_adjust_qunit,
 946 };
 947
 948 #ifdef __KERNEL__
 949
 950 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 951
 952 static int __init init_lustre_quota(void)
 953 {
 954 #ifdef HAVE_QUOTA_SUPPORT
 955         int rc = 0;
 956
 957         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 958                                                 proc_lustre_root,
 959                                                 NULL, NULL);
 960         if (IS_ERR(lquota_type_proc_dir)) {
 961                 CERROR("LProcFS failed in lquota-init\n");
 962                 rc = PTR_ERR(lquota_type_proc_dir);
 963                 return rc;
 964         }
 965
 966         rc = qunit_cache_init();
 967         if (rc)
 968                 return rc;
 969
 970         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 971         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 972 #endif
 973         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
 974         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
 975         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
 976         return 0;
 977 }
 978
 979 static void /*__exit*/ exit_lustre_quota(void)
 980 {
 981         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
 982         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
 983         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
 984 #ifdef HAVE_QUOTA_SUPPORT
 985         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 986         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 987
 988         qunit_cache_cleanup();
 989
 990         if (lquota_type_proc_dir)
 991                 lprocfs_remove(&lquota_type_proc_dir);
 992 #endif
 993 }
 994
 995 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 996 MODULE_DESCRIPTION("Lustre Quota");
 997 MODULE_LICENSE("GPL");
 998
 999 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
1000
1001 #ifdef HAVE_QUOTA_SUPPORT
1002 EXPORT_SYMBOL(mds_quota_interface);
1003 EXPORT_SYMBOL(filter_quota_interface);
1004 #endif
1005 EXPORT_SYMBOL(mdc_quota_interface);
1006 EXPORT_SYMBOL(osc_quota_interface);
1007 EXPORT_SYMBOL(lov_quota_interface);
1008 #endif /* __KERNEL */