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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011 Whamcloud, Inc.
  33  *
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  */
  39
  40 #ifndef EXPORT_SYMTAB
  41 # define EXPORT_SYMTAB
  42 #endif
  43 #define DEBUG_SUBSYSTEM S_LQUOTA
  44
  45 #ifdef __KERNEL__
  46 # include <linux/version.h>
  47 # include <linux/module.h>
  48 # include <linux/init.h>
  49 # include <linux/fs.h>
  50 # include <linux/jbd.h>
  51 # include <linux/smp_lock.h>
  52 # include <linux/buffer_head.h>
  53 # include <linux/workqueue.h>
  54 # include <linux/mount.h>
  55 #else /* __KERNEL__ */
  56 # include <liblustre.h>
  57 #endif
  58
  59 #include <obd_class.h>
  60 #include <lustre_mds.h>
  61 #include <lustre_dlm.h>
  62 #include <lustre_cfg.h>
  63 #include <obd_ost.h>
  64 #include <lustre_fsfilt.h>
  65 #include <lustre_quota.h>
  66 #include <lprocfs_status.h>
  67 #include "quota_internal.h"
  68
  69 #ifdef __KERNEL__
  70
  71 #ifdef HAVE_QUOTA_SUPPORT
  72
  73 static cfs_time_t last_print = 0;
  74 static cfs_spinlock_t last_print_lock = CFS_SPIN_LOCK_UNLOCKED;
  75
  76 static int filter_quota_setup(struct obd_device *obd)
  77 {
  78         int rc = 0;
  79         struct obd_device_target *obt = &obd->u.obt;
  80         ENTRY;
  81
  82         cfs_init_rwsem(&obt->obt_rwsem);
  83         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  84         cfs_sema_init(&obt->obt_quotachecking, 1);
  85         rc = qctxt_init(obd, NULL);
  86         if (rc)
  87                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  88
  89         RETURN(rc);
  90 }
  91
  92 static int filter_quota_cleanup(struct obd_device *obd)
  93 {
  94         ENTRY;
  95         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  96         RETURN(0);
  97 }
  98
  99 static int filter_quota_setinfo(struct obd_device *obd, void *data)
 100 {
 101         struct obd_export *exp = data;
 102         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 103         struct obd_import *imp = exp->exp_imp_reverse;
 104         ENTRY;
 105
 106         LASSERT(imp != NULL);
 107
 108         /* setup the quota context import */
 109         cfs_spin_lock(&qctxt->lqc_lock);
 110         if (qctxt->lqc_import != NULL) {
 111                 cfs_spin_unlock(&qctxt->lqc_lock);
 112                 if (qctxt->lqc_import == imp)
 113                         CDEBUG(D_WARNING, "%s: lqc_import(%p) of obd(%p) was "
 114                                "activated already.\n", obd->obd_name, imp, obd);
 115                 else
 116                         CERROR("%s: lqc_import(%p:%p) of obd(%p) was "
 117                                "activated by others.\n", obd->obd_name,
 118                                qctxt->lqc_import, imp, obd);
 119         } else {
 120                 qctxt->lqc_import = imp;
 121                 /* make imp's connect flags equal relative exp's connect flags
 122                  * adding it to avoid the scan export list */
 123                 imp->imp_connect_data.ocd_connect_flags |=
 124                                 (exp->exp_connect_flags &
 125                                  (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 126                 cfs_spin_unlock(&qctxt->lqc_lock);
 127                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated "
 128                        "now.\n", obd->obd_name, imp, obd);
 129
 130                 cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 131                 /* start quota slave recovery thread. (release high limits) */
 132                 qslave_start_recovery(obd, qctxt);
 133         }
 134         RETURN(0);
 135 }
 136
 137 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 138 {
 139         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 140         struct obd_import *imp = exp->exp_imp_reverse;
 141         ENTRY;
 142
 143         /* lquota may be not set up before destroying export, b=14896 */
 144         if (!obd->obd_set_up)
 145                 RETURN(0);
 146
 147         if (unlikely(imp == NULL))
 148                 RETURN(0);
 149
 150         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 151          * should be invalid b=12374 */
 152         cfs_spin_lock(&qctxt->lqc_lock);
 153         if (qctxt->lqc_import == imp) {
 154                 qctxt->lqc_import = NULL;
 155                 cfs_spin_unlock(&qctxt->lqc_lock);
 156                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is invalid now.\n",
 157                        obd->obd_name, imp, obd);
 158                 ptlrpc_cleanup_imp(imp);
 159                 dqacq_interrupt(qctxt);
 160         } else {
 161                 cfs_spin_unlock(&qctxt->lqc_lock);
 162         }
 163         RETURN(0);
 164 }
 165
 166 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 167 {
 168         ENTRY;
 169
 170         if (!ll_sb_any_quota_active(obd->u.obt.obt_sb))
 171                 RETURN(0);
 172
 173         if (ignore) {
 174                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 175                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 176         } else {
 177                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 178         }
 179
 180         RETURN(0);
 181 }
 182
 183 #define GET_OA_ID(flag, oa) (flag == USRQUOTA ? oa->o_uid : oa->o_gid)
 184 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 185 {
 186         struct obd_device_target *obt = &obd->u.obt;
 187         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 188         int err, cnt, rc = 0;
 189         struct obd_quotactl *oqctl;
 190         ENTRY;
 191
 192         if (!ll_sb_any_quota_active(obt->obt_sb))
 193                 RETURN(0);
 194
 195         OBD_ALLOC_PTR(oqctl);
 196         if (!oqctl)
 197                 RETURN(-ENOMEM);
 198
 199         /* set over quota flags for a uid/gid */
 200         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 201         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 202
 203         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 204                 struct lustre_qunit_size *lqs = NULL;
 205
 206                 lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)),
 207                                        qctxt, 0);
 208                 if (IS_ERR(lqs)) {
 209                         rc = PTR_ERR(lqs);
 210                         CDEBUG(D_QUOTA, "search lqs for %s %d failed, "
 211                                "(rc = %d)\n",
 212                                cnt == USRQUOTA ? "user" : "group",
 213                                GET_OA_ID(cnt, oa), rc);
 214                         break;
 215                 } else if (lqs == NULL) {
 216                         /* continue to check group quota if the file's owner
 217                          * doesn't have quota limit. LU-530 */
 218                         continue;
 219                 } else {
 220                         cfs_spin_lock(&lqs->lqs_lock);
 221                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 222                                 oa->o_flags |= (cnt == USRQUOTA) ?
 223                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 224                                 cfs_spin_unlock(&lqs->lqs_lock);
 225                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 226                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 227                                        qctxt->lqc_sync_blk);
 228                                 /* this is for quota_search_lqs */
 229                                 lqs_putref(lqs);
 230                                 continue;
 231                         }
 232                         cfs_spin_unlock(&lqs->lqs_lock);
 233                         /* this is for quota_search_lqs */
 234                         lqs_putref(lqs);
 235                 }
 236
 237                 memset(oqctl, 0, sizeof(*oqctl));
 238
 239                 oqctl->qc_cmd = Q_GETQUOTA;
 240                 oqctl->qc_type = cnt;
 241                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 242                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 243                 if (err) {
 244                         if (!rc)
 245                                 rc = err;
 246                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 247                                                              OBD_MD_FLGRPQUOTA);
 248                         CDEBUG(D_QUOTA, "fsfilt getquota for %s %d failed, "
 249                                "(rc = %d)\n",
 250                                cnt == USRQUOTA ? "user" : "group",
 251                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid, err);
 252                         continue;
 253                 }
 254
 255                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 256                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 257                     oqctl->qc_dqblk.dqb_bhardlimit)) {
 258                         oa->o_flags |= (cnt == USRQUOTA) ?
 259                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 260                         CDEBUG(D_QUOTA, "out of quota for %s %d\n",
 261                                cnt == USRQUOTA ? "user" : "group",
 262                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid);
 263                 }
 264         }
 265         OBD_FREE_PTR(oqctl);
 266         RETURN(rc);
 267 }
 268
 269 /**
 270  * check whether the left quota of certain uid and gid can satisfy a block_write
 271  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 272  */
 273 static int quota_check_common(struct obd_device *obd, const unsigned int id[],
 274                               int pending[], int count, int cycle, int isblk,
 275                               struct inode *inode, int frags)
 276 {
 277         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 278         int i;
 279         struct qunit_data qdata[MAXQUOTAS];
 280         int mb = 0;
 281         int rc = 0, rc2[2] = { 0, 0 };
 282         ENTRY;
 283
 284         cfs_spin_lock(&qctxt->lqc_lock);
 285         if (!qctxt->lqc_valid){
 286                 cfs_spin_unlock(&qctxt->lqc_lock);
 287                 RETURN(rc);
 288         }
 289         cfs_spin_unlock(&qctxt->lqc_lock);
 290
 291         for (i = 0; i < MAXQUOTAS; i++) {
 292                 struct lustre_qunit_size *lqs = NULL;
 293
 294                 qdata[i].qd_id = id[i];
 295                 qdata[i].qd_flags = i;
 296                 if (isblk)
 297                         QDATA_SET_BLK(&qdata[i]);
 298                 qdata[i].qd_count = 0;
 299
 300                 /* ignore root user */
 301                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 302                         continue;
 303
 304                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 305                 if (lqs == NULL || IS_ERR(lqs))
 306                         continue;
 307
 308                 if (IS_ERR(lqs)) {
 309                         CERROR("can not find lqs for check_common: "
 310                                "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n",
 311                                id[i], i % 2 ? 'g': 'u', isblk, count,
 312                                PTR_ERR(lqs));
 313                         RETURN(PTR_ERR(lqs));
 314                 }
 315
 316                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 317                 cfs_spin_lock(&lqs->lqs_lock);
 318                 if (!cycle) {
 319                         if (isblk) {
 320                                 pending[i] = count * CFS_PAGE_SIZE;
 321                                 /* in order to complete this write, we need extra
 322                                  * meta blocks. This function can get it through
 323                                  * data needed to be written b=16542 */
 324                                 if (inode) {
 325                                         mb = pending[i];
 326                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 327                                                              &mb, inode,
 328                                                              frags);
 329                                         if (rc)
 330                                                 CERROR("%s: can't get extra "
 331                                                        "meta blocks\n",
 332                                                        obd->obd_name);
 333                                         else
 334                                                 pending[i] += mb;
 335                                 }
 336                                 LASSERTF(pending[i] >= 0, "pending is not valid"
 337                                          ", count=%d, mb=%d\n", count, mb);
 338                                 lqs->lqs_bwrite_pending += pending[i];
 339                         } else {
 340                                 pending[i] = count;
 341                                 lqs->lqs_iwrite_pending += pending[i];
 342                         }
 343                 }
 344
 345                 /* if xx_rec < 0, that means quota are releasing,
 346                  * and it may return before we use quota. So if
 347                  * we find this situation, we assuming it has
 348                  * returned b=18491 */
 349                 if (isblk && lqs->lqs_blk_rec < 0) {
 350                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 351                                 qdata[i].qd_count = 0;
 352                         else
 353                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 354                 }
 355                 if (!isblk && lqs->lqs_ino_rec < 0) {
 356                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 357                                 qdata[i].qd_count = 0;
 358                         else
 359                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 360                 }
 361
 362                 CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]"
 363                        " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]"
 364                        " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count,
 365                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 366                        qdata[i].qd_count, mb, pending[i]);
 367                 if (rc2[i] == QUOTA_RET_OK) {
 368                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 369                                 rc2[i] = QUOTA_RET_ACQUOTA;
 370                         if (!isblk && qdata[i].qd_count <
 371                             lqs->lqs_iwrite_pending)
 372                                 rc2[i] = QUOTA_RET_ACQUOTA;
 373                 }
 374
 375                 cfs_spin_unlock(&lqs->lqs_lock);
 376
 377                 if (lqs->lqs_blk_rec  < 0 &&
 378                     qdata[i].qd_count <
 379                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 380                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 381
 382                 /* When cycle is zero, lqs_*_pending will be changed. We will
 383                  * get reference of the lqs here and put reference of lqs in
 384                  * quota_pending_commit b=14784 */
 385                 if (!cycle)
 386                         lqs_getref(lqs);
 387
 388                 /* this is for quota_search_lqs */
 389                 lqs_putref(lqs);
 390         }
 391
 392         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 393                 RETURN(QUOTA_RET_ACQUOTA);
 394         else
 395                 RETURN(rc);
 396 }
 397
 398 int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag)
 399 {
 400         struct lustre_qunit_size *lqs;
 401         int i, q_set = 0;
 402
 403         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 404                 RETURN(0);
 405
 406         for (i = 0; i < MAXQUOTAS; i++) {
 407                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 408                                        &obd->u.obt.obt_qctxt, 0);
 409                 if (lqs && !IS_ERR(lqs)) {
 410                         if (lqs->lqs_flags & flag)
 411                                 q_set = 1;
 412                         lqs_putref(lqs);
 413                 }
 414         }
 415
 416         return q_set;
 417 }
 418
 419 static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
 420                                 const unsigned int id[], int pending[],
 421                                 int count, quota_acquire acquire,
 422                                 struct obd_trans_info *oti, int isblk,
 423                                 struct inode *inode, int frags)
 424 {
 425         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 426         struct timeval work_start;
 427         struct timeval work_end;
 428         long timediff;
 429         struct l_wait_info lwi = { 0 };
 430         int rc = 0, cycle = 0, count_err = 1;
 431         ENTRY;
 432
 433         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 434                 RETURN(0);
 435
 436         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 437                 /* If the client has been evicted or if it
 438                  * timed out and tried to reconnect already,
 439                  * abort the request immediately */
 440                 RETURN(-ENOTCONN);
 441
 442         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 443         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 444         /* Unfortunately, if quota master is too busy to handle the
 445          * pre-dqacq in time and quota hash on ost is used up, we
 446          * have to wait for the completion of in flight dqacq/dqrel,
 447          * in order to get enough quota for write b=12588 */
 448         cfs_gettimeofday(&work_start);
 449         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 450                                         inode, frags)) &
 451                QUOTA_RET_ACQUOTA) {
 452
 453                 cfs_spin_lock(&qctxt->lqc_lock);
 454                 if (!qctxt->lqc_import && oti) {
 455                         cfs_spin_unlock(&qctxt->lqc_lock);
 456                         LASSERT(oti->oti_thread);
 457                         /* The recovery thread doesn't have watchdog
 458                          * attached. LU-369 */
 459                         if (oti->oti_thread->t_watchdog)
 460                                 lc_watchdog_disable(oti->oti_thread->\
 461                                                 t_watchdog);
 462                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 463                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 464                                      &lwi);
 465                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 466                         if (oti->oti_thread->t_watchdog)
 467                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 468                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 469                 } else {
 470                         cfs_spin_unlock(&qctxt->lqc_lock);
 471                 }
 472
 473                 cycle++;
 474                 if (isblk)
 475                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 476                 /* after acquire(), we should run quota_check_common again
 477                  * so that we confirm there are enough quota to finish write */
 478                 rc = acquire(obd, id, oti, isblk);
 479
 480                 /* please reference to dqacq_completion for the below */
 481                 /* a new request is finished, try again */
 482                 if (rc == QUOTA_REQ_RETURNED) {
 483                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 484                         continue;
 485                 }
 486
 487                 /* it is out of quota already */
 488                 if (rc == -EDQUOT) {
 489                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 490                         break;
 491                 }
 492
 493                 /* Related quota has been disabled by master, but enabled by
 494                  * slave, do not try again. */
 495                 if (unlikely(rc == -ESRCH)) {
 496                         CERROR("mismatched quota configuration, stop try.\n");
 497                         break;
 498                 }
 499
 500                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 501                         /* The client has been evicted or tried to
 502                          * to reconnect already, abort the request */
 503                         RETURN(-ENOTCONN);
 504
 505                 /* -EBUSY and others, wait a second and try again */
 506                 if (rc < 0) {
 507                         cfs_waitq_t        waitq;
 508                         struct l_wait_info lwi;
 509
 510                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 511                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 512                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 513                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 514                                count_err++);
 515
 516                         cfs_waitq_init(&waitq);
 517                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 518                                           NULL);
 519                         l_wait_event(waitq, 0, &lwi);
 520                 }
 521
 522                 if (rc < 0 || cycle % 10 == 0) {
 523                         cfs_spin_lock(&last_print_lock);
 524                         if (last_print == 0 ||
 525                             cfs_time_before((last_print + cfs_time_seconds(30)),
 526                                             cfs_time_current())) {
 527                                 last_print = cfs_time_current();
 528                                 cfs_spin_unlock(&last_print_lock);
 529                                 CWARN("still haven't managed to acquire quota "
 530                                       "space from the quota master after %d "
 531                                       "retries (err=%d, rc=%d)\n",
 532                                       cycle, count_err - 1, rc);
 533                         } else {
 534                                 cfs_spin_unlock(&last_print_lock);
 535                         }
 536                 }
 537
 538                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 539                        cycle);
 540         }
 541         cfs_gettimeofday(&work_end);
 542         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 543         lprocfs_counter_add(qctxt->lqc_stats,
 544                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 545                                     LQUOTA_WAIT_FOR_CHK_INO,
 546                             timediff);
 547
 548         if (rc > 0)
 549                 rc = 0;
 550         RETURN(rc);
 551 }
 552
 553 /**
 554  * when a block_write or inode_create rpc is finished, adjust the record for
 555  * pending blocks and inodes
 556  */
 557 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 558                                 int pending[], int isblk)
 559 {
 560         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 561         struct timeval work_start;
 562         struct timeval work_end;
 563         long timediff;
 564         int i;
 565         struct qunit_data qdata[MAXQUOTAS];
 566         ENTRY;
 567
 568         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 569         CLASSERT(MAXQUOTAS < 4);
 570         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 571                 RETURN(0);
 572
 573         cfs_gettimeofday(&work_start);
 574         for (i = 0; i < MAXQUOTAS; i++) {
 575                 struct lustre_qunit_size *lqs = NULL;
 576
 577                 LASSERT(pending[i] >= 0);
 578                 if (pending[i] == 0)
 579                         continue;
 580
 581                 qdata[i].qd_id = id[i];
 582                 qdata[i].qd_flags = i;
 583                 if (isblk)
 584                         QDATA_SET_BLK(&qdata[i]);
 585                 qdata[i].qd_count = 0;
 586
 587                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 588                         continue;
 589
 590                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 591                 if (lqs == NULL || IS_ERR(lqs)) {
 592                         CERROR("can not find lqs for pending_commit: "
 593                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 594                                "maybe cause unexpected lqs refcount error!\n",
 595                                id[i], i ? 'g': 'u', pending[i], isblk,
 596                                lqs ? PTR_ERR(lqs) : -1);
 597                         continue;
 598                 }
 599
 600                 cfs_spin_lock(&lqs->lqs_lock);
 601                 if (isblk) {
 602                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 603                                  "there are too many blocks! [id %u] [%c] "
 604                                  "[bwrite_pending %lu] [pending %u]\n",
 605                                  id[i], i % 2 ? 'g' : 'u',
 606                                  lqs->lqs_bwrite_pending, pending[i]);
 607
 608                         lqs->lqs_bwrite_pending -= pending[i];
 609                 } else {
 610                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 611                                 "there are too many files! [id %u] [%c] "
 612                                 "[iwrite_pending %lu] [pending %u]\n",
 613                                 id[i], i % 2 ? 'g' : 'u',
 614                                 lqs->lqs_iwrite_pending, pending[i]);
 615
 616                         lqs->lqs_iwrite_pending -= pending[i];
 617                 }
 618                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 619                        obd->obd_name,
 620                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 621                        i, pending[i], isblk);
 622                 cfs_spin_unlock(&lqs->lqs_lock);
 623
 624                 /* for quota_search_lqs in pending_commit */
 625                 lqs_putref(lqs);
 626                 /* for quota_search_lqs in quota_check */
 627                 lqs_putref(lqs);
 628         }
 629         cfs_gettimeofday(&work_end);
 630         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 631         lprocfs_counter_add(qctxt->lqc_stats,
 632                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 633                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 634                             timediff);
 635
 636         RETURN(0);
 637 }
 638
 639 static int mds_quota_init(void)
 640 {
 641         return lustre_dquot_init();
 642 }
 643
 644 static int mds_quota_exit(void)
 645 {
 646         lustre_dquot_exit();
 647         return 0;
 648 }
 649
 650 static int mds_quota_setup(struct obd_device *obd)
 651 {
 652         struct obd_device_target *obt = &obd->u.obt;
 653         struct mds_obd *mds = &obd->u.mds;
 654         int rc;
 655         ENTRY;
 656
 657         if (unlikely(mds->mds_quota)) {
 658                 CWARN("try to reinitialize quota context!\n");
 659                 RETURN(0);
 660         }
 661
 662         cfs_init_rwsem(&obt->obt_rwsem);
 663         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 664         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 665         cfs_sema_init(&obt->obt_quotachecking, 1);
 666         /* initialize quota master and quota context */
 667         cfs_init_rwsem(&mds->mds_qonoff_sem);
 668         rc = qctxt_init(obd, dqacq_handler);
 669         if (rc) {
 670                 CERROR("%s: initialize quota context failed! (rc:%d)\n",
 671                        obd->obd_name, rc);
 672                 RETURN(rc);
 673         }
 674         mds->mds_quota = 1;
 675         RETURN(rc);
 676 }
 677
 678 static int mds_quota_cleanup(struct obd_device *obd)
 679 {
 680         ENTRY;
 681         if (unlikely(!obd->u.mds.mds_quota))
 682                 RETURN(0);
 683
 684         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 685         RETURN(0);
 686 }
 687
 688 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 689 {
 690         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 691         ENTRY;
 692
 693         if (unlikely(!obd->u.mds.mds_quota))
 694                 RETURN(0);
 695
 696         if (data != NULL)
 697                 QUOTA_MASTER_READY(qctxt);
 698         else
 699                 QUOTA_MASTER_UNREADY(qctxt);
 700         RETURN(0);
 701 }
 702
 703 static int mds_quota_fs_cleanup(struct obd_device *obd)
 704 {
 705         struct mds_obd *mds = &obd->u.mds;
 706         struct obd_quotactl oqctl;
 707         ENTRY;
 708
 709         if (unlikely(!mds->mds_quota))
 710                 RETURN(0);
 711
 712         mds->mds_quota = 0;
 713         memset(&oqctl, 0, sizeof(oqctl));
 714         oqctl.qc_type = UGQUOTA;
 715
 716         cfs_down_write(&mds->mds_qonoff_sem);
 717         mds_admin_quota_off(obd, &oqctl);
 718         cfs_up_write(&mds->mds_qonoff_sem);
 719         RETURN(0);
 720 }
 721
 722 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 723                                 struct obd_trans_info *oti, int isblk)
 724 {
 725         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 726         int rc;
 727         ENTRY;
 728
 729         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 730         RETURN(rc);
 731 }
 732
 733 #endif /* HAVE_QUOTA_SUPPORT */
 734 #endif /* __KERNEL__ */
 735
 736 struct osc_quota_info {
 737         /** linkage for quota hash table */
 738         cfs_hlist_node_t oqi_hash;
 739         obd_uid          oqi_id;
 740 };
 741
 742 /* SLAB cache for client quota context */
 743 cfs_mem_cache_t *qinfo_cachep = NULL;
 744
 745 static inline struct osc_quota_info *osc_oqi_alloc(obd_uid id)
 746 {
 747         struct osc_quota_info *oqi;
 748
 749         OBD_SLAB_ALLOC_PTR(oqi, qinfo_cachep);
 750         if (oqi != NULL)
 751                 oqi->oqi_id = id;
 752
 753         return oqi;
 754 }
 755
 756 int osc_quota_chkdq(struct client_obd *cli, const unsigned int qid[])
 757 {
 758         int type;
 759         ENTRY;
 760
 761         for (type = 0; type < MAXQUOTAS; type++) {
 762                 struct osc_quota_info *oqi;
 763
 764                 oqi = cfs_hash_lookup(cli->cl_quota_hash[type], &qid[type]);
 765                 if (oqi) {
 766                         obd_uid id = oqi->oqi_id;
 767
 768                         LASSERTF(id == qid[type],
 769                                  "The ids don't match %u != %u\n",
 770                                  id, qid[type]);
 771
 772                         /* the slot is busy, the user is about to run out of
 773                          * quota space on this OST */
 774                         CDEBUG(D_QUOTA, "chkdq found noquota for %s %d\n",
 775                                type == USRQUOTA ? "user" : "grout", qid[type]);
 776                         RETURN(NO_QUOTA);
 777                 }
 778         }
 779
 780         RETURN(QUOTA_OK);
 781 }
 782
 783 #define MD_QUOTA_FLAG(type) ((type == USRQUOTA) ? OBD_MD_FLUSRQUOTA \
 784                                                 : OBD_MD_FLGRPQUOTA)
 785 #define FL_QUOTA_FLAG(type) ((type == USRQUOTA) ? OBD_FL_NO_USRQUOTA \
 786                                                 : OBD_FL_NO_GRPQUOTA)
 787
 788 int osc_quota_setdq(struct client_obd *cli, const unsigned int qid[],
 789                     obd_flag valid, obd_flag flags)
 790 {
 791         int type;
 792         int rc = 0;
 793         ENTRY;
 794
 795         if ((valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) == 0)
 796                 RETURN(0);
 797
 798         for (type = 0; type < MAXQUOTAS; type++) {
 799                 struct osc_quota_info *oqi;
 800
 801                 if ((valid & MD_QUOTA_FLAG(type)) == 0)
 802                         continue;
 803
 804                 /* lookup the ID in the per-type hash table */
 805                 oqi = cfs_hash_lookup(cli->cl_quota_hash[type], &qid[type]);
 806                 if ((flags & FL_QUOTA_FLAG(type)) != 0) {
 807                         /* This ID is getting close to its quota limit, let's
 808                          * switch to sync I/O */
 809                         if (oqi != NULL)
 810                                 continue;
 811
 812                         oqi = osc_oqi_alloc(qid[type]);
 813                         if (oqi == NULL) {
 814                                 rc = -ENOMEM;
 815                                 break;
 816                         }
 817
 818                         rc = cfs_hash_add_unique(cli->cl_quota_hash[type],
 819                                                  &qid[type], &oqi->oqi_hash);
 820                         /* race with others? */
 821                         if (rc == -EALREADY) {
 822                                 rc = 0;
 823                                 OBD_SLAB_FREE_PTR(oqi, qinfo_cachep);
 824                         }
 825
 826                         CDEBUG(D_QUOTA, "%s: setdq to insert for %s %d (%d)\n",
 827                                cli->cl_import->imp_obd->obd_name,
 828                                type == USRQUOTA ? "user" : "group",
 829                                qid[type], rc);
 830                 } else {
 831                         /* This ID is now off the hook, let's remove it from
 832                          * the hash table */
 833                         if (oqi == NULL)
 834                                 continue;
 835
 836                         oqi = cfs_hash_del_key(cli->cl_quota_hash[type],
 837                                                &qid[type]);
 838                         if (oqi)
 839                                 OBD_SLAB_FREE_PTR(oqi, qinfo_cachep);
 840
 841                         CDEBUG(D_QUOTA, "%s: setdq to remove for %s %d (%p)\n",
 842                                cli->cl_import->imp_obd->obd_name,
 843                                type == USRQUOTA ? "user" : "group",
 844                                qid[type], oqi);
 845                 }
 846         }
 847
 848         RETURN(rc);
 849 }
 850
 851 /*
 852  * Hash operations for uid/gid <-> osc_quota_info
 853  */
 854 static unsigned
 855 oqi_hashfn(cfs_hash_t *hs, const void *key, unsigned mask)
 856 {
 857         return cfs_hash_u32_hash(*((__u32*)key), mask);
 858 }
 859
 860 static int
 861 oqi_keycmp(const void *key, cfs_hlist_node_t *hnode)
 862 {
 863         struct osc_quota_info *oqi;
 864         obd_uid uid;
 865
 866         LASSERT(key != NULL);
 867         uid = *((obd_uid*)key);
 868         oqi = cfs_hlist_entry(hnode, struct osc_quota_info, oqi_hash);
 869
 870         return uid == oqi->oqi_id;
 871 }
 872
 873 static void *
 874 oqi_key(cfs_hlist_node_t *hnode)
 875 {
 876         struct osc_quota_info *oqi;
 877         oqi = cfs_hlist_entry(hnode, struct osc_quota_info, oqi_hash);
 878         return &oqi->oqi_id;
 879 }
 880
 881 static void *
 882 oqi_object(cfs_hlist_node_t *hnode)
 883 {
 884         return cfs_hlist_entry(hnode, struct osc_quota_info, oqi_hash);
 885 }
 886
 887 static void
 888 oqi_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 889 {
 890 }
 891
 892 static void
 893 oqi_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 894 {
 895 }
 896
 897 static void
 898 oqi_exit(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 899 {
 900         struct osc_quota_info *oqi;
 901
 902         oqi = cfs_hlist_entry(hnode, struct osc_quota_info, oqi_hash);
 903
 904         OBD_SLAB_FREE_PTR(oqi, qinfo_cachep);
 905 }
 906
 907 #define HASH_QUOTA_BKT_BITS 5
 908 #define HASH_QUOTA_CUR_BITS 5
 909 #define HASH_QUOTA_MAX_BITS 15
 910
 911 static cfs_hash_ops_t quota_hash_ops = {
 912         .hs_hash       = oqi_hashfn,
 913         .hs_keycmp     = oqi_keycmp,
 914         .hs_key        = oqi_key,
 915         .hs_object     = oqi_object,
 916         .hs_get        = oqi_get,
 917         .hs_put_locked = oqi_put_locked,
 918         .hs_exit       = oqi_exit,
 919 };
 920
 921 int osc_quota_setup(struct obd_device *obd)
 922 {
 923         struct client_obd *cli = &obd->u.cli;
 924         int i, type;
 925         ENTRY;
 926
 927         for (type = 0; type < MAXQUOTAS; type++) {
 928                 cli->cl_quota_hash[type] = cfs_hash_create("QUOTA_HASH",
 929                                                            HASH_QUOTA_CUR_BITS,
 930                                                            HASH_QUOTA_MAX_BITS,
 931                                                            HASH_QUOTA_BKT_BITS,
 932                                                            0,
 933                                                            CFS_HASH_MIN_THETA,
 934                                                            CFS_HASH_MAX_THETA,
 935                                                            &quota_hash_ops,
 936                                                            CFS_HASH_DEFAULT);
 937                 if (cli->cl_quota_hash[type] == NULL)
 938                         break;
 939         }
 940
 941         if (type == MAXQUOTAS)
 942                 RETURN(0);
 943
 944         for (i = 0; i < type; i++)
 945                 cfs_hash_putref(cli->cl_quota_hash[i]);
 946
 947         RETURN(-ENOMEM);
 948 }
 949
 950 int osc_quota_cleanup(struct obd_device *obd)
 951 {
 952         struct client_obd *cli = &obd->u.cli;
 953         int type;
 954         ENTRY;
 955
 956         for (type = 0; type < MAXQUOTAS; type++)
 957                 cfs_hash_putref(cli->cl_quota_hash[type]);
 958
 959         RETURN(0);
 960 }
 961
 962 int osc_quota_init(void)
 963 {
 964         ENTRY;
 965
 966         LASSERT(qinfo_cachep == NULL);
 967         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 968                                             sizeof(struct osc_quota_info),
 969                                             0, 0);
 970         if (!qinfo_cachep)
 971                 RETURN(-ENOMEM);
 972
 973         RETURN(0);
 974 }
 975
 976 int osc_quota_exit(void)
 977 {
 978         int rc;
 979         ENTRY;
 980
 981         rc = cfs_mem_cache_destroy(qinfo_cachep);
 982         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 983         qinfo_cachep = NULL;
 984
 985         RETURN(0);
 986 }
 987
 988
 989 #ifdef __KERNEL__
 990 #ifdef HAVE_QUOTA_SUPPORT
 991 quota_interface_t mds_quota_interface = {
 992         .quota_init     = mds_quota_init,
 993         .quota_exit     = mds_quota_exit,
 994         .quota_setup    = mds_quota_setup,
 995         .quota_cleanup  = mds_quota_cleanup,
 996         .quota_check    = target_quota_check,
 997         .quota_ctl      = mds_quota_ctl,
 998         .quota_setinfo  = mds_quota_setinfo,
 999         .quota_fs_cleanup = mds_quota_fs_cleanup,
1000         .quota_recovery = mds_quota_recovery,
1001         .quota_adjust   = mds_quota_adjust,
1002         .quota_chkquota = quota_chk_acq_common,
1003         .quota_acquire  = quota_acquire_common,
1004         .quota_pending_commit = quota_pending_commit,
1005 };
1006
1007 quota_interface_t filter_quota_interface = {
1008         .quota_setup    = filter_quota_setup,
1009         .quota_cleanup  = filter_quota_cleanup,
1010         .quota_check    = target_quota_check,
1011         .quota_ctl      = filter_quota_ctl,
1012         .quota_setinfo  = filter_quota_setinfo,
1013         .quota_clearinfo = filter_quota_clearinfo,
1014         .quota_enforce  = filter_quota_enforce,
1015         .quota_getflag  = filter_quota_getflag,
1016         .quota_acquire  = quota_acquire_common,
1017         .quota_adjust   = filter_quota_adjust,
1018         .quota_chkquota = quota_chk_acq_common,
1019         .quota_adjust_qunit   = filter_quota_adjust_qunit,
1020         .quota_pending_commit = quota_pending_commit,
1021 };
1022 #endif
1023 #endif /* __KERNEL__ */
1024
1025 quota_interface_t mdc_quota_interface = {
1026         .quota_ctl      = client_quota_ctl,
1027         .quota_check    = client_quota_check,
1028         .quota_poll_check = client_quota_poll_check,
1029 };
1030
1031 quota_interface_t lmv_quota_interface = {
1032         .quota_ctl      = lmv_quota_ctl,
1033         .quota_check    = lmv_quota_check,
1034 };
1035
1036 quota_interface_t osc_quota_interface = {
1037         .quota_ctl      = client_quota_ctl,
1038         .quota_check    = client_quota_check,
1039         .quota_poll_check = client_quota_poll_check,
1040         .quota_init     = osc_quota_init,
1041         .quota_exit     = osc_quota_exit,
1042         .quota_setup    = osc_quota_setup,
1043         .quota_chkdq    = osc_quota_chkdq,
1044         .quota_setdq    = osc_quota_setdq,
1045         .quota_cleanup  = osc_quota_cleanup,
1046         .quota_adjust_qunit = client_quota_adjust_qunit,
1047 };
1048
1049 quota_interface_t lov_quota_interface = {
1050         .quota_ctl      = lov_quota_ctl,
1051         .quota_check    = lov_quota_check,
1052         .quota_adjust_qunit = lov_quota_adjust_qunit,
1053 };
1054
1055 #ifdef __KERNEL__
1056
1057 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
1058
1059 static int __init init_lustre_quota(void)
1060 {
1061 #ifdef HAVE_QUOTA_SUPPORT
1062         int rc = 0;
1063
1064         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
1065                                                 proc_lustre_root,
1066                                                 NULL, NULL);
1067         if (IS_ERR(lquota_type_proc_dir)) {
1068                 CERROR("LProcFS failed in lquota-init\n");
1069                 rc = PTR_ERR(lquota_type_proc_dir);
1070                 return rc;
1071         }
1072
1073         rc = qunit_cache_init();
1074         if (rc)
1075                 return rc;
1076
1077         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
1078         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
1079 #endif
1080         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
1081         PORTAL_SYMBOL_REGISTER(lmv_quota_interface);
1082         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
1083         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
1084         return 0;
1085 }
1086
1087 static void /*__exit*/ exit_lustre_quota(void)
1088 {
1089         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
1090         PORTAL_SYMBOL_UNREGISTER(lmv_quota_interface);
1091         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
1092         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
1093 #ifdef HAVE_QUOTA_SUPPORT
1094         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
1095         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
1096
1097         qunit_cache_cleanup();
1098
1099         if (lquota_type_proc_dir)
1100                 lprocfs_remove(&lquota_type_proc_dir);
1101 #endif
1102 }
1103
1104 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1105 MODULE_DESCRIPTION("Lustre Quota");
1106 MODULE_LICENSE("GPL");
1107
1108 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
1109
1110 #ifdef HAVE_QUOTA_SUPPORT
1111 EXPORT_SYMBOL(mds_quota_interface);
1112 EXPORT_SYMBOL(filter_quota_interface);
1113 #endif
1114 EXPORT_SYMBOL(mdc_quota_interface);
1115 EXPORT_SYMBOL(lmv_quota_interface);
1116 EXPORT_SYMBOL(osc_quota_interface);
1117 EXPORT_SYMBOL(lov_quota_interface);
1118 #endif /* __KERNEL */