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 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_LQUOTA
  41
  42 #ifdef __KERNEL__
  43 # include <linux/version.h>
  44 # include <linux/module.h>
  45 # include <linux/init.h>
  46 # include <linux/fs.h>
  47 # include <linux/jbd.h>
  48 # include <linux/smp_lock.h>
  49 # include <linux/buffer_head.h>
  50 # include <linux/workqueue.h>
  51 # include <linux/mount.h>
  52 #else /* __KERNEL__ */
  53 # include <liblustre.h>
  54 #endif
  55
  56 #include <obd_class.h>
  57 #include <lustre_mds.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_cfg.h>
  60 #include <obd_ost.h>
  61 #include <lustre_fsfilt.h>
  62 #include <lustre_quota.h>
  63 #include <lprocfs_status.h>
  64 #include "quota_internal.h"
  65
  66 #ifdef __KERNEL__
  67
  68 #ifdef HAVE_QUOTA_SUPPORT
  69
  70 static cfs_time_t last_print = 0;
  71 static cfs_spinlock_t last_print_lock = CFS_SPIN_LOCK_UNLOCKED;
  72
  73 static int filter_quota_setup(struct obd_device *obd)
  74 {
  75         int rc = 0;
  76         struct obd_device_target *obt = &obd->u.obt;
  77         ENTRY;
  78
  79         cfs_init_rwsem(&obt->obt_rwsem);
  80         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  81         cfs_sema_init(&obt->obt_quotachecking, 1);
  82         rc = qctxt_init(obd, NULL);
  83         if (rc)
  84                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  85
  86         RETURN(rc);
  87 }
  88
  89 static int filter_quota_cleanup(struct obd_device *obd)
  90 {
  91         ENTRY;
  92         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  93         RETURN(0);
  94 }
  95
  96 static int filter_quota_setinfo(struct obd_device *obd, void *data)
  97 {
  98         struct obd_export *exp = data;
  99         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 100         struct obd_import *imp = exp->exp_imp_reverse;
 101         ENTRY;
 102
 103         LASSERT(imp != NULL);
 104
 105         /* setup the quota context import */
 106         cfs_spin_lock(&qctxt->lqc_lock);
 107         if (qctxt->lqc_import != NULL) {
 108                 cfs_spin_unlock(&qctxt->lqc_lock);
 109                 if (qctxt->lqc_import == imp)
 110                         CDEBUG(D_WARNING, "%s: lqc_import(%p) of obd(%p) was "
 111                                "activated already.\n", obd->obd_name, imp, obd);
 112                 else
 113                         CERROR("%s: lqc_import(%p:%p) of obd(%p) was "
 114                                "activated by others.\n", obd->obd_name,
 115                                qctxt->lqc_import, imp, obd);
 116         } else {
 117                 qctxt->lqc_import = imp;
 118                 /* make imp's connect flags equal relative exp's connect flags
 119                  * adding it to avoid the scan export list */
 120                 imp->imp_connect_data.ocd_connect_flags |=
 121                                 (exp->exp_connect_flags &
 122                                  (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 123                 cfs_spin_unlock(&qctxt->lqc_lock);
 124                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated "
 125                        "now.\n", obd->obd_name, imp, obd);
 126
 127                 cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 128                 /* start quota slave recovery thread. (release high limits) */
 129                 qslave_start_recovery(obd, qctxt);
 130         }
 131         RETURN(0);
 132 }
 133
 134 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 135 {
 136         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 137         struct obd_import *imp = exp->exp_imp_reverse;
 138         ENTRY;
 139
 140         /* lquota may be not set up before destroying export, b=14896 */
 141         if (!obd->obd_set_up)
 142                 RETURN(0);
 143
 144         if (unlikely(imp == NULL))
 145                 RETURN(0);
 146
 147         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 148          * should be invalid b=12374 */
 149         cfs_spin_lock(&qctxt->lqc_lock);
 150         if (qctxt->lqc_import == imp) {
 151                 qctxt->lqc_import = NULL;
 152                 cfs_spin_unlock(&qctxt->lqc_lock);
 153                 CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is invalid now.\n",
 154                        obd->obd_name, imp, obd);
 155                 ptlrpc_cleanup_imp(imp);
 156                 dqacq_interrupt(qctxt);
 157         } else {
 158                 cfs_spin_unlock(&qctxt->lqc_lock);
 159         }
 160         RETURN(0);
 161 }
 162
 163 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 164 {
 165         ENTRY;
 166
 167         if (!ll_sb_any_quota_active(obd->u.obt.obt_sb))
 168                 RETURN(0);
 169
 170         if (ignore) {
 171                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 172                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 173         } else {
 174                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 175         }
 176
 177         RETURN(0);
 178 }
 179
 180 #define GET_OA_ID(flag, oa) (flag == USRQUOTA ? oa->o_uid : oa->o_gid)
 181 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 182 {
 183         struct obd_device_target *obt = &obd->u.obt;
 184         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 185         int err, cnt, rc = 0;
 186         struct obd_quotactl *oqctl;
 187         ENTRY;
 188
 189         if (!ll_sb_any_quota_active(obt->obt_sb))
 190                 RETURN(0);
 191
 192         OBD_ALLOC_PTR(oqctl);
 193         if (!oqctl)
 194                 RETURN(-ENOMEM);
 195
 196         /* set over quota flags for a uid/gid */
 197         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 198         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 199
 200         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 201                 struct lustre_qunit_size *lqs = NULL;
 202
 203                 lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)),
 204                                        qctxt, 0);
 205                 if (IS_ERR(lqs)) {
 206                         rc = PTR_ERR(lqs);
 207                         CDEBUG(D_QUOTA, "search lqs for %s %d failed, "
 208                                "(rc = %d)\n",
 209                                cnt == USRQUOTA ? "user" : "group",
 210                                GET_OA_ID(cnt, oa), rc);
 211                         break;
 212                 } else if (lqs == NULL) {
 213                         /* continue to check group quota if the file's owner
 214                          * doesn't have quota limit. LU-530 */
 215                         continue;
 216                 } else {
 217                         cfs_spin_lock(&lqs->lqs_lock);
 218                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 219                                 oa->o_flags |= (cnt == USRQUOTA) ?
 220                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 221                                 cfs_spin_unlock(&lqs->lqs_lock);
 222                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 223                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 224                                        qctxt->lqc_sync_blk);
 225                                 /* this is for quota_search_lqs */
 226                                 lqs_putref(lqs);
 227                                 continue;
 228                         }
 229                         cfs_spin_unlock(&lqs->lqs_lock);
 230                         /* this is for quota_search_lqs */
 231                         lqs_putref(lqs);
 232                 }
 233
 234                 memset(oqctl, 0, sizeof(*oqctl));
 235
 236                 oqctl->qc_cmd = Q_GETQUOTA;
 237                 oqctl->qc_type = cnt;
 238                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 239                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 240                 if (err) {
 241                         if (!rc)
 242                                 rc = err;
 243                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 244                                                              OBD_MD_FLGRPQUOTA);
 245                         CDEBUG(D_QUOTA, "fsfilt getquota for %s %d failed, "
 246                                "(rc = %d)\n",
 247                                cnt == USRQUOTA ? "user" : "group",
 248                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid, err);
 249                         continue;
 250                 }
 251
 252                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 253                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 254                     oqctl->qc_dqblk.dqb_bhardlimit)) {
 255                         oa->o_flags |= (cnt == USRQUOTA) ?
 256                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 257                         CDEBUG(D_QUOTA, "out of quota for %s %d\n",
 258                                cnt == USRQUOTA ? "user" : "group",
 259                                cnt == USRQUOTA ? oa->o_uid : oa->o_gid);
 260                 }
 261         }
 262         OBD_FREE_PTR(oqctl);
 263         RETURN(rc);
 264 }
 265
 266 /**
 267  * check whether the left quota of certain uid and gid can satisfy a block_write
 268  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 269  */
 270 static int quota_check_common(struct obd_device *obd, const unsigned int id[],
 271                               int pending[], int count, int cycle, int isblk,
 272                               struct inode *inode, int frags)
 273 {
 274         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 275         int i;
 276         struct qunit_data qdata[MAXQUOTAS];
 277         int mb = 0;
 278         int rc = 0, rc2[2] = { 0, 0 };
 279         ENTRY;
 280
 281         cfs_spin_lock(&qctxt->lqc_lock);
 282         if (!qctxt->lqc_valid){
 283                 cfs_spin_unlock(&qctxt->lqc_lock);
 284                 RETURN(rc);
 285         }
 286         cfs_spin_unlock(&qctxt->lqc_lock);
 287
 288         for (i = 0; i < MAXQUOTAS; i++) {
 289                 struct lustre_qunit_size *lqs = NULL;
 290
 291                 qdata[i].qd_id = id[i];
 292                 qdata[i].qd_flags = i;
 293                 if (isblk)
 294                         QDATA_SET_BLK(&qdata[i]);
 295                 qdata[i].qd_count = 0;
 296
 297                 /* ignore root user */
 298                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 299                         continue;
 300
 301                 lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0);
 302                 if (lqs == NULL || IS_ERR(lqs))
 303                         continue;
 304
 305                 if (IS_ERR(lqs)) {
 306                         CERROR("can not find lqs for check_common: "
 307                                "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n",
 308                                id[i], i % 2 ? 'g': 'u', isblk, count,
 309                                PTR_ERR(lqs));
 310                         RETURN(PTR_ERR(lqs));
 311                 }
 312
 313                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 314                 cfs_spin_lock(&lqs->lqs_lock);
 315                 if (!cycle) {
 316                         if (isblk) {
 317                                 pending[i] = count * CFS_PAGE_SIZE;
 318                                 /* in order to complete this write, we need extra
 319                                  * meta blocks. This function can get it through
 320                                  * data needed to be written b=16542 */
 321                                 if (inode) {
 322                                         mb = pending[i];
 323                                         rc = fsfilt_get_mblk(obd, qctxt->lqc_sb,
 324                                                              &mb, inode,
 325                                                              frags);
 326                                         if (rc)
 327                                                 CERROR("%s: can't get extra "
 328                                                        "meta blocks\n",
 329                                                        obd->obd_name);
 330                                         else
 331                                                 pending[i] += mb;
 332                                 }
 333                                 lqs->lqs_bwrite_pending += pending[i];
 334                         } else {
 335                                 pending[i] = count;
 336                                 lqs->lqs_iwrite_pending += pending[i];
 337                         }
 338                 }
 339
 340                 /* if xx_rec < 0, that means quota are releasing,
 341                  * and it may return before we use quota. So if
 342                  * we find this situation, we assuming it has
 343                  * returned b=18491 */
 344                 if (isblk && lqs->lqs_blk_rec < 0) {
 345                         if (qdata[i].qd_count < -lqs->lqs_blk_rec)
 346                                 qdata[i].qd_count = 0;
 347                         else
 348                                 qdata[i].qd_count += lqs->lqs_blk_rec;
 349                 }
 350                 if (!isblk && lqs->lqs_ino_rec < 0) {
 351                         if (qdata[i].qd_count < -lqs->lqs_ino_rec)
 352                                 qdata[i].qd_count = 0;
 353                         else
 354                                 qdata[i].qd_count += lqs->lqs_ino_rec;
 355                 }
 356
 357                 CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]"
 358                        " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]"
 359                        " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count,
 360                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 361                        qdata[i].qd_count, mb, pending[i]);
 362                 if (rc2[i] == QUOTA_RET_OK) {
 363                         if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending)
 364                                 rc2[i] = QUOTA_RET_ACQUOTA;
 365                         if (!isblk && qdata[i].qd_count <
 366                             lqs->lqs_iwrite_pending)
 367                                 rc2[i] = QUOTA_RET_ACQUOTA;
 368                 }
 369
 370                 cfs_spin_unlock(&lqs->lqs_lock);
 371
 372                 if (lqs->lqs_blk_rec  < 0 &&
 373                     qdata[i].qd_count <
 374                     lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb)
 375                         OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5);
 376
 377                 /* When cycle is zero, lqs_*_pending will be changed. We will
 378                  * get reference of the lqs here and put reference of lqs in
 379                  * quota_pending_commit b=14784 */
 380                 if (!cycle)
 381                         lqs_getref(lqs);
 382
 383                 /* this is for quota_search_lqs */
 384                 lqs_putref(lqs);
 385         }
 386
 387         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 388                 RETURN(QUOTA_RET_ACQUOTA);
 389         else
 390                 RETURN(rc);
 391 }
 392
 393 int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag)
 394 {
 395         struct lustre_qunit_size *lqs;
 396         int i, q_set = 0;
 397
 398         if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb))
 399                 RETURN(0);
 400
 401         for (i = 0; i < MAXQUOTAS; i++) {
 402                 lqs = quota_search_lqs(LQS_KEY(i, id[i]),
 403                                        &obd->u.obt.obt_qctxt, 0);
 404                 if (lqs && !IS_ERR(lqs)) {
 405                         if (lqs->lqs_flags & flag)
 406                                 q_set = 1;
 407                         lqs_putref(lqs);
 408                 }
 409         }
 410
 411         return q_set;
 412 }
 413
 414 static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
 415                                 const unsigned int id[], int pending[],
 416                                 int count, quota_acquire acquire,
 417                                 struct obd_trans_info *oti, int isblk,
 418                                 struct inode *inode, int frags)
 419 {
 420         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 421         struct timeval work_start;
 422         struct timeval work_end;
 423         long timediff;
 424         struct l_wait_info lwi = { 0 };
 425         int rc = 0, cycle = 0, count_err = 1;
 426         ENTRY;
 427
 428         if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
 429                 RETURN(0);
 430
 431         if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 432                 /* If the client has been evicted or if it
 433                  * timed out and tried to reconnect already,
 434                  * abort the request immediately */
 435                 RETURN(-ENOTCONN);
 436
 437         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 438         pending[USRQUOTA] = pending[GRPQUOTA] = 0;
 439         /* Unfortunately, if quota master is too busy to handle the
 440          * pre-dqacq in time and quota hash on ost is used up, we
 441          * have to wait for the completion of in flight dqacq/dqrel,
 442          * in order to get enough quota for write b=12588 */
 443         cfs_gettimeofday(&work_start);
 444         while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
 445                                         inode, frags)) &
 446                QUOTA_RET_ACQUOTA) {
 447
 448                 cfs_spin_lock(&qctxt->lqc_lock);
 449                 if (!qctxt->lqc_import && oti) {
 450                         cfs_spin_unlock(&qctxt->lqc_lock);
 451                         LASSERT(oti->oti_thread);
 452                         /* The recovery thread doesn't have watchdog
 453                          * attached. LU-369 */
 454                         if (oti->oti_thread->t_watchdog)
 455                                 lc_watchdog_disable(oti->oti_thread->\
 456                                                 t_watchdog);
 457                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 458                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 459                                      &lwi);
 460                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 461                         if (oti->oti_thread->t_watchdog)
 462                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 463                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 464                 } else {
 465                         cfs_spin_unlock(&qctxt->lqc_lock);
 466                 }
 467
 468                 cycle++;
 469                 if (isblk)
 470                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 471                 /* after acquire(), we should run quota_check_common again
 472                  * so that we confirm there are enough quota to finish write */
 473                 rc = acquire(obd, id, oti, isblk);
 474
 475                 /* please reference to dqacq_completion for the below */
 476                 /* a new request is finished, try again */
 477                 if (rc == QUOTA_REQ_RETURNED) {
 478                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 479                         continue;
 480                 }
 481
 482                 /* it is out of quota already */
 483                 if (rc == -EDQUOT) {
 484                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 485                         break;
 486                 }
 487
 488                 /* Related quota has been disabled by master, but enabled by
 489                  * slave, do not try again. */
 490                 if (unlikely(rc == -ESRCH)) {
 491                         CERROR("mismatched quota configuration, stop try.\n");
 492                         break;
 493                 }
 494
 495                 if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
 496                         /* The client has been evicted or tried to
 497                          * to reconnect already, abort the request */
 498                         RETURN(-ENOTCONN);
 499
 500                 /* -EBUSY and others, wait a second and try again */
 501                 if (rc < 0) {
 502                         cfs_waitq_t        waitq;
 503                         struct l_wait_info lwi;
 504
 505                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 506                                 lc_watchdog_touch(oti->oti_thread->t_watchdog,
 507                                        CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
 508                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 509                                count_err++);
 510
 511                         cfs_waitq_init(&waitq);
 512                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 513                                           NULL);
 514                         l_wait_event(waitq, 0, &lwi);
 515                 }
 516
 517                 if (rc < 0 || cycle % 10 == 0) {
 518                         cfs_spin_lock(&last_print_lock);
 519                         if (last_print == 0 ||
 520                             cfs_time_before((last_print + cfs_time_seconds(30)),
 521                                             cfs_time_current())) {
 522                                 last_print = cfs_time_current();
 523                                 cfs_spin_unlock(&last_print_lock);
 524                                 CWARN("still haven't managed to acquire quota "
 525                                       "space from the quota master after %d "
 526                                       "retries (err=%d, rc=%d)\n",
 527                                       cycle, count_err - 1, rc);
 528                         } else {
 529                                 cfs_spin_unlock(&last_print_lock);
 530                         }
 531                 }
 532
 533                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 534                        cycle);
 535         }
 536         cfs_gettimeofday(&work_end);
 537         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 538         lprocfs_counter_add(qctxt->lqc_stats,
 539                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 540                                     LQUOTA_WAIT_FOR_CHK_INO,
 541                             timediff);
 542
 543         if (rc > 0)
 544                 rc = 0;
 545         RETURN(rc);
 546 }
 547
 548 /**
 549  * when a block_write or inode_create rpc is finished, adjust the record for
 550  * pending blocks and inodes
 551  */
 552 static int quota_pending_commit(struct obd_device *obd, const unsigned int id[],
 553                                 int pending[], int isblk)
 554 {
 555         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 556         struct timeval work_start;
 557         struct timeval work_end;
 558         long timediff;
 559         int i;
 560         struct qunit_data qdata[MAXQUOTAS];
 561         ENTRY;
 562
 563         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 564         CLASSERT(MAXQUOTAS < 4);
 565         if (!ll_sb_any_quota_active(qctxt->lqc_sb))
 566                 RETURN(0);
 567
 568         cfs_gettimeofday(&work_start);
 569         for (i = 0; i < MAXQUOTAS; i++) {
 570                 struct lustre_qunit_size *lqs = NULL;
 571
 572                 LASSERT(pending[i] >= 0);
 573                 if (pending[i] == 0)
 574                         continue;
 575
 576                 qdata[i].qd_id = id[i];
 577                 qdata[i].qd_flags = i;
 578                 if (isblk)
 579                         QDATA_SET_BLK(&qdata[i]);
 580                 qdata[i].qd_count = 0;
 581
 582                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 583                         continue;
 584
 585                 lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0);
 586                 if (lqs == NULL || IS_ERR(lqs)) {
 587                         CERROR("can not find lqs for pending_commit: "
 588                                "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), "
 589                                "maybe cause unexpected lqs refcount error!\n",
 590                                id[i], i ? 'g': 'u', pending[i], isblk,
 591                                lqs ? PTR_ERR(lqs) : -1);
 592                         continue;
 593                 }
 594
 595                 cfs_spin_lock(&lqs->lqs_lock);
 596                 if (isblk) {
 597                         LASSERTF(lqs->lqs_bwrite_pending >= pending[i],
 598                                  "there are too many blocks! [id %u] [%c] "
 599                                  "[bwrite_pending %lu] [pending %u]\n",
 600                                  id[i], i % 2 ? 'g' : 'u',
 601                                  lqs->lqs_bwrite_pending, pending[i]);
 602
 603                         lqs->lqs_bwrite_pending -= pending[i];
 604                 } else {
 605                         LASSERTF(lqs->lqs_iwrite_pending >= pending[i],
 606                                 "there are too many files! [id %u] [%c] "
 607                                 "[iwrite_pending %lu] [pending %u]\n",
 608                                 id[i], i % 2 ? 'g' : 'u',
 609                                 lqs->lqs_iwrite_pending, pending[i]);
 610
 611                         lqs->lqs_iwrite_pending -= pending[i];
 612                 }
 613                 CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n",
 614                        obd->obd_name,
 615                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 616                        i, pending[i], isblk);
 617                 cfs_spin_unlock(&lqs->lqs_lock);
 618
 619                 /* for quota_search_lqs in pending_commit */
 620                 lqs_putref(lqs);
 621                 /* for quota_search_lqs in quota_check */
 622                 lqs_putref(lqs);
 623         }
 624         cfs_gettimeofday(&work_end);
 625         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 626         lprocfs_counter_add(qctxt->lqc_stats,
 627                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 628                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 629                             timediff);
 630
 631         RETURN(0);
 632 }
 633
 634 static int mds_quota_init(void)
 635 {
 636         return lustre_dquot_init();
 637 }
 638
 639 static int mds_quota_exit(void)
 640 {
 641         lustre_dquot_exit();
 642         return 0;
 643 }
 644
 645 static int mds_quota_setup(struct obd_device *obd)
 646 {
 647         struct obd_device_target *obt = &obd->u.obt;
 648         struct mds_obd *mds = &obd->u.mds;
 649         int rc;
 650         ENTRY;
 651
 652         if (unlikely(mds->mds_quota)) {
 653                 CWARN("try to reinitialize quota context!\n");
 654                 RETURN(0);
 655         }
 656
 657         cfs_init_rwsem(&obt->obt_rwsem);
 658         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 659         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 660         cfs_sema_init(&obt->obt_quotachecking, 1);
 661         /* initialize quota master and quota context */
 662         cfs_init_rwsem(&mds->mds_qonoff_sem);
 663         rc = qctxt_init(obd, dqacq_handler);
 664         if (rc) {
 665                 CERROR("%s: initialize quota context failed! (rc:%d)\n",
 666                        obd->obd_name, rc);
 667                 RETURN(rc);
 668         }
 669         mds->mds_quota = 1;
 670         RETURN(rc);
 671 }
 672
 673 static int mds_quota_cleanup(struct obd_device *obd)
 674 {
 675         ENTRY;
 676         if (unlikely(!obd->u.mds.mds_quota))
 677                 RETURN(0);
 678
 679         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 680         RETURN(0);
 681 }
 682
 683 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 684 {
 685         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 686         ENTRY;
 687
 688         if (unlikely(!obd->u.mds.mds_quota))
 689                 RETURN(0);
 690
 691         if (data != NULL)
 692                 QUOTA_MASTER_READY(qctxt);
 693         else
 694                 QUOTA_MASTER_UNREADY(qctxt);
 695         RETURN(0);
 696 }
 697
 698 static int mds_quota_fs_cleanup(struct obd_device *obd)
 699 {
 700         struct mds_obd *mds = &obd->u.mds;
 701         struct obd_quotactl oqctl;
 702         ENTRY;
 703
 704         if (unlikely(!mds->mds_quota))
 705                 RETURN(0);
 706
 707         mds->mds_quota = 0;
 708         memset(&oqctl, 0, sizeof(oqctl));
 709         oqctl.qc_type = UGQUOTA;
 710
 711         cfs_down_write(&mds->mds_qonoff_sem);
 712         mds_admin_quota_off(obd, &oqctl);
 713         cfs_up_write(&mds->mds_qonoff_sem);
 714         RETURN(0);
 715 }
 716
 717 static int quota_acquire_common(struct obd_device *obd, const unsigned int id[],
 718                                 struct obd_trans_info *oti, int isblk)
 719 {
 720         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 721         int rc;
 722         ENTRY;
 723
 724         rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti);
 725         RETURN(rc);
 726 }
 727
 728 #endif /* HAVE_QUOTA_SUPPORT */
 729 #endif /* __KERNEL__ */
 730
 731 struct osc_quota_info {
 732         cfs_list_t              oqi_hash;       /* hash list */
 733         struct client_obd      *oqi_cli;        /* osc obd */
 734         unsigned int            oqi_id;         /* uid/gid of a file */
 735         short                   oqi_type;       /* quota type */
 736 };
 737
 738 cfs_spinlock_t qinfo_list_lock = CFS_SPIN_LOCK_UNLOCKED;
 739
 740 static cfs_list_t qinfo_hash[NR_DQHASH];
 741 /* SLAB cache for client quota context */
 742 cfs_mem_cache_t *qinfo_cachep = NULL;
 743
 744 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 745                          __attribute__((__const__));
 746
 747 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 748 {
 749         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 750         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 751         return tmp;
 752 }
 753
 754 /* caller must hold qinfo_list_lock */
 755 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 756 {
 757         cfs_list_t *head = qinfo_hash +
 758                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 759
 760         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 761         cfs_list_add(&oqi->oqi_hash, head);
 762 }
 763
 764 /* caller must hold qinfo_list_lock */
 765 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 766 {
 767         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 768         cfs_list_del_init(&oqi->oqi_hash);
 769 }
 770
 771 /* caller must hold qinfo_list_lock */
 772 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 773                                                 unsigned int id, int type)
 774 {
 775         unsigned int hashent = hashfn(cli, id, type);
 776         struct osc_quota_info *oqi;
 777         ENTRY;
 778
 779         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 780         cfs_list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 781                 if (oqi->oqi_cli == cli &&
 782                     oqi->oqi_id == id && oqi->oqi_type == type)
 783                         return oqi;
 784         }
 785         RETURN(NULL);
 786 }
 787
 788 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 789                                           unsigned int id, int type)
 790 {
 791         struct osc_quota_info *oqi;
 792         ENTRY;
 793
 794         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_IO, sizeof(*oqi));
 795         if(!oqi)
 796                 RETURN(NULL);
 797
 798         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 799         oqi->oqi_cli = cli;
 800         oqi->oqi_id = id;
 801         oqi->oqi_type = type;
 802
 803         RETURN(oqi);
 804 }
 805
 806 static void free_qinfo(struct osc_quota_info *oqi)
 807 {
 808         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 809 }
 810
 811 int osc_quota_chkdq(struct client_obd *cli, const unsigned int qid[])
 812 {
 813         unsigned int id;
 814         int cnt, rc = QUOTA_OK;
 815         ENTRY;
 816
 817         cfs_spin_lock(&qinfo_list_lock);
 818         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 819                 struct osc_quota_info *oqi = NULL;
 820
 821                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 822                 oqi = find_qinfo(cli, id, cnt);
 823                 if (oqi) {
 824                         rc = NO_QUOTA;
 825                         break;
 826                 }
 827         }
 828         cfs_spin_unlock(&qinfo_list_lock);
 829
 830         if (rc == NO_QUOTA)
 831                 CDEBUG(D_QUOTA, "chkdq found noquota for %s %d\n",
 832                        cnt == USRQUOTA ? "user" : "group", id);
 833         RETURN(rc);
 834 }
 835
 836 int osc_quota_setdq(struct client_obd *cli, const unsigned int qid[],
 837                     obd_flag valid, obd_flag flags)
 838 {
 839         unsigned int id;
 840         obd_flag noquota;
 841         int cnt, rc = 0;
 842         ENTRY;
 843
 844
 845         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 846                 struct osc_quota_info *oqi = NULL, *old;
 847
 848                 if (!(valid & ((cnt == USRQUOTA) ?
 849                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 850                         continue;
 851
 852                 id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA];
 853                 noquota = (cnt == USRQUOTA) ?
 854                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 855
 856                 if (noquota) {
 857                         oqi = alloc_qinfo(cli, id, cnt);
 858                         if (!oqi) {
 859                                 rc = -ENOMEM;
 860                                 CDEBUG(D_QUOTA, "setdq for %s %d failed, "
 861                                        "(rc = %d)\n",
 862                                        cnt == USRQUOTA ? "user" : "group",
 863                                        id, rc);
 864                                 break;
 865                         }
 866                 }
 867
 868                 cfs_spin_lock(&qinfo_list_lock);
 869                 old = find_qinfo(cli, id, cnt);
 870                 if (old && !noquota)
 871                         remove_qinfo_hash(old);
 872                 else if (!old && noquota)
 873                         insert_qinfo_hash(oqi);
 874                 cfs_spin_unlock(&qinfo_list_lock);
 875
 876                 if (old && !noquota)
 877                         CDEBUG(D_QUOTA, "setdq to remove for %s %d\n",
 878                                cnt == USRQUOTA ? "user" : "group", id);
 879                 else if (!old && noquota)
 880                         CDEBUG(D_QUOTA, "setdq to insert for %s %d\n",
 881                                cnt == USRQUOTA ? "user" : "group", id);
 882
 883                 if (old) {
 884                         if (noquota)
 885                                 free_qinfo(oqi);
 886                         else
 887                                 free_qinfo(old);
 888                 }
 889         }
 890
 891         RETURN(rc);
 892 }
 893
 894 int osc_quota_cleanup(struct obd_device *obd)
 895 {
 896         struct client_obd *cli = &obd->u.cli;
 897         struct osc_quota_info *oqi, *n;
 898         int i;
 899         ENTRY;
 900
 901         cfs_spin_lock(&qinfo_list_lock);
 902         for (i = 0; i < NR_DQHASH; i++) {
 903                 cfs_list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 904                         if (oqi->oqi_cli != cli)
 905                                 continue;
 906                         remove_qinfo_hash(oqi);
 907                         free_qinfo(oqi);
 908                 }
 909         }
 910         cfs_spin_unlock(&qinfo_list_lock);
 911
 912         RETURN(0);
 913 }
 914
 915 int osc_quota_init(void)
 916 {
 917         int i;
 918         ENTRY;
 919
 920         LASSERT(qinfo_cachep == NULL);
 921         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 922                                             sizeof(struct osc_quota_info),
 923                                             0, 0);
 924         if (!qinfo_cachep)
 925                 RETURN(-ENOMEM);
 926
 927         for (i = 0; i < NR_DQHASH; i++)
 928                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 929
 930         RETURN(0);
 931 }
 932
 933 int osc_quota_exit(void)
 934 {
 935         struct osc_quota_info *oqi, *n;
 936         int i, rc;
 937         ENTRY;
 938
 939         cfs_spin_lock(&qinfo_list_lock);
 940         for (i = 0; i < NR_DQHASH; i++) {
 941                 cfs_list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 942                         remove_qinfo_hash(oqi);
 943                         free_qinfo(oqi);
 944                 }
 945         }
 946         cfs_spin_unlock(&qinfo_list_lock);
 947
 948         rc = cfs_mem_cache_destroy(qinfo_cachep);
 949         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 950         qinfo_cachep = NULL;
 951
 952         RETURN(0);
 953 }
 954
 955 #ifdef __KERNEL__
 956 #ifdef HAVE_QUOTA_SUPPORT
 957 quota_interface_t mds_quota_interface = {
 958         .quota_init     = mds_quota_init,
 959         .quota_exit     = mds_quota_exit,
 960         .quota_setup    = mds_quota_setup,
 961         .quota_cleanup  = mds_quota_cleanup,
 962         .quota_check    = target_quota_check,
 963         .quota_ctl      = mds_quota_ctl,
 964         .quota_setinfo  = mds_quota_setinfo,
 965         .quota_fs_cleanup = mds_quota_fs_cleanup,
 966         .quota_recovery = mds_quota_recovery,
 967         .quota_adjust   = mds_quota_adjust,
 968         .quota_chkquota = quota_chk_acq_common,
 969         .quota_acquire  = quota_acquire_common,
 970         .quota_pending_commit = quota_pending_commit,
 971 };
 972
 973 quota_interface_t filter_quota_interface = {
 974         .quota_setup    = filter_quota_setup,
 975         .quota_cleanup  = filter_quota_cleanup,
 976         .quota_check    = target_quota_check,
 977         .quota_ctl      = filter_quota_ctl,
 978         .quota_setinfo  = filter_quota_setinfo,
 979         .quota_clearinfo = filter_quota_clearinfo,
 980         .quota_enforce  = filter_quota_enforce,
 981         .quota_getflag  = filter_quota_getflag,
 982         .quota_acquire  = quota_acquire_common,
 983         .quota_adjust   = filter_quota_adjust,
 984         .quota_chkquota = quota_chk_acq_common,
 985         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 986         .quota_pending_commit = quota_pending_commit,
 987 };
 988 #endif
 989 #endif /* __KERNEL__ */
 990
 991 quota_interface_t mdc_quota_interface = {
 992         .quota_ctl      = client_quota_ctl,
 993         .quota_check    = client_quota_check,
 994         .quota_poll_check = client_quota_poll_check,
 995 };
 996
 997 quota_interface_t lmv_quota_interface = {
 998         .quota_ctl      = lmv_quota_ctl,
 999         .quota_check    = lmv_quota_check,
1000 };
1001
1002 quota_interface_t osc_quota_interface = {
1003         .quota_ctl      = client_quota_ctl,
1004         .quota_check    = client_quota_check,
1005         .quota_poll_check = client_quota_poll_check,
1006         .quota_init     = osc_quota_init,
1007         .quota_exit     = osc_quota_exit,
1008         .quota_chkdq    = osc_quota_chkdq,
1009         .quota_setdq    = osc_quota_setdq,
1010         .quota_cleanup  = osc_quota_cleanup,
1011         .quota_adjust_qunit = client_quota_adjust_qunit,
1012 };
1013
1014 quota_interface_t lov_quota_interface = {
1015         .quota_ctl      = lov_quota_ctl,
1016         .quota_check    = lov_quota_check,
1017         .quota_adjust_qunit = lov_quota_adjust_qunit,
1018 };
1019
1020 #ifdef __KERNEL__
1021
1022 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
1023
1024 static int __init init_lustre_quota(void)
1025 {
1026 #ifdef HAVE_QUOTA_SUPPORT
1027         int rc = 0;
1028
1029         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
1030                                                 proc_lustre_root,
1031                                                 NULL, NULL);
1032         if (IS_ERR(lquota_type_proc_dir)) {
1033                 CERROR("LProcFS failed in lquota-init\n");
1034                 rc = PTR_ERR(lquota_type_proc_dir);
1035                 return rc;
1036         }
1037
1038         rc = qunit_cache_init();
1039         if (rc)
1040                 return rc;
1041
1042         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
1043         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
1044 #endif
1045         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
1046         PORTAL_SYMBOL_REGISTER(lmv_quota_interface);
1047         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
1048         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
1049         return 0;
1050 }
1051
1052 static void /*__exit*/ exit_lustre_quota(void)
1053 {
1054         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
1055         PORTAL_SYMBOL_UNREGISTER(lmv_quota_interface);
1056         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
1057         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
1058 #ifdef HAVE_QUOTA_SUPPORT
1059         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
1060         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
1061
1062         qunit_cache_cleanup();
1063
1064         if (lquota_type_proc_dir)
1065                 lprocfs_remove(&lquota_type_proc_dir);
1066 #endif
1067 }
1068
1069 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1070 MODULE_DESCRIPTION("Lustre Quota");
1071 MODULE_LICENSE("GPL");
1072
1073 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
1074
1075 #ifdef HAVE_QUOTA_SUPPORT
1076 EXPORT_SYMBOL(mds_quota_interface);
1077 EXPORT_SYMBOL(filter_quota_interface);
1078 #endif
1079 EXPORT_SYMBOL(mdc_quota_interface);
1080 EXPORT_SYMBOL(lmv_quota_interface);
1081 EXPORT_SYMBOL(osc_quota_interface);
1082 EXPORT_SYMBOL(lov_quota_interface);
1083 #endif /* __KERNEL */