lustre/quota/qsd_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2012, Intel Corporation.
  25  * Use is subject to license terms.
  26  *
  27  * Author: Johann Lombardi <johann.lombardi@intel.com>
  28  * Author: Niu    Yawei    <yawei.niu@intel.com>
  29  */
  30
  31 /*
  32  * Quota Slave Driver (QSD) management.
  33  *
  34  * The quota slave feature is implemented under the form of a library called
  35  * QSD. Each OSD device should create a QSD instance via qsd_init() which will
  36  * be used to manage quota enforcement for this device. This implies:
  37  * - completing the reintegration procedure with the quota master (aka QMT, see
  38  *   qmt_dev.c) to retrieve the latest quota settings and space distribution.
  39  * - managing quota locks in order to be notified of configuration changes.
  40  * - acquiring space from the QMT when quota space for a given user/group is
  41  *   close to exhaustion.
  42  * - allocating quota space to service threads for local request processing.
  43  *
  44  * Once the QSD instance created, the OSD device should invoke qsd_start()
  45  * when recovery is completed. This notifies the QSD that we are about to
  46  * process new requests on which quota should be strictly enforced.
  47  * Then, qsd_op_begin/end can be used to reserve/release/pre-acquire quota space
  48  * for/after each operation until shutdown where the QSD instance should be
  49  * freed via qsd_fini().
  50  */
  51
  52 #ifndef EXPORT_SYMTAB
  53 # define EXPORT_SYMTAB
  54 #endif
  55
  56 #define DEBUG_SUBSYSTEM S_LQUOTA
  57
  58 #include <obd_class.h>
  59 #include "qsd_internal.h"
  60
  61 cfs_mem_cache_t *upd_kmem;
  62
  63 struct lu_kmem_descr qsd_caches[] = {
  64         {
  65                 .ckd_cache = &upd_kmem,
  66                 .ckd_name  = "upd_kmem",
  67                 .ckd_size  = sizeof(struct qsd_upd_rec)
  68         },
  69         {
  70                 .ckd_cache = NULL
  71         }
  72 };
  73
  74 /* define qsd thread key */
  75 LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
  76 LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
  77 LU_KEY_INIT_GENERIC(qsd);
  78
  79 /* some procfs helpers */
  80 static int lprocfs_qsd_rd_state(char *page, char **start, off_t off,
  81                                 int count, int *eof, void *data)
  82 {
  83         struct qsd_instance     *qsd = (struct qsd_instance *)data;
  84         char                     enabled[5];
  85         int                      rc;
  86
  87         LASSERT(qsd != NULL);
  88
  89         memset(enabled, 0, sizeof(enabled));
  90         if (qsd_type_enabled(qsd, USRQUOTA))
  91                 strcat(enabled, "u");
  92         if (qsd_type_enabled(qsd, GRPQUOTA))
  93                 strcat(enabled, "g");
  94         if (strlen(enabled) == 0)
  95                 strcat(enabled, "none");
  96
  97         rc = snprintf(page, count,
  98                       "target name:    %s\n"
  99                       "pool ID:        %d\n"
 100                       "type:           %s\n"
 101                       "quota enabled:  %s\n"
 102                       "conn to master: %s\n",
 103                       qsd->qsd_svname, qsd->qsd_pool_id,
 104                       qsd->qsd_is_md ? "md" : "dt", enabled,
 105                       qsd->qsd_exp_valid ? "setup" : "not setup yet");
 106
 107         if (qsd->qsd_prepared) {
 108                 memset(enabled, 0, sizeof(enabled));
 109                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 110                         strcat(enabled, "u");
 111                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 112                         strcat(enabled, "g");
 113                 if (strlen(enabled) == 0)
 114                         strcat(enabled, "none");
 115                 rc +=  snprintf(page + rc, count - rc,
 116                                 "space acct:     %s\n"
 117                                 "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 118                                 "group uptodate: glb[%d],slv[%d],reint[%d]\n",
 119                                 enabled,
 120                                 qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 121                                 qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 122                                 qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 123                                 qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 124                                 qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 125                                 qsd->qsd_type_array[GRPQUOTA]->qqi_reint);
 126         }
 127         return rc;
 128 }
 129
 130 static int lprocfs_qsd_rd_enabled(char *page, char **start, off_t off,
 131                                   int count, int *eof, void *data)
 132 {
 133         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 134         char                     enabled[5];
 135
 136         LASSERT(qsd != NULL);
 137
 138         memset(enabled, 0, sizeof(enabled));
 139         if (qsd_type_enabled(qsd, USRQUOTA))
 140                 strcat(enabled, "u");
 141         if (qsd_type_enabled(qsd, GRPQUOTA))
 142                 strcat(enabled, "g");
 143         if (strlen(enabled) == 0)
 144                 strcat(enabled, "none");
 145
 146         return snprintf(page, count, "%s\n", enabled);
 147 }
 148
 149 /* force reintegration procedure to be executed.
 150  * Used for test/debugging purpose */
 151 static int lprocfs_qsd_wr_force_reint(struct file *file, const char *buffer,
 152                                       unsigned long count, void *data)
 153 {
 154         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 155         int                      rc = 0, qtype;
 156
 157         LASSERT(qsd != NULL);
 158
 159         write_lock(&qsd->qsd_lock);
 160         if (qsd->qsd_stopping) {
 161                 /* don't mess up with shutdown procedure, it is already
 162                  * complicated enough */
 163                 rc = -ESHUTDOWN;
 164         } else if (!qsd->qsd_prepared) {
 165                 rc = -EAGAIN;
 166         } else {
 167                 /* mark all indexes as stale */
 168                 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 169                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 170                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 171                 }
 172         }
 173         write_unlock(&qsd->qsd_lock);
 174
 175         if (rc)
 176                 return rc;
 177
 178         /* kick off reintegration */
 179         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 180                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 181                 if (rc)
 182                         break;
 183         }
 184         return rc == 0 ? count : rc;
 185 }
 186
 187 static int lprocfs_qsd_rd_timeout(char *page, char **start, off_t off,
 188                                   int count, int *eof, void *data)
 189 {
 190         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 191         LASSERT(qsd != NULL);
 192
 193         return snprintf(page, count, "%d\n", qsd_wait_timeout(qsd));
 194 }
 195
 196 static int lprocfs_qsd_wr_timeout(struct file *file, const char *buffer,
 197                                   unsigned long count, void *data)
 198 {
 199         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 200         int                      timeout, rc;
 201         LASSERT(qsd != NULL);
 202
 203         rc = lprocfs_write_helper(buffer, count, &timeout);
 204         if (rc)
 205                 return rc;
 206         if (timeout < 0)
 207                 return -EINVAL;
 208
 209         qsd->qsd_timeout = timeout;
 210         return count;
 211 }
 212
 213 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 214         { "info", lprocfs_qsd_rd_state, 0, 0},
 215         { "enabled", lprocfs_qsd_rd_enabled, 0, 0},
 216         { "force_reint", 0, lprocfs_qsd_wr_force_reint, 0},
 217         { "timeout", lprocfs_qsd_rd_timeout, lprocfs_qsd_wr_timeout, 0},
 218         { NULL }
 219 };
 220
 221 /*
 222  * Callback function invoked by the OSP layer when the connection to the master
 223  * has been set up.
 224  *
 225  * \param data - is a pointer to the qsd_instance
 226  *
 227  * \retval - 0 on success, appropriate error on failure
 228  */
 229 static int qsd_conn_callback(void *data)
 230 {
 231         struct qsd_instance *qsd = (struct qsd_instance *)data;
 232         int                  type;
 233         ENTRY;
 234
 235         /* qsd_exp should now be valid */
 236         LASSERT(qsd->qsd_exp);
 237
 238         /* grab reference on namespace */
 239         ldlm_namespace_get(class_exp2obd(qsd->qsd_exp)->obd_namespace);
 240         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 241
 242         write_lock(&qsd->qsd_lock);
 243         /* notify that qsd_exp is now valid */
 244         qsd->qsd_exp_valid = true;
 245         write_unlock(&qsd->qsd_lock);
 246
 247         /* Now that the connection to master is setup, we can initiate the
 248          * reintegration procedure for quota types which are enabled.
 249          * It is worth noting that, if the qsd_instance hasn't been started
 250          * already, then we can only complete the first two steps of the
 251          * reintegration procedure (i.e. global lock enqueue and slave
 252          * index transfer) since the space usage reconciliation (i.e.
 253          * step 3) will have to wait for qsd_start() to be called */
 254         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 255                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 256                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 257         }
 258
 259         RETURN(0);
 260 }
 261
 262 /*
 263  * Release qsd_qtype_info structure which contains data associated with a
 264  * given quota type. This releases the accounting objects.
 265  * It's called on OSD cleanup when the qsd instance is released.
 266  *
 267  * \param env - is the environment passed by the caller
 268  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 269  *              to be released
 270  * \param qtype - is the quota type to be shutdown
 271  */
 272 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 273                            int qtype)
 274 {
 275         struct qsd_qtype_info   *qqi;
 276         ENTRY;
 277
 278         if (qsd->qsd_type_array[qtype] == NULL)
 279                 RETURN_EXIT;
 280         qqi = qsd->qsd_type_array[qtype];
 281         qsd->qsd_type_array[qtype] = NULL;
 282
 283         /* all deferred work lists should be empty */
 284         LASSERT(cfs_list_empty(&qqi->qqi_deferred_glb));
 285         LASSERT(cfs_list_empty(&qqi->qqi_deferred_slv));
 286
 287         /* shutdown lquota site */
 288         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 289                 lquota_site_free(env, qqi->qqi_site);
 290                 qqi->qqi_site = NULL;
 291         }
 292
 293         /* by now, all qqi users should have gone away */
 294         LASSERT(cfs_atomic_read(&qqi->qqi_ref) == 1);
 295         lu_ref_fini(&qqi->qqi_reference);
 296
 297         /* release accounting object */
 298         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 299                 lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
 300                 qqi->qqi_acct_obj = NULL;
 301         }
 302
 303         /* release slv index */
 304         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 305                 lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
 306                 qqi->qqi_slv_obj = NULL;
 307                 qqi->qqi_slv_ver = 0;
 308         }
 309
 310         /* release global index */
 311         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 312                 lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
 313                 qqi->qqi_glb_obj = NULL;
 314                 qqi->qqi_glb_ver = 0;
 315         }
 316
 317         OBD_FREE_PTR(qqi);
 318         EXIT;
 319 }
 320
 321 /*
 322  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 323  * This opens the accounting object and initializes the proc file.
 324  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 325  * instance.
 326  *
 327  * \param env  - the environment passed by the caller
 328  * \param qsd  - is the qsd instance which will be in charge of the new
 329  *               qsd_qtype_info instance.
 330  * \param qtype - is quota type to set up
 331  *
 332  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 333  *           appropriate error on failure
 334  */
 335 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 336                           int qtype)
 337 {
 338         struct qsd_qtype_info   *qqi;
 339         int                      rc;
 340         struct obd_uuid          uuid;
 341         ENTRY;
 342
 343         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 344
 345         /* allocate structure for this quota type */
 346         OBD_ALLOC_PTR(qqi);
 347         if (qqi == NULL)
 348                 RETURN(-ENOMEM);
 349         qsd->qsd_type_array[qtype] = qqi;
 350         cfs_atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 351
 352         /* set backpointer and other parameters */
 353         qqi->qqi_qsd   = qsd;
 354         qqi->qqi_qtype = qtype;
 355         lu_ref_init(&qqi->qqi_reference);
 356         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id, QSD_RES_TYPE(qsd),
 357                             qtype);
 358         qqi->qqi_glb_uptodate = false;
 359         qqi->qqi_slv_uptodate = false;
 360         qqi->qqi_reint        = false;
 361         cfs_waitq_init(&qqi->qqi_reint_thread.t_ctl_waitq);
 362         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 363         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 364         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 365         memset(&qqi->qqi_lockh, 0, sizeof(qqi->qqi_lockh));
 366
 367         /* open accounting object */
 368         LASSERT(qqi->qqi_acct_obj == NULL);
 369         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 370         if (IS_ERR(qqi->qqi_acct_obj)) {
 371                 LCONSOLE_WARN("%s: No %s space accounting support. Please "
 372                               "consider running tunefs.lustre --quota on an "
 373                               "unmounted filesystem to enable quota accounting."
 374                               "\n", qsd->qsd_svname,
 375                               QTYPE_NAME(qtype));
 376                 qqi->qqi_acct_obj = NULL;
 377                 qsd->qsd_acct_failed = true;
 378         }
 379
 380         /* open global index copy */
 381         LASSERT(qqi->qqi_glb_obj == NULL);
 382         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 383                                                        qsd->qsd_root,
 384                                                        &qqi->qqi_fid, true);
 385         if (IS_ERR(qqi->qqi_glb_obj)) {
 386                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 387                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 388                        PTR_ERR(qqi->qqi_glb_obj));
 389                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 390         }
 391         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 392
 393         /* open slave index copy */
 394         LASSERT(qqi->qqi_slv_obj == NULL);
 395         obd_str2uuid(&uuid, qsd->qsd_svname);
 396         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 397                                                        qsd->qsd_root,
 398                                                        &qqi->qqi_fid, &uuid,
 399                                                        true);
 400         if (IS_ERR(qqi->qqi_slv_obj)) {
 401                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 402                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 403                        PTR_ERR(qqi->qqi_slv_obj));
 404                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 405         }
 406         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 407
 408         /* allocate site */
 409         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 410         if (IS_ERR(qqi->qqi_site)) {
 411                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 412                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 413                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 414         }
 415
 416         /* register proc entry for accounting & global index copy objects */
 417         rc = lprocfs_seq_create(qsd->qsd_proc,
 418                                 qtype == USRQUOTA ? "acct_user" : "acct_group",
 419                                 0444, &lprocfs_quota_seq_fops,
 420                                 qqi->qqi_acct_obj);
 421         if (rc) {
 422                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 423                        qsd->qsd_svname, rc);
 424                 GOTO(out, rc);
 425         }
 426
 427         rc = lprocfs_seq_create(qsd->qsd_proc,
 428                                 qtype == USRQUOTA ? "limit_user" : "limit_group",
 429                                 0444, &lprocfs_quota_seq_fops,
 430                                 qqi->qqi_glb_obj);
 431         if (rc) {
 432                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 433                        qsd->qsd_svname, rc);
 434                 GOTO(out, rc);
 435         }
 436         EXIT;
 437 out:
 438         if (rc)
 439                 qsd_qtype_fini(env, qsd, qtype);
 440         return rc;
 441 }
 442
 443 /*
 444  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 445  * structures associated with the quota slave (on-disk objects, lquota entry
 446  * tables, ...).
 447  * This function should be called when the OSD is shutting down.
 448  *
 449  * \param env - is the environment passed by the caller
 450  * \param qsd - is the qsd instance to shutdown
 451  */
 452 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 453 {
 454         int     qtype;
 455         ENTRY;
 456
 457         if (unlikely(qsd == NULL))
 458                 RETURN_EXIT;
 459
 460         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 461         write_lock(&qsd->qsd_lock);
 462         qsd->qsd_stopping = true;
 463         write_unlock(&qsd->qsd_lock);
 464
 465         /* remove from the list of fsinfo */
 466         if (!cfs_list_empty(&qsd->qsd_link)) {
 467                 LASSERT(qsd->qsd_fsinfo != NULL);
 468                 down(&qsd->qsd_fsinfo->qfs_sem);
 469                 cfs_list_del_init(&qsd->qsd_link);
 470                 up(&qsd->qsd_fsinfo->qfs_sem);
 471         }
 472
 473         /* remove qsd proc entry */
 474         if (qsd->qsd_proc != NULL) {
 475                 lprocfs_remove(&qsd->qsd_proc);
 476                 qsd->qsd_proc = NULL;
 477         }
 478
 479         /* stop the writeback thread */
 480         qsd_stop_upd_thread(qsd);
 481
 482         /* shutdown the reintegration threads */
 483         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 484                 if (qsd->qsd_type_array[qtype] == NULL)
 485                         continue;
 486                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 487         }
 488
 489         /* release reference on namespace */
 490         if (qsd->qsd_ns != NULL) {
 491                 ldlm_namespace_put(qsd->qsd_ns);
 492                 qsd->qsd_ns = NULL;
 493         }
 494
 495         /* free per-quota type data */
 496         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
 497                 qsd_qtype_fini(env, qsd, qtype);
 498
 499         /* deregister connection to the quota master */
 500         qsd->qsd_exp_valid = false;
 501         lustre_deregister_osp_item(&qsd->qsd_exp);
 502
 503         /* release per-filesystem information */
 504         if (qsd->qsd_fsinfo != NULL)
 505                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 506
 507         /* release quota root directory */
 508         if (qsd->qsd_root != NULL) {
 509                 lu_object_put(env, &qsd->qsd_root->do_lu);
 510                 qsd->qsd_root = NULL;
 511         }
 512
 513         /* release reference on dt_device */
 514         if (qsd->qsd_dev != NULL) {
 515                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 516                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 517                 qsd->qsd_dev = NULL;
 518         }
 519
 520         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 521         OBD_FREE_PTR(qsd);
 522         EXIT;
 523 }
 524 EXPORT_SYMBOL(qsd_fini);
 525
 526 /*
 527  * Create a new qsd_instance to be associated with backend osd device
 528  * identified by \dev.
 529  *
 530  * \param env    - the environment passed by the caller
 531  * \param svname - is the service name of the OSD device creating this instance
 532  * \param dev    - is the dt_device where to store quota index files
 533  * \param osd_proc - is the procfs parent directory where to create procfs file
 534  *                   related to this new qsd instance
 535  *
 536  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 537  *           appropriate error on failure
 538  */
 539 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 540                               struct dt_device *dev,
 541                               cfs_proc_dir_entry_t *osd_proc)
 542 {
 543         struct qsd_thread_info  *qti = qsd_info(env);
 544         struct qsd_instance     *qsd;
 545         int                      rc, type, idx;
 546         ENTRY;
 547
 548         /* only configure qsd for MDT & OST */
 549         type = server_name2index(svname, &idx, NULL);
 550         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 551                 RETURN(NULL);
 552
 553         /* allocate qsd instance */
 554         OBD_ALLOC_PTR(qsd);
 555         if (qsd == NULL)
 556                 RETURN(ERR_PTR(-ENOMEM));
 557
 558         /* generic initializations */
 559         rwlock_init(&qsd->qsd_lock);
 560         CFS_INIT_LIST_HEAD(&qsd->qsd_link);
 561         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 562         cfs_waitq_init(&qsd->qsd_upd_thread.t_ctl_waitq);
 563         CFS_INIT_LIST_HEAD(&qsd->qsd_upd_list);
 564         spin_lock_init(&qsd->qsd_adjust_lock);
 565         CFS_INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 566         qsd->qsd_prepared = false;
 567         qsd->qsd_started = false;
 568
 569         /* copy service name */
 570         strncpy(qsd->qsd_svname, svname, MAX_OBD_NAME);
 571
 572         /* grab reference on osd device */
 573         lu_device_get(&dev->dd_lu_dev);
 574         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 575         qsd->qsd_dev = dev;
 576
 577         /* we only support pool ID 0 (default data or metadata pool) for the
 578          * time being. A different pool ID could be assigned to this target via
 579          * the configuration log in the future */
 580         qsd->qsd_pool_id  = 0;
 581
 582         /* get fsname from svname */
 583         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 584         if (rc) {
 585                 CERROR("%s: fail to extract filesystem name\n", svname);
 586                 GOTO(out, rc);
 587         }
 588
 589         /* look up quota setting for the filesystem the target belongs to */
 590         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 591         if (qsd->qsd_fsinfo == NULL) {
 592                 CERROR("%s: failed to locate filesystem information\n", svname);
 593                 GOTO(out, rc = -EINVAL);
 594         }
 595
 596         /* add in the list of lquota_fsinfo */
 597         down(&qsd->qsd_fsinfo->qfs_sem);
 598         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 599         up(&qsd->qsd_fsinfo->qfs_sem);
 600
 601         /* register procfs directory */
 602         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 603                                          lprocfs_quota_qsd_vars, qsd);
 604         if (IS_ERR(qsd->qsd_proc)) {
 605                 rc = PTR_ERR(qsd->qsd_proc);
 606                 qsd->qsd_proc = NULL;
 607                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 608                        svname, rc);
 609                 GOTO(out, rc);
 610         }
 611         EXIT;
 612 out:
 613         if (rc) {
 614                 qsd_fini(env, qsd);
 615                 return ERR_PTR(rc);
 616         }
 617         RETURN(qsd);
 618 }
 619 EXPORT_SYMBOL(qsd_init);
 620
 621 /*
 622  * Initialize on-disk structures in order to manage quota enforcement for
 623  * the target associated with the qsd instance \qsd and starts the reintegration
 624  * procedure for each quota type as soon as possible.
 625  * The last step of the reintegration will be completed once qsd_start() is
 626  * called, at which points the space reconciliation with the master will be
 627  * executed.
 628  * This function must be called when the server stack is fully configured,
 629  * typically when ->ldo_prepare is called across the stack.
 630  *
 631  * \param env - the environment passed by the caller
 632  * \param qsd - is qsd_instance to prepare
 633  *
 634  * \retval - 0 on success, appropriate error on failure
 635  */
 636 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 637 {
 638         struct qsd_thread_info  *qti = qsd_info(env);
 639         int                      qtype, rc = 0;
 640         ENTRY;
 641
 642         if (unlikely(qsd == NULL))
 643                 RETURN(0);
 644
 645         read_lock(&qsd->qsd_lock);
 646         if (qsd->qsd_prepared) {
 647                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 648                 rc = -EALREADY;
 649         }
 650         read_unlock(&qsd->qsd_lock);
 651         if (rc)
 652                 RETURN(rc);
 653
 654         /* Record whether this qsd instance is managing quota enforcement for a
 655          * MDT (i.e. inode quota) or OST (block quota) */
 656         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 657                 qsd->qsd_is_md = true;
 658                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 659         } else {
 660                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 661         }
 662
 663         /* look-up on-disk directory for the quota slave */
 664         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 665                                                     QSD_DIR);
 666         if (IS_ERR(qsd->qsd_root)) {
 667                 rc = PTR_ERR(qsd->qsd_root);
 668                 qsd->qsd_root = NULL;
 669                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 670                        qsd->qsd_svname, rc);
 671                 RETURN(rc);
 672         }
 673
 674         /* initialize per-quota type data */
 675         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 676                 rc = qsd_qtype_init(env, qsd, qtype);
 677                 if (rc)
 678                         RETURN(rc);
 679         }
 680
 681         /* pools successfully setup, mark the qsd as prepared */
 682         write_lock(&qsd->qsd_lock);
 683         qsd->qsd_prepared = true;
 684         write_unlock(&qsd->qsd_lock);
 685
 686         /* start reintegration thread for each type, if required */
 687         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 688                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 689
 690                 if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
 691                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 692                                        "since space accounting isn't functional"
 693                                        ". Please run tunefs.lustre --quota on "
 694                                        "an unmounted filesystem if not done "
 695                                        "already\n", qsd->qsd_svname);
 696                         break;
 697                 }
 698
 699                 rc = qsd_start_reint_thread(qqi);
 700                 if (rc) {
 701                         CERROR("%s: failed to start reint thread for type %s "
 702                                "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
 703                                rc);
 704                         RETURN(rc);
 705                 }
 706         }
 707
 708         /* start writeback thread */
 709         rc = qsd_start_upd_thread(qsd);
 710         if (rc) {
 711                 CERROR("%s: failed to start writeback thread (%d)\n",
 712                        qsd->qsd_svname, rc);
 713                 RETURN(rc);
 714         }
 715
 716         /* generate osp name */
 717         rc = tgt_name2ospname((char *)qsd->qsd_svname, qti->qti_buf);
 718         if (rc) {
 719                 CERROR("%s: failed to generate ospname (%d)\n",
 720                        qsd->qsd_svname, rc);
 721                 RETURN(rc);
 722         }
 723
 724         /* the connection callback will start the reintegration
 725          * procedure if quota is enabled */
 726         rc = lustre_register_osp_item(qti->qti_buf, &qsd->qsd_exp,
 727                                       qsd_conn_callback, (void *)qsd);
 728         if (rc) {
 729                 CERROR("%s: fail to get connection to master (%d)\n",
 730                        qsd->qsd_svname, rc);
 731                 RETURN(rc);
 732         }
 733
 734         RETURN(0);
 735 }
 736 EXPORT_SYMBOL(qsd_prepare);
 737
 738 /*
 739  * Start a qsd instance. This will complete the last step of the reintegration
 740  * procedure as soon as possible (provided that the master is reachable).
 741  * This should be called when recovery has been completed and quota should now
 742  * be enforced on every operations.
 743  *
 744  * \param env - the environment passed by the caller
 745  * \param qsd - is the qsd instance associated with the osd device to start
 746  */
 747 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 748 {
 749         int     type, rc = 0;
 750         ENTRY;
 751
 752         if (unlikely(qsd == NULL))
 753                 RETURN(0);
 754
 755         write_lock(&qsd->qsd_lock);
 756         if (!qsd->qsd_prepared) {
 757                 CERROR("%s: can't start qsd instance since it wasn't properly "
 758                        "initialized\n", qsd->qsd_svname);
 759                 rc = -EFAULT;
 760         } else if (qsd->qsd_started) {
 761                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 762                 rc = -EALREADY;
 763         } else {
 764                 /* notify that the qsd_instance is now started */
 765                 qsd->qsd_started = true;
 766         }
 767         write_unlock(&qsd->qsd_lock);
 768
 769         if (rc)
 770                 RETURN(rc);
 771
 772         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 773          * up to usage; If usage < granted, release down to usage.  */
 774         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 775                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 776                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 777         }
 778
 779         RETURN(rc);
 780 }
 781 EXPORT_SYMBOL(qsd_start);
 782
 783 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 784
 785 /*
 786  * Global initialization performed at module load time
 787  */
 788 int qsd_glb_init(void)
 789 {
 790         int     rc;
 791
 792         rc = lu_kmem_init(qsd_caches);
 793         if (rc)
 794                 return rc;
 795
 796         qsd_key_init_generic(&qsd_thread_key, NULL);
 797         lu_context_key_register(&qsd_thread_key);
 798         lustre_register_quota_process_config(qsd_process_config);
 799
 800         return 0;
 801 }
 802
 803 /*
 804  * Companion of qsd_glb_init() called at module unload time
 805  */
 806 void qsd_glb_fini(void)
 807 {
 808         lustre_register_quota_process_config(NULL);
 809         lu_kmem_fini(qsd_caches);
 810         lu_context_key_degister(&qsd_thread_key);
 811 }