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, 2013, 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 struct kmem_cache *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         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 239
 240         write_lock(&qsd->qsd_lock);
 241         /* notify that qsd_exp is now valid */
 242         qsd->qsd_exp_valid = true;
 243         write_unlock(&qsd->qsd_lock);
 244
 245         /* Now that the connection to master is setup, we can initiate the
 246          * reintegration procedure for quota types which are enabled.
 247          * It is worth noting that, if the qsd_instance hasn't been started
 248          * already, then we can only complete the first two steps of the
 249          * reintegration procedure (i.e. global lock enqueue and slave
 250          * index transfer) since the space usage reconciliation (i.e.
 251          * step 3) will have to wait for qsd_start() to be called */
 252         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 253                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 254                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 255         }
 256
 257         RETURN(0);
 258 }
 259
 260 /*
 261  * Release qsd_qtype_info structure which contains data associated with a
 262  * given quota type. This releases the accounting objects.
 263  * It's called on OSD cleanup when the qsd instance is released.
 264  *
 265  * \param env - is the environment passed by the caller
 266  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 267  *              to be released
 268  * \param qtype - is the quota type to be shutdown
 269  */
 270 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 271                            int qtype)
 272 {
 273         struct qsd_qtype_info   *qqi;
 274         ENTRY;
 275
 276         if (qsd->qsd_type_array[qtype] == NULL)
 277                 RETURN_EXIT;
 278         qqi = qsd->qsd_type_array[qtype];
 279         qsd->qsd_type_array[qtype] = NULL;
 280
 281         /* all deferred work lists should be empty */
 282         LASSERT(cfs_list_empty(&qqi->qqi_deferred_glb));
 283         LASSERT(cfs_list_empty(&qqi->qqi_deferred_slv));
 284
 285         /* shutdown lquota site */
 286         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 287                 lquota_site_free(env, qqi->qqi_site);
 288                 qqi->qqi_site = NULL;
 289         }
 290
 291         /* by now, all qqi users should have gone away */
 292         LASSERT(cfs_atomic_read(&qqi->qqi_ref) == 1);
 293         lu_ref_fini(&qqi->qqi_reference);
 294
 295         /* release accounting object */
 296         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 297                 lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
 298                 qqi->qqi_acct_obj = NULL;
 299         }
 300
 301         /* release slv index */
 302         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 303                 lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
 304                 qqi->qqi_slv_obj = NULL;
 305                 qqi->qqi_slv_ver = 0;
 306         }
 307
 308         /* release global index */
 309         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 310                 lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
 311                 qqi->qqi_glb_obj = NULL;
 312                 qqi->qqi_glb_ver = 0;
 313         }
 314
 315         OBD_FREE_PTR(qqi);
 316         EXIT;
 317 }
 318
 319 /*
 320  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 321  * This opens the accounting object and initializes the proc file.
 322  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 323  * instance.
 324  *
 325  * \param env  - the environment passed by the caller
 326  * \param qsd  - is the qsd instance which will be in charge of the new
 327  *               qsd_qtype_info instance.
 328  * \param qtype - is quota type to set up
 329  *
 330  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 331  *           appropriate error on failure
 332  */
 333 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 334                           int qtype)
 335 {
 336         struct qsd_qtype_info   *qqi;
 337         int                      rc;
 338         struct obd_uuid          uuid;
 339         ENTRY;
 340
 341         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 342
 343         /* allocate structure for this quota type */
 344         OBD_ALLOC_PTR(qqi);
 345         if (qqi == NULL)
 346                 RETURN(-ENOMEM);
 347         qsd->qsd_type_array[qtype] = qqi;
 348         cfs_atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 349
 350         /* set backpointer and other parameters */
 351         qqi->qqi_qsd   = qsd;
 352         qqi->qqi_qtype = qtype;
 353         lu_ref_init(&qqi->qqi_reference);
 354         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id, QSD_RES_TYPE(qsd),
 355                             qtype);
 356         qqi->qqi_glb_uptodate = false;
 357         qqi->qqi_slv_uptodate = false;
 358         qqi->qqi_reint        = false;
 359         cfs_waitq_init(&qqi->qqi_reint_thread.t_ctl_waitq);
 360         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 361         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 362         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 363
 364         /* open accounting object */
 365         LASSERT(qqi->qqi_acct_obj == NULL);
 366         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 367         if (IS_ERR(qqi->qqi_acct_obj)) {
 368                 CDEBUG(D_QUOTA, "%s: no %s space accounting support rc:%ld\n",
 369                        qsd->qsd_svname, QTYPE_NAME(qtype),
 370                        PTR_ERR(qqi->qqi_acct_obj));
 371                 qqi->qqi_acct_obj = NULL;
 372                 qsd->qsd_acct_failed = true;
 373         }
 374
 375         /* open global index copy */
 376         LASSERT(qqi->qqi_glb_obj == NULL);
 377         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 378                                                        qsd->qsd_root,
 379                                                        &qqi->qqi_fid, true);
 380         if (IS_ERR(qqi->qqi_glb_obj)) {
 381                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 382                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 383                        PTR_ERR(qqi->qqi_glb_obj));
 384                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 385         }
 386         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 387
 388         /* open slave index copy */
 389         LASSERT(qqi->qqi_slv_obj == NULL);
 390         obd_str2uuid(&uuid, qsd->qsd_svname);
 391         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 392                                                        qsd->qsd_root,
 393                                                        &qqi->qqi_fid, &uuid,
 394                                                        true);
 395         if (IS_ERR(qqi->qqi_slv_obj)) {
 396                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 397                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 398                        PTR_ERR(qqi->qqi_slv_obj));
 399                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 400         }
 401         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 402
 403         /* allocate site */
 404         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 405         if (IS_ERR(qqi->qqi_site)) {
 406                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 407                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 408                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 409         }
 410
 411         /* register proc entry for accounting & global index copy objects */
 412         rc = lprocfs_seq_create(qsd->qsd_proc,
 413                                 qtype == USRQUOTA ? "acct_user" : "acct_group",
 414                                 0444, &lprocfs_quota_seq_fops,
 415                                 qqi->qqi_acct_obj);
 416         if (rc) {
 417                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 418                        qsd->qsd_svname, rc);
 419                 GOTO(out, rc);
 420         }
 421
 422         rc = lprocfs_seq_create(qsd->qsd_proc,
 423                                 qtype == USRQUOTA ? "limit_user" : "limit_group",
 424                                 0444, &lprocfs_quota_seq_fops,
 425                                 qqi->qqi_glb_obj);
 426         if (rc) {
 427                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 428                        qsd->qsd_svname, rc);
 429                 GOTO(out, rc);
 430         }
 431         EXIT;
 432 out:
 433         if (rc)
 434                 qsd_qtype_fini(env, qsd, qtype);
 435         return rc;
 436 }
 437
 438 /*
 439  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 440  * structures associated with the quota slave (on-disk objects, lquota entry
 441  * tables, ...).
 442  * This function should be called when the OSD is shutting down.
 443  *
 444  * \param env - is the environment passed by the caller
 445  * \param qsd - is the qsd instance to shutdown
 446  */
 447 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 448 {
 449         int     qtype;
 450         ENTRY;
 451
 452         if (unlikely(qsd == NULL))
 453                 RETURN_EXIT;
 454
 455         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 456         write_lock(&qsd->qsd_lock);
 457         qsd->qsd_stopping = true;
 458         write_unlock(&qsd->qsd_lock);
 459
 460         /* remove qsd proc entry */
 461         if (qsd->qsd_proc != NULL) {
 462                 lprocfs_remove(&qsd->qsd_proc);
 463                 qsd->qsd_proc = NULL;
 464         }
 465
 466         /* stop the writeback thread */
 467         qsd_stop_upd_thread(qsd);
 468
 469         /* shutdown the reintegration threads */
 470         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 471                 if (qsd->qsd_type_array[qtype] == NULL)
 472                         continue;
 473                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 474         }
 475
 476         if (qsd->qsd_ns != NULL) {
 477                 qsd->qsd_ns = NULL;
 478         }
 479
 480         /* free per-quota type data */
 481         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
 482                 qsd_qtype_fini(env, qsd, qtype);
 483
 484         /* deregister connection to the quota master */
 485         qsd->qsd_exp_valid = false;
 486         lustre_deregister_lwp_item(&qsd->qsd_exp);
 487
 488         /* release per-filesystem information */
 489         if (qsd->qsd_fsinfo != NULL) {
 490                 down(&qsd->qsd_fsinfo->qfs_sem);
 491                 /* remove from the list of fsinfo */
 492                 cfs_list_del_init(&qsd->qsd_link);
 493                 up(&qsd->qsd_fsinfo->qfs_sem);
 494                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 495                 qsd->qsd_fsinfo = NULL;
 496         }
 497
 498         /* release quota root directory */
 499         if (qsd->qsd_root != NULL) {
 500                 lu_object_put(env, &qsd->qsd_root->do_lu);
 501                 qsd->qsd_root = NULL;
 502         }
 503
 504         /* release reference on dt_device */
 505         if (qsd->qsd_dev != NULL) {
 506                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 507                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 508                 qsd->qsd_dev = NULL;
 509         }
 510
 511         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 512         OBD_FREE_PTR(qsd);
 513         EXIT;
 514 }
 515 EXPORT_SYMBOL(qsd_fini);
 516
 517 /*
 518  * Create a new qsd_instance to be associated with backend osd device
 519  * identified by \dev.
 520  *
 521  * \param env    - the environment passed by the caller
 522  * \param svname - is the service name of the OSD device creating this instance
 523  * \param dev    - is the dt_device where to store quota index files
 524  * \param osd_proc - is the procfs parent directory where to create procfs file
 525  *                   related to this new qsd instance
 526  *
 527  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 528  *           appropriate error on failure
 529  */
 530 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 531                               struct dt_device *dev,
 532                               cfs_proc_dir_entry_t *osd_proc)
 533 {
 534         struct qsd_thread_info  *qti = qsd_info(env);
 535         struct qsd_instance     *qsd;
 536         int                      rc, type, idx;
 537         ENTRY;
 538
 539         /* only configure qsd for MDT & OST */
 540         type = server_name2index(svname, &idx, NULL);
 541         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 542                 RETURN(NULL);
 543
 544         /* allocate qsd instance */
 545         OBD_ALLOC_PTR(qsd);
 546         if (qsd == NULL)
 547                 RETURN(ERR_PTR(-ENOMEM));
 548
 549         /* generic initializations */
 550         rwlock_init(&qsd->qsd_lock);
 551         CFS_INIT_LIST_HEAD(&qsd->qsd_link);
 552         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 553         cfs_waitq_init(&qsd->qsd_upd_thread.t_ctl_waitq);
 554         CFS_INIT_LIST_HEAD(&qsd->qsd_upd_list);
 555         spin_lock_init(&qsd->qsd_adjust_lock);
 556         CFS_INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 557         qsd->qsd_prepared = false;
 558         qsd->qsd_started = false;
 559
 560         /* copy service name */
 561         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 562             >= sizeof(qsd->qsd_svname))
 563                 GOTO(out, rc = -E2BIG);
 564
 565         /* grab reference on osd device */
 566         lu_device_get(&dev->dd_lu_dev);
 567         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 568         qsd->qsd_dev = dev;
 569
 570         /* we only support pool ID 0 (default data or metadata pool) for the
 571          * time being. A different pool ID could be assigned to this target via
 572          * the configuration log in the future */
 573         qsd->qsd_pool_id  = 0;
 574
 575         /* get fsname from svname */
 576         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 577         if (rc) {
 578                 CERROR("%s: fail to extract filesystem name\n", svname);
 579                 GOTO(out, rc);
 580         }
 581
 582         /* look up quota setting for the filesystem the target belongs to */
 583         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 584         if (qsd->qsd_fsinfo == NULL) {
 585                 CERROR("%s: failed to locate filesystem information\n", svname);
 586                 GOTO(out, rc = -EINVAL);
 587         }
 588
 589         /* add in the list of lquota_fsinfo */
 590         down(&qsd->qsd_fsinfo->qfs_sem);
 591         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 592         up(&qsd->qsd_fsinfo->qfs_sem);
 593
 594         /* register procfs directory */
 595         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 596                                          lprocfs_quota_qsd_vars, qsd);
 597         if (IS_ERR(qsd->qsd_proc)) {
 598                 rc = PTR_ERR(qsd->qsd_proc);
 599                 qsd->qsd_proc = NULL;
 600                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 601                        svname, rc);
 602                 GOTO(out, rc);
 603         }
 604         EXIT;
 605 out:
 606         if (rc) {
 607                 qsd_fini(env, qsd);
 608                 return ERR_PTR(rc);
 609         }
 610         RETURN(qsd);
 611 }
 612 EXPORT_SYMBOL(qsd_init);
 613
 614 /*
 615  * Initialize on-disk structures in order to manage quota enforcement for
 616  * the target associated with the qsd instance \qsd and starts the reintegration
 617  * procedure for each quota type as soon as possible.
 618  * The last step of the reintegration will be completed once qsd_start() is
 619  * called, at which points the space reconciliation with the master will be
 620  * executed.
 621  * This function must be called when the server stack is fully configured,
 622  * typically when ->ldo_prepare is called across the stack.
 623  *
 624  * \param env - the environment passed by the caller
 625  * \param qsd - is qsd_instance to prepare
 626  *
 627  * \retval - 0 on success, appropriate error on failure
 628  */
 629 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 630 {
 631         struct qsd_thread_info  *qti = qsd_info(env);
 632         int                      qtype, rc = 0;
 633         ENTRY;
 634
 635         if (unlikely(qsd == NULL))
 636                 RETURN(0);
 637
 638         read_lock(&qsd->qsd_lock);
 639         if (qsd->qsd_prepared) {
 640                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 641                 rc = -EALREADY;
 642         }
 643         read_unlock(&qsd->qsd_lock);
 644         if (rc)
 645                 RETURN(rc);
 646
 647         /* Record whether this qsd instance is managing quota enforcement for a
 648          * MDT (i.e. inode quota) or OST (block quota) */
 649         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 650                 qsd->qsd_is_md = true;
 651                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 652         } else {
 653                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 654         }
 655
 656         /* look-up on-disk directory for the quota slave */
 657         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 658                                                     QSD_DIR);
 659         if (IS_ERR(qsd->qsd_root)) {
 660                 rc = PTR_ERR(qsd->qsd_root);
 661                 qsd->qsd_root = NULL;
 662                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 663                        qsd->qsd_svname, rc);
 664                 RETURN(rc);
 665         }
 666
 667         /* initialize per-quota type data */
 668         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 669                 rc = qsd_qtype_init(env, qsd, qtype);
 670                 if (rc)
 671                         RETURN(rc);
 672         }
 673
 674         /* pools successfully setup, mark the qsd as prepared */
 675         write_lock(&qsd->qsd_lock);
 676         qsd->qsd_prepared = true;
 677         write_unlock(&qsd->qsd_lock);
 678
 679         /* start reintegration thread for each type, if required */
 680         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 681                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 682
 683                 if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
 684                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 685                                        "since space accounting isn't functional"
 686                                        ". Please run tunefs.lustre --quota on "
 687                                        "an unmounted filesystem if not done "
 688                                        "already\n", qsd->qsd_svname);
 689                         break;
 690                 }
 691
 692                 rc = qsd_start_reint_thread(qqi);
 693                 if (rc) {
 694                         CERROR("%s: failed to start reint thread for type %s "
 695                                "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
 696                                rc);
 697                         RETURN(rc);
 698                 }
 699         }
 700
 701         /* start writeback thread */
 702         rc = qsd_start_upd_thread(qsd);
 703         if (rc) {
 704                 CERROR("%s: failed to start writeback thread (%d)\n",
 705                        qsd->qsd_svname, rc);
 706                 RETURN(rc);
 707         }
 708
 709         /* generate osp name */
 710         rc = tgt_name2lwpname((char *)qsd->qsd_svname, qti->qti_buf);
 711         if (rc) {
 712                 CERROR("%s: failed to generate ospname (%d)\n",
 713                        qsd->qsd_svname, rc);
 714                 RETURN(rc);
 715         }
 716
 717         /* the connection callback will start the reintegration
 718          * procedure if quota is enabled */
 719         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 720                                       qsd_conn_callback, (void *)qsd);
 721         if (rc) {
 722                 CERROR("%s: fail to get connection to master (%d)\n",
 723                        qsd->qsd_svname, rc);
 724                 RETURN(rc);
 725         }
 726
 727         RETURN(0);
 728 }
 729 EXPORT_SYMBOL(qsd_prepare);
 730
 731 /*
 732  * Start a qsd instance. This will complete the last step of the reintegration
 733  * procedure as soon as possible (provided that the master is reachable).
 734  * This should be called when recovery has been completed and quota should now
 735  * be enforced on every operations.
 736  *
 737  * \param env - the environment passed by the caller
 738  * \param qsd - is the qsd instance associated with the osd device to start
 739  */
 740 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 741 {
 742         int     type, rc = 0;
 743         ENTRY;
 744
 745         if (unlikely(qsd == NULL))
 746                 RETURN(0);
 747
 748         write_lock(&qsd->qsd_lock);
 749         if (!qsd->qsd_prepared) {
 750                 CERROR("%s: can't start qsd instance since it wasn't properly "
 751                        "initialized\n", qsd->qsd_svname);
 752                 rc = -EFAULT;
 753         } else if (qsd->qsd_started) {
 754                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 755                 rc = -EALREADY;
 756         } else {
 757                 /* notify that the qsd_instance is now started */
 758                 qsd->qsd_started = true;
 759         }
 760         write_unlock(&qsd->qsd_lock);
 761
 762         if (rc)
 763                 RETURN(rc);
 764
 765         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 766          * up to usage; If usage < granted, release down to usage.  */
 767         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 768                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 769                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 770         }
 771
 772         RETURN(rc);
 773 }
 774 EXPORT_SYMBOL(qsd_start);
 775
 776 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 777
 778 /*
 779  * Global initialization performed at module load time
 780  */
 781 int qsd_glb_init(void)
 782 {
 783         int     rc;
 784
 785         rc = lu_kmem_init(qsd_caches);
 786         if (rc)
 787                 return rc;
 788
 789         qsd_key_init_generic(&qsd_thread_key, NULL);
 790         lu_context_key_register(&qsd_thread_key);
 791         lustre_register_quota_process_config(qsd_process_config);
 792
 793         return 0;
 794 }
 795
 796 /*
 797  * Companion of qsd_glb_init() called at module unload time
 798  */
 799 void qsd_glb_fini(void)
 800 {
 801         lustre_register_quota_process_config(NULL);
 802         lu_kmem_fini(qsd_caches);
 803         lu_context_key_degister(&qsd_thread_key);
 804 }