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 #define DEBUG_SUBSYSTEM S_LQUOTA
  53
  54 #include <obd_class.h>
  55 #include "qsd_internal.h"
  56
  57 struct kmem_cache *upd_kmem;
  58
  59 struct lu_kmem_descr qsd_caches[] = {
  60         {
  61                 .ckd_cache = &upd_kmem,
  62                 .ckd_name  = "upd_kmem",
  63                 .ckd_size  = sizeof(struct qsd_upd_rec)
  64         },
  65         {
  66                 .ckd_cache = NULL
  67         }
  68 };
  69
  70 /* define qsd thread key */
  71 LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
  72 LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
  73 LU_KEY_INIT_GENERIC(qsd);
  74
  75 /* some procfs helpers */
  76 static int lprocfs_qsd_rd_state(char *page, char **start, off_t off,
  77                                 int count, int *eof, void *data)
  78 {
  79         struct qsd_instance     *qsd = (struct qsd_instance *)data;
  80         char                     enabled[5];
  81         int                      rc;
  82
  83         LASSERT(qsd != NULL);
  84
  85         memset(enabled, 0, sizeof(enabled));
  86         if (qsd_type_enabled(qsd, USRQUOTA))
  87                 strcat(enabled, "u");
  88         if (qsd_type_enabled(qsd, GRPQUOTA))
  89                 strcat(enabled, "g");
  90         if (strlen(enabled) == 0)
  91                 strcat(enabled, "none");
  92
  93         rc = snprintf(page, count,
  94                       "target name:    %s\n"
  95                       "pool ID:        %d\n"
  96                       "type:           %s\n"
  97                       "quota enabled:  %s\n"
  98                       "conn to master: %s\n",
  99                       qsd->qsd_svname, qsd->qsd_pool_id,
 100                       qsd->qsd_is_md ? "md" : "dt", enabled,
 101                       qsd->qsd_exp_valid ? "setup" : "not setup yet");
 102
 103         if (qsd->qsd_prepared) {
 104                 memset(enabled, 0, sizeof(enabled));
 105                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 106                         strcat(enabled, "u");
 107                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 108                         strcat(enabled, "g");
 109                 if (strlen(enabled) == 0)
 110                         strcat(enabled, "none");
 111                 rc +=  snprintf(page + rc, count - rc,
 112                                 "space acct:     %s\n"
 113                                 "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 114                                 "group uptodate: glb[%d],slv[%d],reint[%d]\n",
 115                                 enabled,
 116                                 qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 117                                 qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 118                                 qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 119                                 qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 120                                 qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 121                                 qsd->qsd_type_array[GRPQUOTA]->qqi_reint);
 122         }
 123         return rc;
 124 }
 125
 126 static int lprocfs_qsd_rd_enabled(char *page, char **start, off_t off,
 127                                   int count, int *eof, void *data)
 128 {
 129         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 130         char                     enabled[5];
 131
 132         LASSERT(qsd != NULL);
 133
 134         memset(enabled, 0, sizeof(enabled));
 135         if (qsd_type_enabled(qsd, USRQUOTA))
 136                 strcat(enabled, "u");
 137         if (qsd_type_enabled(qsd, GRPQUOTA))
 138                 strcat(enabled, "g");
 139         if (strlen(enabled) == 0)
 140                 strcat(enabled, "none");
 141
 142         return snprintf(page, count, "%s\n", enabled);
 143 }
 144
 145 /* force reintegration procedure to be executed.
 146  * Used for test/debugging purpose */
 147 static int lprocfs_qsd_wr_force_reint(struct file *file, const char *buffer,
 148                                       unsigned long count, void *data)
 149 {
 150         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 151         int                      rc = 0, qtype;
 152
 153         LASSERT(qsd != NULL);
 154
 155         write_lock(&qsd->qsd_lock);
 156         if (qsd->qsd_stopping) {
 157                 /* don't mess up with shutdown procedure, it is already
 158                  * complicated enough */
 159                 rc = -ESHUTDOWN;
 160         } else if (!qsd->qsd_prepared) {
 161                 rc = -EAGAIN;
 162         } else {
 163                 /* mark all indexes as stale */
 164                 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 165                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 166                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 167                 }
 168         }
 169         write_unlock(&qsd->qsd_lock);
 170
 171         if (rc)
 172                 return rc;
 173
 174         /* kick off reintegration */
 175         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 176                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 177                 if (rc)
 178                         break;
 179         }
 180         return rc == 0 ? count : rc;
 181 }
 182
 183 static int lprocfs_qsd_rd_timeout(char *page, char **start, off_t off,
 184                                   int count, int *eof, void *data)
 185 {
 186         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 187         LASSERT(qsd != NULL);
 188
 189         return snprintf(page, count, "%d\n", qsd_wait_timeout(qsd));
 190 }
 191
 192 static int lprocfs_qsd_wr_timeout(struct file *file, const char *buffer,
 193                                   unsigned long count, void *data)
 194 {
 195         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 196         int                      timeout, rc;
 197         LASSERT(qsd != NULL);
 198
 199         rc = lprocfs_write_helper(buffer, count, &timeout);
 200         if (rc)
 201                 return rc;
 202         if (timeout < 0)
 203                 return -EINVAL;
 204
 205         qsd->qsd_timeout = timeout;
 206         return count;
 207 }
 208
 209 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 210         { "info", lprocfs_qsd_rd_state, 0, 0},
 211         { "enabled", lprocfs_qsd_rd_enabled, 0, 0},
 212         { "force_reint", 0, lprocfs_qsd_wr_force_reint, 0},
 213         { "timeout", lprocfs_qsd_rd_timeout, lprocfs_qsd_wr_timeout, 0},
 214         { NULL }
 215 };
 216
 217 /*
 218  * Callback function invoked by the OSP layer when the connection to the master
 219  * has been set up.
 220  *
 221  * \param data - is a pointer to the qsd_instance
 222  *
 223  * \retval - 0 on success, appropriate error on failure
 224  */
 225 static int qsd_conn_callback(void *data)
 226 {
 227         struct qsd_instance *qsd = (struct qsd_instance *)data;
 228         int                  type;
 229         ENTRY;
 230
 231         /* qsd_exp should now be valid */
 232         LASSERT(qsd->qsd_exp);
 233
 234         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 235
 236         write_lock(&qsd->qsd_lock);
 237         /* notify that qsd_exp is now valid */
 238         qsd->qsd_exp_valid = true;
 239         write_unlock(&qsd->qsd_lock);
 240
 241         /* Now that the connection to master is setup, we can initiate the
 242          * reintegration procedure for quota types which are enabled.
 243          * It is worth noting that, if the qsd_instance hasn't been started
 244          * already, then we can only complete the first two steps of the
 245          * reintegration procedure (i.e. global lock enqueue and slave
 246          * index transfer) since the space usage reconciliation (i.e.
 247          * step 3) will have to wait for qsd_start() to be called */
 248         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 249                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 250                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 251         }
 252
 253         RETURN(0);
 254 }
 255
 256 /*
 257  * Release qsd_qtype_info structure which contains data associated with a
 258  * given quota type. This releases the accounting objects.
 259  * It's called on OSD cleanup when the qsd instance is released.
 260  *
 261  * \param env - is the environment passed by the caller
 262  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 263  *              to be released
 264  * \param qtype - is the quota type to be shutdown
 265  */
 266 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 267                            int qtype)
 268 {
 269         struct qsd_qtype_info   *qqi;
 270         ENTRY;
 271
 272         if (qsd->qsd_type_array[qtype] == NULL)
 273                 RETURN_EXIT;
 274         qqi = qsd->qsd_type_array[qtype];
 275         qsd->qsd_type_array[qtype] = NULL;
 276
 277         /* all deferred work lists should be empty */
 278         LASSERT(cfs_list_empty(&qqi->qqi_deferred_glb));
 279         LASSERT(cfs_list_empty(&qqi->qqi_deferred_slv));
 280
 281         /* shutdown lquota site */
 282         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 283                 lquota_site_free(env, qqi->qqi_site);
 284                 qqi->qqi_site = NULL;
 285         }
 286
 287         /* by now, all qqi users should have gone away */
 288         LASSERT(cfs_atomic_read(&qqi->qqi_ref) == 1);
 289         lu_ref_fini(&qqi->qqi_reference);
 290
 291         /* release accounting object */
 292         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 293                 lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
 294                 qqi->qqi_acct_obj = NULL;
 295         }
 296
 297         /* release slv index */
 298         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 299                 lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
 300                 qqi->qqi_slv_obj = NULL;
 301                 qqi->qqi_slv_ver = 0;
 302         }
 303
 304         /* release global index */
 305         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 306                 lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
 307                 qqi->qqi_glb_obj = NULL;
 308                 qqi->qqi_glb_ver = 0;
 309         }
 310
 311         OBD_FREE_PTR(qqi);
 312         EXIT;
 313 }
 314
 315 /*
 316  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 317  * This opens the accounting object and initializes the proc file.
 318  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 319  * instance.
 320  *
 321  * \param env  - the environment passed by the caller
 322  * \param qsd  - is the qsd instance which will be in charge of the new
 323  *               qsd_qtype_info instance.
 324  * \param qtype - is quota type to set up
 325  *
 326  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 327  *           appropriate error on failure
 328  */
 329 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 330                           int qtype)
 331 {
 332         struct qsd_qtype_info   *qqi;
 333         int                      rc;
 334         struct obd_uuid          uuid;
 335         ENTRY;
 336
 337         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 338
 339         /* allocate structure for this quota type */
 340         OBD_ALLOC_PTR(qqi);
 341         if (qqi == NULL)
 342                 RETURN(-ENOMEM);
 343         qsd->qsd_type_array[qtype] = qqi;
 344         cfs_atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 345
 346         /* set backpointer and other parameters */
 347         qqi->qqi_qsd   = qsd;
 348         qqi->qqi_qtype = qtype;
 349         lu_ref_init(&qqi->qqi_reference);
 350         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id, QSD_RES_TYPE(qsd),
 351                             qtype);
 352         qqi->qqi_glb_uptodate = false;
 353         qqi->qqi_slv_uptodate = false;
 354         qqi->qqi_reint        = false;
 355         cfs_waitq_init(&qqi->qqi_reint_thread.t_ctl_waitq);
 356         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 357         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 358         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 359
 360         /* open accounting object */
 361         LASSERT(qqi->qqi_acct_obj == NULL);
 362         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 363         if (IS_ERR(qqi->qqi_acct_obj)) {
 364                 CDEBUG(D_QUOTA, "%s: no %s space accounting support rc:%ld\n",
 365                        qsd->qsd_svname, QTYPE_NAME(qtype),
 366                        PTR_ERR(qqi->qqi_acct_obj));
 367                 qqi->qqi_acct_obj = NULL;
 368                 qsd->qsd_acct_failed = true;
 369         }
 370
 371         /* open global index copy */
 372         LASSERT(qqi->qqi_glb_obj == NULL);
 373         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 374                                                        qsd->qsd_root,
 375                                                        &qqi->qqi_fid, true);
 376         if (IS_ERR(qqi->qqi_glb_obj)) {
 377                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 378                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 379                        PTR_ERR(qqi->qqi_glb_obj));
 380                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 381         }
 382         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 383
 384         /* open slave index copy */
 385         LASSERT(qqi->qqi_slv_obj == NULL);
 386         obd_str2uuid(&uuid, qsd->qsd_svname);
 387         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 388                                                        qsd->qsd_root,
 389                                                        &qqi->qqi_fid, &uuid,
 390                                                        true);
 391         if (IS_ERR(qqi->qqi_slv_obj)) {
 392                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 393                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 394                        PTR_ERR(qqi->qqi_slv_obj));
 395                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 396         }
 397         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 398
 399         /* allocate site */
 400         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 401         if (IS_ERR(qqi->qqi_site)) {
 402                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 403                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 404                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 405         }
 406
 407         /* register proc entry for accounting & global index copy objects */
 408         rc = lprocfs_seq_create(qsd->qsd_proc,
 409                                 qtype == USRQUOTA ? "acct_user" : "acct_group",
 410                                 0444, &lprocfs_quota_seq_fops,
 411                                 qqi->qqi_acct_obj);
 412         if (rc) {
 413                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 414                        qsd->qsd_svname, rc);
 415                 GOTO(out, rc);
 416         }
 417
 418         rc = lprocfs_seq_create(qsd->qsd_proc,
 419                                 qtype == USRQUOTA ? "limit_user" : "limit_group",
 420                                 0444, &lprocfs_quota_seq_fops,
 421                                 qqi->qqi_glb_obj);
 422         if (rc) {
 423                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 424                        qsd->qsd_svname, rc);
 425                 GOTO(out, rc);
 426         }
 427         EXIT;
 428 out:
 429         if (rc)
 430                 qsd_qtype_fini(env, qsd, qtype);
 431         return rc;
 432 }
 433
 434 /*
 435  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 436  * structures associated with the quota slave (on-disk objects, lquota entry
 437  * tables, ...).
 438  * This function should be called when the OSD is shutting down.
 439  *
 440  * \param env - is the environment passed by the caller
 441  * \param qsd - is the qsd instance to shutdown
 442  */
 443 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 444 {
 445         int     qtype;
 446         ENTRY;
 447
 448         if (unlikely(qsd == NULL))
 449                 RETURN_EXIT;
 450
 451         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 452         write_lock(&qsd->qsd_lock);
 453         qsd->qsd_stopping = true;
 454         write_unlock(&qsd->qsd_lock);
 455
 456         /* remove qsd proc entry */
 457         if (qsd->qsd_proc != NULL) {
 458                 lprocfs_remove(&qsd->qsd_proc);
 459                 qsd->qsd_proc = NULL;
 460         }
 461
 462         /* stop the writeback thread */
 463         qsd_stop_upd_thread(qsd);
 464
 465         /* shutdown the reintegration threads */
 466         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 467                 if (qsd->qsd_type_array[qtype] == NULL)
 468                         continue;
 469                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 470         }
 471
 472         if (qsd->qsd_ns != NULL) {
 473                 qsd->qsd_ns = NULL;
 474         }
 475
 476         /* free per-quota type data */
 477         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
 478                 qsd_qtype_fini(env, qsd, qtype);
 479
 480         /* deregister connection to the quota master */
 481         qsd->qsd_exp_valid = false;
 482         lustre_deregister_lwp_item(&qsd->qsd_exp);
 483
 484         /* release per-filesystem information */
 485         if (qsd->qsd_fsinfo != NULL) {
 486                 down(&qsd->qsd_fsinfo->qfs_sem);
 487                 /* remove from the list of fsinfo */
 488                 cfs_list_del_init(&qsd->qsd_link);
 489                 up(&qsd->qsd_fsinfo->qfs_sem);
 490                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 491                 qsd->qsd_fsinfo = NULL;
 492         }
 493
 494         /* release quota root directory */
 495         if (qsd->qsd_root != NULL) {
 496                 lu_object_put(env, &qsd->qsd_root->do_lu);
 497                 qsd->qsd_root = NULL;
 498         }
 499
 500         /* release reference on dt_device */
 501         if (qsd->qsd_dev != NULL) {
 502                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 503                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 504                 qsd->qsd_dev = NULL;
 505         }
 506
 507         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 508         OBD_FREE_PTR(qsd);
 509         EXIT;
 510 }
 511 EXPORT_SYMBOL(qsd_fini);
 512
 513 /*
 514  * Create a new qsd_instance to be associated with backend osd device
 515  * identified by \dev.
 516  *
 517  * \param env    - the environment passed by the caller
 518  * \param svname - is the service name of the OSD device creating this instance
 519  * \param dev    - is the dt_device where to store quota index files
 520  * \param osd_proc - is the procfs parent directory where to create procfs file
 521  *                   related to this new qsd instance
 522  *
 523  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 524  *           appropriate error on failure
 525  */
 526 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 527                               struct dt_device *dev,
 528                               cfs_proc_dir_entry_t *osd_proc)
 529 {
 530         struct qsd_thread_info  *qti = qsd_info(env);
 531         struct qsd_instance     *qsd;
 532         int                      rc, type, idx;
 533         ENTRY;
 534
 535         /* only configure qsd for MDT & OST */
 536         type = server_name2index(svname, &idx, NULL);
 537         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 538                 RETURN(ERR_PTR(-EINVAL));
 539
 540         /* allocate qsd instance */
 541         OBD_ALLOC_PTR(qsd);
 542         if (qsd == NULL)
 543                 RETURN(ERR_PTR(-ENOMEM));
 544
 545         /* generic initializations */
 546         rwlock_init(&qsd->qsd_lock);
 547         CFS_INIT_LIST_HEAD(&qsd->qsd_link);
 548         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 549         cfs_waitq_init(&qsd->qsd_upd_thread.t_ctl_waitq);
 550         CFS_INIT_LIST_HEAD(&qsd->qsd_upd_list);
 551         spin_lock_init(&qsd->qsd_adjust_lock);
 552         CFS_INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 553         qsd->qsd_prepared = false;
 554         qsd->qsd_started = false;
 555
 556         /* copy service name */
 557         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 558             >= sizeof(qsd->qsd_svname))
 559                 GOTO(out, rc = -E2BIG);
 560
 561         /* grab reference on osd device */
 562         lu_device_get(&dev->dd_lu_dev);
 563         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 564         qsd->qsd_dev = dev;
 565
 566         /* we only support pool ID 0 (default data or metadata pool) for the
 567          * time being. A different pool ID could be assigned to this target via
 568          * the configuration log in the future */
 569         qsd->qsd_pool_id  = 0;
 570
 571         /* get fsname from svname */
 572         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 573         if (rc) {
 574                 CERROR("%s: fail to extract filesystem name\n", svname);
 575                 GOTO(out, rc);
 576         }
 577
 578         /* look up quota setting for the filesystem the target belongs to */
 579         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 580         if (qsd->qsd_fsinfo == NULL) {
 581                 CERROR("%s: failed to locate filesystem information\n", svname);
 582                 GOTO(out, rc = -EINVAL);
 583         }
 584
 585         /* add in the list of lquota_fsinfo */
 586         down(&qsd->qsd_fsinfo->qfs_sem);
 587         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 588         up(&qsd->qsd_fsinfo->qfs_sem);
 589
 590         /* register procfs directory */
 591         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 592                                          lprocfs_quota_qsd_vars, qsd);
 593         if (IS_ERR(qsd->qsd_proc)) {
 594                 rc = PTR_ERR(qsd->qsd_proc);
 595                 qsd->qsd_proc = NULL;
 596                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 597                        svname, rc);
 598                 GOTO(out, rc);
 599         }
 600         EXIT;
 601 out:
 602         if (rc) {
 603                 qsd_fini(env, qsd);
 604                 return ERR_PTR(rc);
 605         }
 606         RETURN(qsd);
 607 }
 608 EXPORT_SYMBOL(qsd_init);
 609
 610 /*
 611  * Initialize on-disk structures in order to manage quota enforcement for
 612  * the target associated with the qsd instance \qsd and starts the reintegration
 613  * procedure for each quota type as soon as possible.
 614  * The last step of the reintegration will be completed once qsd_start() is
 615  * called, at which points the space reconciliation with the master will be
 616  * executed.
 617  * This function must be called when the server stack is fully configured,
 618  * typically when ->ldo_prepare is called across the stack.
 619  *
 620  * \param env - the environment passed by the caller
 621  * \param qsd - is qsd_instance to prepare
 622  *
 623  * \retval - 0 on success, appropriate error on failure
 624  */
 625 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 626 {
 627         struct qsd_thread_info  *qti = qsd_info(env);
 628         int                      qtype, rc = 0;
 629         ENTRY;
 630
 631         if (unlikely(qsd == NULL))
 632                 RETURN(0);
 633
 634         read_lock(&qsd->qsd_lock);
 635         if (qsd->qsd_prepared) {
 636                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 637                 rc = -EALREADY;
 638         }
 639         read_unlock(&qsd->qsd_lock);
 640         if (rc)
 641                 RETURN(rc);
 642
 643         /* Record whether this qsd instance is managing quota enforcement for a
 644          * MDT (i.e. inode quota) or OST (block quota) */
 645         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 646                 qsd->qsd_is_md = true;
 647                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 648         } else {
 649                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 650         }
 651
 652         /* look-up on-disk directory for the quota slave */
 653         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 654                                                     QSD_DIR);
 655         if (IS_ERR(qsd->qsd_root)) {
 656                 rc = PTR_ERR(qsd->qsd_root);
 657                 qsd->qsd_root = NULL;
 658                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 659                        qsd->qsd_svname, rc);
 660                 RETURN(rc);
 661         }
 662
 663         /* initialize per-quota type data */
 664         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 665                 rc = qsd_qtype_init(env, qsd, qtype);
 666                 if (rc)
 667                         RETURN(rc);
 668         }
 669
 670         /* pools successfully setup, mark the qsd as prepared */
 671         write_lock(&qsd->qsd_lock);
 672         qsd->qsd_prepared = true;
 673         write_unlock(&qsd->qsd_lock);
 674
 675         /* start reintegration thread for each type, if required */
 676         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 677                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 678
 679                 if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
 680                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 681                                        "since space accounting isn't functional"
 682                                        ". Please run tunefs.lustre --quota on "
 683                                        "an unmounted filesystem if not done "
 684                                        "already\n", qsd->qsd_svname);
 685                         break;
 686                 }
 687
 688                 rc = qsd_start_reint_thread(qqi);
 689                 if (rc) {
 690                         CERROR("%s: failed to start reint thread for type %s "
 691                                "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
 692                                rc);
 693                         RETURN(rc);
 694                 }
 695         }
 696
 697         /* start writeback thread */
 698         rc = qsd_start_upd_thread(qsd);
 699         if (rc) {
 700                 CERROR("%s: failed to start writeback thread (%d)\n",
 701                        qsd->qsd_svname, rc);
 702                 RETURN(rc);
 703         }
 704
 705         /* generate osp name */
 706         rc = tgt_name2lwpname((char *)qsd->qsd_svname, qti->qti_buf);
 707         if (rc) {
 708                 CERROR("%s: failed to generate ospname (%d)\n",
 709                        qsd->qsd_svname, rc);
 710                 RETURN(rc);
 711         }
 712
 713         /* the connection callback will start the reintegration
 714          * procedure if quota is enabled */
 715         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 716                                       qsd_conn_callback, (void *)qsd);
 717         if (rc) {
 718                 CERROR("%s: fail to get connection to master (%d)\n",
 719                        qsd->qsd_svname, rc);
 720                 RETURN(rc);
 721         }
 722
 723         RETURN(0);
 724 }
 725 EXPORT_SYMBOL(qsd_prepare);
 726
 727 /*
 728  * Start a qsd instance. This will complete the last step of the reintegration
 729  * procedure as soon as possible (provided that the master is reachable).
 730  * This should be called when recovery has been completed and quota should now
 731  * be enforced on every operations.
 732  *
 733  * \param env - the environment passed by the caller
 734  * \param qsd - is the qsd instance associated with the osd device to start
 735  */
 736 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 737 {
 738         int     type, rc = 0;
 739         ENTRY;
 740
 741         if (unlikely(qsd == NULL))
 742                 RETURN(0);
 743
 744         write_lock(&qsd->qsd_lock);
 745         if (!qsd->qsd_prepared) {
 746                 CERROR("%s: can't start qsd instance since it wasn't properly "
 747                        "initialized\n", qsd->qsd_svname);
 748                 rc = -EFAULT;
 749         } else if (qsd->qsd_started) {
 750                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 751                 rc = -EALREADY;
 752         } else {
 753                 /* notify that the qsd_instance is now started */
 754                 qsd->qsd_started = true;
 755         }
 756         write_unlock(&qsd->qsd_lock);
 757
 758         if (rc)
 759                 RETURN(rc);
 760
 761         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 762          * up to usage; If usage < granted, release down to usage.  */
 763         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 764                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 765                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 766         }
 767
 768         RETURN(rc);
 769 }
 770 EXPORT_SYMBOL(qsd_start);
 771
 772 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 773
 774 /*
 775  * Global initialization performed at module load time
 776  */
 777 int qsd_glb_init(void)
 778 {
 779         int     rc;
 780
 781         rc = lu_kmem_init(qsd_caches);
 782         if (rc)
 783                 return rc;
 784
 785         qsd_key_init_generic(&qsd_thread_key, NULL);
 786         lu_context_key_register(&qsd_thread_key);
 787         lustre_register_quota_process_config(qsd_process_config);
 788
 789         return 0;
 790 }
 791
 792 /*
 793  * Companion of qsd_glb_init() called at module unload time
 794  */
 795 void qsd_glb_fini(void)
 796 {
 797         lustre_register_quota_process_config(NULL);
 798         lu_kmem_fini(qsd_caches);
 799         lu_context_key_degister(&qsd_thread_key);
 800 }