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, Inc.
  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                 rc +=  snprintf(page + rc, count - rc,
 109                                 "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 110                                 "group uptodate: glb[%d],slv[%d],reint[%d]\n",
 111                                 qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 112                                 qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 113                                 qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 114                                 qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 115                                 qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 116                                 qsd->qsd_type_array[GRPQUOTA]->qqi_reint);
 117         return rc;
 118 }
 119
 120 static int lprocfs_qsd_rd_enabled(char *page, char **start, off_t off,
 121                                   int count, int *eof, void *data)
 122 {
 123         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 124         char                     enabled[5];
 125
 126         LASSERT(qsd != NULL);
 127
 128         memset(enabled, 0, sizeof(enabled));
 129         if (qsd_type_enabled(qsd, USRQUOTA))
 130                 strcat(enabled, "u");
 131         if (qsd_type_enabled(qsd, GRPQUOTA))
 132                 strcat(enabled, "g");
 133         if (strlen(enabled) == 0)
 134                 strcat(enabled, "none");
 135
 136         return snprintf(page, count, "%s\n", enabled);
 137 }
 138
 139 /* force reintegration procedure to be executed.
 140  * Used for test/debugging purpose */
 141 static int lprocfs_qsd_wr_force_reint(struct file *file, const char *buffer,
 142                                       unsigned long count, void *data)
 143 {
 144         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 145         int                      rc = 0, qtype;
 146
 147         LASSERT(qsd != NULL);
 148
 149         cfs_write_lock(&qsd->qsd_lock);
 150         if (qsd->qsd_stopping) {
 151                 /* don't mess up with shutdown procedure, it is already
 152                  * complicated enough */
 153                 rc = -ESHUTDOWN;
 154         } else if (!qsd->qsd_prepared) {
 155                 rc = -EAGAIN;
 156         } else {
 157                 /* mark all indexes as stale */
 158                 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 159                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 160                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 161                 }
 162         }
 163         cfs_write_unlock(&qsd->qsd_lock);
 164
 165         if (rc)
 166                 return rc;
 167
 168         /* kick off reintegration */
 169         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 170                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 171                 if (rc)
 172                         break;
 173         }
 174         return rc == 0 ? count : rc;
 175 }
 176
 177 static int lprocfs_qsd_rd_timeout(char *page, char **start, off_t off,
 178                                   int count, int *eof, void *data)
 179 {
 180         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 181         LASSERT(qsd != NULL);
 182
 183         return snprintf(page, count, "%d\n", qsd_wait_timeout(qsd));
 184 }
 185
 186 static int lprocfs_qsd_wr_timeout(struct file *file, const char *buffer,
 187                                   unsigned long count, void *data)
 188 {
 189         struct qsd_instance     *qsd = (struct qsd_instance *)data;
 190         int                      timeout, rc;
 191         LASSERT(qsd != NULL);
 192
 193         rc = lprocfs_write_helper(buffer, count, &timeout);
 194         if (rc)
 195                 return rc;
 196         if (timeout < 0)
 197                 return -EINVAL;
 198
 199         qsd->qsd_timeout = timeout;
 200         return count;
 201 }
 202
 203 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 204         { "info", lprocfs_qsd_rd_state, 0, 0},
 205         { "enabled", lprocfs_qsd_rd_enabled, 0, 0},
 206         { "force_reint", 0, lprocfs_qsd_wr_force_reint, 0},
 207         { "timeout", lprocfs_qsd_rd_timeout, lprocfs_qsd_wr_timeout, 0},
 208         { NULL }
 209 };
 210
 211 /*
 212  * Callback function invoked by the OSP layer when the connection to the master
 213  * has been set up.
 214  *
 215  * \param data - is a pointer to the qsd_instance
 216  *
 217  * \retval - 0 on success, appropriate error on failure
 218  */
 219 static int qsd_conn_callback(void *data)
 220 {
 221         struct qsd_instance *qsd = (struct qsd_instance *)data;
 222         int                  type;
 223         ENTRY;
 224
 225         /* qsd_exp should now be valid */
 226         LASSERT(qsd->qsd_exp);
 227
 228         /* grab reference on namespace */
 229         ldlm_namespace_get(class_exp2obd(qsd->qsd_exp)->obd_namespace);
 230         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 231
 232         cfs_write_lock(&qsd->qsd_lock);
 233         /* notify that qsd_exp is now valid */
 234         qsd->qsd_exp_valid = true;
 235         cfs_write_unlock(&qsd->qsd_lock);
 236
 237         /* Now that the connection to master is setup, we can initiate the
 238          * reintegration procedure for quota types which are enabled.
 239          * It is worth noting that, if the qsd_instance hasn't been started
 240          * already, then we can only complete the first two steps of the
 241          * reintegration procedure (i.e. global lock enqueue and slave
 242          * index transfer) since the space usage reconciliation (i.e.
 243          * step 3) will have to wait for qsd_start() to be called */
 244         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 245                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 246                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 247         }
 248
 249         RETURN(0);
 250 }
 251
 252 /*
 253  * Release qsd_qtype_info structure which contains data associated with a
 254  * given quota type. This releases the accounting objects.
 255  * It's called on OSD cleanup when the qsd instance is released.
 256  *
 257  * \param env - is the environment passed by the caller
 258  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 259  *              to be released
 260  * \param qtype - is the quota type to be shutdown
 261  */
 262 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 263                            int qtype)
 264 {
 265         struct qsd_qtype_info   *qqi;
 266         ENTRY;
 267
 268         if (qsd->qsd_type_array[qtype] == NULL)
 269                 RETURN_EXIT;
 270         qqi = qsd->qsd_type_array[qtype];
 271         qsd->qsd_type_array[qtype] = NULL;
 272
 273         /* all deferred work lists should be empty */
 274         LASSERT(cfs_list_empty(&qqi->qqi_deferred_glb));
 275         LASSERT(cfs_list_empty(&qqi->qqi_deferred_slv));
 276
 277         /* shutdown lquota site */
 278         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 279                 lquota_site_free(env, qqi->qqi_site);
 280                 qqi->qqi_site = NULL;
 281         }
 282
 283         /* by now, all qqi users should have gone away */
 284         LASSERT(cfs_atomic_read(&qqi->qqi_ref) == 1);
 285         lu_ref_fini(&qqi->qqi_reference);
 286
 287         /* release accounting object */
 288         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 289                 lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
 290                 qqi->qqi_acct_obj = NULL;
 291         }
 292
 293         /* release slv index */
 294         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 295                 lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
 296                 qqi->qqi_slv_obj = NULL;
 297                 qqi->qqi_slv_ver = 0;
 298         }
 299
 300         /* release global index */
 301         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 302                 lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
 303                 qqi->qqi_glb_obj = NULL;
 304                 qqi->qqi_glb_ver = 0;
 305         }
 306
 307         OBD_FREE_PTR(qqi);
 308         EXIT;
 309 }
 310
 311 /*
 312  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 313  * This opens the accounting object and initializes the proc file.
 314  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 315  * instance.
 316  *
 317  * \param env  - the environment passed by the caller
 318  * \param qsd  - is the qsd instance which will be in charge of the new
 319  *               qsd_qtype_info instance.
 320  * \param qtype - is quota type to set up
 321  *
 322  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 323  *           appropriate error on failure
 324  */
 325 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 326                           int qtype)
 327 {
 328         struct qsd_qtype_info   *qqi;
 329         int                      rc;
 330         struct obd_uuid          uuid;
 331         ENTRY;
 332
 333         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 334
 335         /* allocate structure for this quota type */
 336         OBD_ALLOC_PTR(qqi);
 337         if (qqi == NULL)
 338                 RETURN(-ENOMEM);
 339         qsd->qsd_type_array[qtype] = qqi;
 340         cfs_atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 341
 342         /* set backpointer and other parameters */
 343         qqi->qqi_qsd   = qsd;
 344         qqi->qqi_qtype = qtype;
 345         lu_ref_init(&qqi->qqi_reference);
 346         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id, QSD_RES_TYPE(qsd),
 347                             qtype);
 348         qqi->qqi_glb_uptodate = false;
 349         qqi->qqi_slv_uptodate = false;
 350         qqi->qqi_reint        = false;
 351         cfs_waitq_init(&qqi->qqi_reint_thread.t_ctl_waitq);
 352         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 353         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 354         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 355         memset(&qqi->qqi_lockh, 0, sizeof(qqi->qqi_lockh));
 356
 357         /* open accounting object */
 358         LASSERT(qqi->qqi_acct_obj == NULL);
 359         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 360         if (qqi->qqi_acct_obj == NULL) {
 361                 LCONSOLE_ERROR("%s: No %s space accounting support. Please use "
 362                                "tunefs.lustre --quota option to enable quota "
 363                                "accounting.\n",
 364                                qsd->qsd_svname, QTYPE_NAME(qtype));
 365                 GOTO(out, rc = -ENOENT);
 366         }
 367
 368         /* open global index copy */
 369         LASSERT(qqi->qqi_glb_obj == NULL);
 370         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 371                                                        qsd->qsd_root,
 372                                                        &qqi->qqi_fid, true);
 373         if (IS_ERR(qqi->qqi_glb_obj)) {
 374                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 375                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 376                        PTR_ERR(qqi->qqi_glb_obj));
 377                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 378         }
 379         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 380
 381         /* open slave index copy */
 382         LASSERT(qqi->qqi_slv_obj == NULL);
 383         obd_str2uuid(&uuid, qsd->qsd_svname);
 384         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 385                                                        qsd->qsd_root,
 386                                                        &qqi->qqi_fid, &uuid,
 387                                                        true);
 388         if (IS_ERR(qqi->qqi_slv_obj)) {
 389                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 390                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 391                        PTR_ERR(qqi->qqi_slv_obj));
 392                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 393         }
 394         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 395
 396         /* allocate site */
 397         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 398         if (IS_ERR(qqi->qqi_site)) {
 399                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 400                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 401                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 402         }
 403
 404         /* register proc entry for accounting & global index copy objects */
 405         rc = lprocfs_seq_create(qsd->qsd_proc,
 406                                 qtype == USRQUOTA ? "acct_user" : "acct_group",
 407                                 0444, &lprocfs_quota_seq_fops,
 408                                 qqi->qqi_acct_obj);
 409         if (rc) {
 410                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 411                        qsd->qsd_svname, rc);
 412                 GOTO(out, rc);
 413         }
 414
 415         rc = lprocfs_seq_create(qsd->qsd_proc,
 416                                 qtype == USRQUOTA ? "limit_user" : "limit_group",
 417                                 0444, &lprocfs_quota_seq_fops,
 418                                 qqi->qqi_glb_obj);
 419         if (rc) {
 420                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 421                        qsd->qsd_svname, rc);
 422                 GOTO(out, rc);
 423         }
 424         EXIT;
 425 out:
 426         if (rc)
 427                 qsd_qtype_fini(env, qsd, qtype);
 428         return rc;
 429 }
 430
 431 /*
 432  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 433  * structures associated with the quota slave (on-disk objects, lquota entry
 434  * tables, ...).
 435  * This function should be called when the OSD is shutting down.
 436  *
 437  * \param env - is the environment passed by the caller
 438  * \param qsd - is the qsd instance to shutdown
 439  */
 440 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 441 {
 442         int     qtype;
 443         ENTRY;
 444
 445         if (unlikely(qsd == NULL))
 446                 RETURN_EXIT;
 447
 448         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 449         cfs_write_lock(&qsd->qsd_lock);
 450         qsd->qsd_stopping = true;
 451         cfs_write_unlock(&qsd->qsd_lock);
 452
 453         /* remove from the list of fsinfo */
 454         if (!cfs_list_empty(&qsd->qsd_link)) {
 455                 LASSERT(qsd->qsd_fsinfo != NULL);
 456                 cfs_down(&qsd->qsd_fsinfo->qfs_sem);
 457                 cfs_list_del_init(&qsd->qsd_link);
 458                 cfs_up(&qsd->qsd_fsinfo->qfs_sem);
 459         }
 460
 461         /* remove qsd proc entry */
 462         if (qsd->qsd_proc != NULL) {
 463                 lprocfs_remove(&qsd->qsd_proc);
 464                 qsd->qsd_proc = NULL;
 465         }
 466
 467         /* stop the writeback thread */
 468         qsd_stop_upd_thread(qsd);
 469
 470         /* shutdown the reintegration threads */
 471         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 472                 if (qsd->qsd_type_array[qtype] == NULL)
 473                         continue;
 474                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 475         }
 476
 477         /* release reference on namespace */
 478         if (qsd->qsd_ns != NULL) {
 479                 ldlm_namespace_put(qsd->qsd_ns);
 480                 qsd->qsd_ns = NULL;
 481         }
 482
 483         /* free per-quota type data */
 484         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
 485                 qsd_qtype_fini(env, qsd, qtype);
 486
 487         /* deregister connection to the quota master */
 488         qsd->qsd_exp_valid = false;
 489         lustre_deregister_osp_item(&qsd->qsd_exp);
 490
 491         /* release per-filesystem information */
 492         if (qsd->qsd_fsinfo != NULL)
 493                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 494
 495         /* release quota root directory */
 496         if (qsd->qsd_root != NULL) {
 497                 lu_object_put(env, &qsd->qsd_root->do_lu);
 498                 qsd->qsd_root = NULL;
 499         }
 500
 501         /* release reference on dt_device */
 502         if (qsd->qsd_dev != NULL) {
 503                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 504                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 505                 qsd->qsd_dev = NULL;
 506         }
 507
 508         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 509         OBD_FREE_PTR(qsd);
 510         EXIT;
 511 }
 512 EXPORT_SYMBOL(qsd_fini);
 513
 514 /*
 515  * Create a new qsd_instance to be associated with backend osd device
 516  * identified by \dev.
 517  *
 518  * \param env    - the environment passed by the caller
 519  * \param svname - is the service name of the OSD device creating this instance
 520  * \param dev    - is the dt_device where to store quota index files
 521  * \param osd_proc - is the procfs parent directory where to create procfs file
 522  *                   related to this new qsd instance
 523  *
 524  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 525  *           appropriate error on failure
 526  */
 527 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 528                               struct dt_device *dev,
 529                               cfs_proc_dir_entry_t *osd_proc)
 530 {
 531         struct qsd_thread_info  *qti = qsd_info(env);
 532         struct qsd_instance     *qsd;
 533         int                      rc, type, idx;
 534         ENTRY;
 535
 536         /* only configure qsd for MDT & OST */
 537         type = server_name2index(svname, &idx, NULL);
 538         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 539                 RETURN(NULL);
 540
 541         /* allocate qsd instance */
 542         OBD_ALLOC_PTR(qsd);
 543         if (qsd == NULL)
 544                 RETURN(ERR_PTR(-ENOMEM));
 545
 546         /* generic initializations */
 547         cfs_rwlock_init(&qsd->qsd_lock);
 548         CFS_INIT_LIST_HEAD(&qsd->qsd_link);
 549         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 550         cfs_waitq_init(&qsd->qsd_upd_thread.t_ctl_waitq);
 551         CFS_INIT_LIST_HEAD(&qsd->qsd_upd_list);
 552         cfs_spin_lock_init(&qsd->qsd_adjust_lock);
 553         CFS_INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 554         qsd->qsd_prepared = false;
 555         qsd->qsd_started = false;
 556
 557         /* copy service name */
 558         strncpy(qsd->qsd_svname, svname, MAX_OBD_NAME);
 559
 560         /* grab reference on osd device */
 561         lu_device_get(&dev->dd_lu_dev);
 562         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 563         qsd->qsd_dev = dev;
 564
 565         /* we only support pool ID 0 (default data or metadata pool) for the
 566          * time being. A different pool ID could be assigned to this target via
 567          * the configuration log in the future */
 568         qsd->qsd_pool_id  = 0;
 569
 570         /* get fsname from svname */
 571         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 572         if (rc) {
 573                 CERROR("%s: fail to extract filesystem name\n", svname);
 574                 GOTO(out, rc);
 575         }
 576
 577         /* look up quota setting for the filesystem the target belongs to */
 578         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 579         if (qsd->qsd_fsinfo == NULL) {
 580                 CERROR("%s: failed to locate filesystem information\n", svname);
 581                 GOTO(out, rc = -EINVAL);
 582         }
 583
 584         /* add in the list of lquota_fsinfo */
 585         cfs_down(&qsd->qsd_fsinfo->qfs_sem);
 586         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 587         cfs_up(&qsd->qsd_fsinfo->qfs_sem);
 588
 589         /* register procfs directory */
 590         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 591                                          lprocfs_quota_qsd_vars, qsd);
 592         if (IS_ERR(qsd->qsd_proc)) {
 593                 rc = PTR_ERR(qsd->qsd_proc);
 594                 qsd->qsd_proc = NULL;
 595                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 596                        svname, rc);
 597                 GOTO(out, rc);
 598         }
 599         EXIT;
 600 out:
 601         if (rc) {
 602                 qsd_fini(env, qsd);
 603                 return ERR_PTR(rc);
 604         }
 605         RETURN(qsd);
 606 }
 607 EXPORT_SYMBOL(qsd_init);
 608
 609 /*
 610  * Initialize on-disk structures in order to manage quota enforcement for
 611  * the target associated with the qsd instance \qsd and starts the reintegration
 612  * procedure for each quota type as soon as possible.
 613  * The last step of the reintegration will be completed once qsd_start() is
 614  * called, at which points the space reconciliation with the master will be
 615  * executed.
 616  * This function must be called when the server stack is fully configured,
 617  * typically when ->ldo_prepare is called across the stack.
 618  *
 619  * \param env - the environment passed by the caller
 620  * \param qsd - is qsd_instance to prepare
 621  *
 622  * \retval - 0 on success, appropriate error on failure
 623  */
 624 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 625 {
 626         struct qsd_thread_info  *qti = qsd_info(env);
 627         int                      qtype, rc = 0;
 628         ENTRY;
 629
 630         if (unlikely(qsd == NULL))
 631                 RETURN(0);
 632
 633         cfs_read_lock(&qsd->qsd_lock);
 634         if (qsd->qsd_prepared) {
 635                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 636                 rc = -EALREADY;
 637         }
 638         cfs_read_unlock(&qsd->qsd_lock);
 639         if (rc)
 640                 RETURN(rc);
 641
 642         /* Record whether this qsd instance is managing quota enforcement for a
 643          * MDT (i.e. inode quota) or OST (block quota) */
 644         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 645                 qsd->qsd_is_md = true;
 646                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 647         } else {
 648                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 649         }
 650
 651         /* look-up on-disk directory for the quota slave */
 652         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 653                                                     QSD_DIR);
 654         if (IS_ERR(qsd->qsd_root)) {
 655                 rc = PTR_ERR(qsd->qsd_root);
 656                 qsd->qsd_root = NULL;
 657                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 658                        qsd->qsd_svname, rc);
 659                 RETURN(rc);
 660         }
 661
 662         /* initialize per-quota type data */
 663         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 664                 rc = qsd_qtype_init(env, qsd, qtype);
 665                 if (rc)
 666                         RETURN(rc);
 667         }
 668
 669         /* pools successfully setup, mark the qsd as prepared */
 670         cfs_write_lock(&qsd->qsd_lock);
 671         qsd->qsd_prepared = true;
 672         cfs_write_unlock(&qsd->qsd_lock);
 673
 674         /* start reintegration thread for each type, if required */
 675         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 676                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 677
 678                 rc = qsd_start_reint_thread(qqi);
 679                 if (rc) {
 680                         CERROR("%s: failed to start reint thread for type %s "
 681                                "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
 682                                rc);
 683                         RETURN(rc);
 684                 }
 685         }
 686
 687         /* start writeback thread */
 688         rc = qsd_start_upd_thread(qsd);
 689         if (rc) {
 690                 CERROR("%s: failed to start writeback thread (%d)\n",
 691                        qsd->qsd_svname, rc);
 692                 RETURN(rc);
 693         }
 694
 695         /* generate osp name */
 696         rc = tgt_name2ospname((char *)qsd->qsd_svname, qti->qti_buf);
 697         if (rc) {
 698                 CERROR("%s: failed to generate ospname (%d)\n",
 699                        qsd->qsd_svname, rc);
 700                 RETURN(rc);
 701         }
 702
 703         /* the connection callback will start the reintegration
 704          * procedure if quota is enabled */
 705         rc = lustre_register_osp_item(qti->qti_buf, &qsd->qsd_exp,
 706                                       qsd_conn_callback, (void *)qsd);
 707         if (rc) {
 708                 CERROR("%s: fail to get connection to master (%d)\n",
 709                        qsd->qsd_svname, rc);
 710                 RETURN(rc);
 711         }
 712
 713         RETURN(0);
 714 }
 715 EXPORT_SYMBOL(qsd_prepare);
 716
 717 /*
 718  * Start a qsd instance. This will complete the last step of the reintegration
 719  * procedure as soon as possible (provided that the master is reachable).
 720  * This should be called when recovery has been completed and quota should now
 721  * be enforced on every operations.
 722  *
 723  * \param env - the environment passed by the caller
 724  * \param qsd - is the qsd instance associated with the osd device to start
 725  */
 726 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 727 {
 728         int     type, rc = 0;
 729         ENTRY;
 730
 731         if (unlikely(qsd == NULL))
 732                 RETURN(0);
 733
 734         cfs_write_lock(&qsd->qsd_lock);
 735         if (!qsd->qsd_prepared) {
 736                 CERROR("%s: can't start qsd instance since it was properly "
 737                        "initialized\n", qsd->qsd_svname);
 738                 rc = -EFAULT;
 739         } else if (qsd->qsd_started) {
 740                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 741                 rc = -EALREADY;
 742         } else {
 743                 /* notify that the qsd_instance is now started */
 744                 qsd->qsd_started = true;
 745         }
 746         cfs_write_unlock(&qsd->qsd_lock);
 747
 748         if (rc)
 749                 RETURN(rc);
 750
 751         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 752          * up to usage; If usage < granted, release down to usage.  */
 753         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 754                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 755                 cfs_waitq_signal(&qqi->qqi_reint_thread.t_ctl_waitq);
 756         }
 757
 758         RETURN(rc);
 759 }
 760 EXPORT_SYMBOL(qsd_start);
 761
 762 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 763
 764 /*
 765  * Global initialization performed at module load time
 766  */
 767 int qsd_glb_init(void)
 768 {
 769         int     rc;
 770
 771         rc = lu_kmem_init(qsd_caches);
 772         if (rc)
 773                 return rc;
 774
 775         qsd_key_init_generic(&qsd_thread_key, NULL);
 776         lu_context_key_register(&qsd_thread_key);
 777         lustre_register_quota_process_config(qsd_process_config);
 778
 779         return 0;
 780 }
 781
 782 /*
 783  * Companion of qsd_glb_init() called at module unload time
 784  */
 785 void qsd_glb_fini(void)
 786 {
 787         lustre_register_quota_process_config(NULL);
 788         lu_kmem_fini(qsd_caches);
 789         lu_context_key_degister(&qsd_thread_key);
 790 }