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, 2014, 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 qsd_state_seq_show(struct seq_file *m, void *data)
  77 {
  78         struct qsd_instance     *qsd = m->private;
  79         char                     enabled[5];
  80         int                      rc;
  81
  82         LASSERT(qsd != NULL);
  83
  84         memset(enabled, 0, sizeof(enabled));
  85         if (qsd_type_enabled(qsd, USRQUOTA))
  86                 strcat(enabled, "u");
  87         if (qsd_type_enabled(qsd, GRPQUOTA))
  88                 strcat(enabled, "g");
  89         if (strlen(enabled) == 0)
  90                 strcat(enabled, "none");
  91
  92         rc = seq_printf(m, "target name:    %s\n"
  93                         "pool ID:        %d\n"
  94                         "type:           %s\n"
  95                         "quota enabled:  %s\n"
  96                         "conn to master: %s\n",
  97                         qsd->qsd_svname, qsd->qsd_pool_id,
  98                         qsd->qsd_is_md ? "md" : "dt", enabled,
  99                         qsd->qsd_exp_valid ? "setup" : "not setup yet");
 100
 101         if (qsd->qsd_prepared) {
 102                 memset(enabled, 0, sizeof(enabled));
 103                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 104                         strcat(enabled, "u");
 105                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 106                         strcat(enabled, "g");
 107                 if (strlen(enabled) == 0)
 108                         strcat(enabled, "none");
 109                 rc += seq_printf(m, "space acct:     %s\n"
 110                                 "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 111                                 "group uptodate: glb[%d],slv[%d],reint[%d]\n",
 112                                 enabled,
 113                                 qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 114                                 qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 115                                 qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 116                                 qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 117                                 qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 118                                 qsd->qsd_type_array[GRPQUOTA]->qqi_reint);
 119         }
 120         return rc;
 121 }
 122 LPROC_SEQ_FOPS_RO(qsd_state);
 123
 124 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
 125 {
 126         struct qsd_instance     *qsd = m->private;
 127         char                     enabled[5];
 128
 129         LASSERT(qsd != NULL);
 130
 131         memset(enabled, 0, sizeof(enabled));
 132         if (qsd_type_enabled(qsd, USRQUOTA))
 133                 strcat(enabled, "u");
 134         if (qsd_type_enabled(qsd, GRPQUOTA))
 135                 strcat(enabled, "g");
 136         if (strlen(enabled) == 0)
 137                 strcat(enabled, "none");
 138
 139         return seq_printf(m, "%s\n", enabled);
 140 }
 141 LPROC_SEQ_FOPS_RO(qsd_enabled);
 142
 143 /* force reintegration procedure to be executed.
 144  * Used for test/debugging purpose */
 145 static ssize_t
 146 lprocfs_force_reint_seq_write(struct file *file, const char *buffer,
 147                                 size_t count, loff_t *off)
 148 {
 149         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 150         int                  rc = 0, qtype;
 151
 152         LASSERT(qsd != NULL);
 153
 154         write_lock(&qsd->qsd_lock);
 155         if (qsd->qsd_stopping) {
 156                 /* don't mess up with shutdown procedure, it is already
 157                  * complicated enough */
 158                 rc = -ESHUTDOWN;
 159         } else if (!qsd->qsd_prepared) {
 160                 rc = -EAGAIN;
 161         } else {
 162                 /* mark all indexes as stale */
 163                 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 164                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 165                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 166                 }
 167         }
 168         write_unlock(&qsd->qsd_lock);
 169
 170         if (rc)
 171                 return rc;
 172
 173         /* kick off reintegration */
 174         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 175                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 176                 if (rc)
 177                         break;
 178         }
 179         return rc == 0 ? count : rc;
 180 }
 181 LPROC_SEQ_FOPS_WO_TYPE(qsd, force_reint);
 182
 183 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 184 {
 185         struct qsd_instance *qsd = m->private;
 186         LASSERT(qsd != NULL);
 187
 188         return seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 189 }
 190
 191 static ssize_t
 192 qsd_timeout_seq_write(struct file *file, const char *buffer,
 193                         size_t count, loff_t *off)
 194 {
 195         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 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 LPROC_SEQ_FOPS(qsd_timeout);
 209
 210 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 211         { .name =       "info",
 212           .fops =       &qsd_state_fops         },
 213         { .name =       "enabled",
 214           .fops =       &qsd_enabled_fops       },
 215         { .name =       "force_reint",
 216           .fops =       &qsd_force_reint_fops   },
 217         { .name =       "timeout",
 218           .fops =       &qsd_timeout_fops       },
 219         { NULL }
 220 };
 221
 222 /*
 223  * Callback function invoked by the OSP layer when the connection to the master
 224  * has been set up.
 225  *
 226  * \param data - is a pointer to the qsd_instance
 227  *
 228  * \retval - 0 on success, appropriate error on failure
 229  */
 230 static int qsd_conn_callback(void *data)
 231 {
 232         struct qsd_instance *qsd = (struct qsd_instance *)data;
 233         int                  type;
 234         ENTRY;
 235
 236         /* qsd_exp should now be valid */
 237         LASSERT(qsd->qsd_exp);
 238
 239         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 240
 241         write_lock(&qsd->qsd_lock);
 242         /* notify that qsd_exp is now valid */
 243         qsd->qsd_exp_valid = true;
 244         write_unlock(&qsd->qsd_lock);
 245
 246         /* Now that the connection to master is setup, we can initiate the
 247          * reintegration procedure for quota types which are enabled.
 248          * It is worth noting that, if the qsd_instance hasn't been started
 249          * already, then we can only complete the first two steps of the
 250          * reintegration procedure (i.e. global lock enqueue and slave
 251          * index transfer) since the space usage reconciliation (i.e.
 252          * step 3) will have to wait for qsd_start() to be called */
 253         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 254                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 255                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 256         }
 257
 258         RETURN(0);
 259 }
 260
 261 /*
 262  * Release qsd_qtype_info structure which contains data associated with a
 263  * given quota type. This releases the accounting objects.
 264  * It's called on OSD cleanup when the qsd instance is released.
 265  *
 266  * \param env - is the environment passed by the caller
 267  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 268  *              to be released
 269  * \param qtype - is the quota type to be shutdown
 270  */
 271 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 272                            int qtype)
 273 {
 274         struct qsd_qtype_info   *qqi;
 275         int repeat = 0;
 276         ENTRY;
 277
 278         if (qsd->qsd_type_array[qtype] == NULL)
 279                 RETURN_EXIT;
 280         qqi = qsd->qsd_type_array[qtype];
 281         qsd->qsd_type_array[qtype] = NULL;
 282
 283         /* all deferred work lists should be empty */
 284         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 285         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 286
 287         /* shutdown lquota site */
 288         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 289                 lquota_site_free(env, qqi->qqi_site);
 290                 qqi->qqi_site = NULL;
 291         }
 292
 293         /* The qqi may still be holding by global locks which are being
 294          * canceled asynchronously (LU-4365), see the following steps:
 295          *
 296          * - On server umount, we try to clear all quota locks first by
 297          *   disconnecting LWP (which will invalidate import and cleanup
 298          *   all locks on it), however, if quota reint process is holding
 299          *   the global lock for reintegration at that time, global lock
 300          *   will fail to be cleared on LWP disconnection.
 301          *
 302          * - Umount process goes on and stops reint process, the global
 303          *   lock will be dropped on reint process exit, however, the lock
 304          *   cancel in done in asynchronous way, so the
 305          *   qsd_glb_blocking_ast() might haven't been called yet when we
 306          *   get here.
 307          */
 308         while (atomic_read(&qqi->qqi_ref) > 1) {
 309                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 310                        atomic_read(&qqi->qqi_ref), repeat);
 311                 repeat++;
 312                 set_current_state(TASK_INTERRUPTIBLE);
 313                 schedule_timeout(cfs_time_seconds(1));
 314         }
 315
 316         /* by now, all qqi users should have gone away */
 317         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 318         lu_ref_fini(&qqi->qqi_reference);
 319
 320         /* release accounting object */
 321         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 322                 lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
 323                 qqi->qqi_acct_obj = NULL;
 324         }
 325
 326         /* release slv index */
 327         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 328                 lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
 329                 qqi->qqi_slv_obj = NULL;
 330                 qqi->qqi_slv_ver = 0;
 331         }
 332
 333         /* release global index */
 334         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 335                 lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
 336                 qqi->qqi_glb_obj = NULL;
 337                 qqi->qqi_glb_ver = 0;
 338         }
 339
 340         OBD_FREE_PTR(qqi);
 341         EXIT;
 342 }
 343
 344 /*
 345  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 346  * This opens the accounting object and initializes the proc file.
 347  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 348  * instance.
 349  *
 350  * \param env  - the environment passed by the caller
 351  * \param qsd  - is the qsd instance which will be in charge of the new
 352  *               qsd_qtype_info instance.
 353  * \param qtype - is quota type to set up
 354  *
 355  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 356  *           appropriate error on failure
 357  */
 358 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 359                           int qtype)
 360 {
 361         struct qsd_qtype_info   *qqi;
 362         int                      rc;
 363         struct obd_uuid          uuid;
 364         ENTRY;
 365
 366         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 367
 368         /* allocate structure for this quota type */
 369         OBD_ALLOC_PTR(qqi);
 370         if (qqi == NULL)
 371                 RETURN(-ENOMEM);
 372         qsd->qsd_type_array[qtype] = qqi;
 373         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 374
 375         /* set backpointer and other parameters */
 376         qqi->qqi_qsd   = qsd;
 377         qqi->qqi_qtype = qtype;
 378         lu_ref_init(&qqi->qqi_reference);
 379         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id, QSD_RES_TYPE(qsd),
 380                             qtype);
 381         qqi->qqi_glb_uptodate = false;
 382         qqi->qqi_slv_uptodate = false;
 383         qqi->qqi_reint        = false;
 384         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 385         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 386         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 387         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 388
 389         /* open accounting object */
 390         LASSERT(qqi->qqi_acct_obj == NULL);
 391         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 392         if (IS_ERR(qqi->qqi_acct_obj)) {
 393                 CDEBUG(D_QUOTA, "%s: no %s space accounting support rc:%ld\n",
 394                        qsd->qsd_svname, QTYPE_NAME(qtype),
 395                        PTR_ERR(qqi->qqi_acct_obj));
 396                 qqi->qqi_acct_obj = NULL;
 397                 qsd->qsd_acct_failed = true;
 398         }
 399
 400         /* open global index copy */
 401         LASSERT(qqi->qqi_glb_obj == NULL);
 402         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 403                                                        qsd->qsd_root,
 404                                                        &qqi->qqi_fid, true);
 405         if (IS_ERR(qqi->qqi_glb_obj)) {
 406                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 407                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 408                        PTR_ERR(qqi->qqi_glb_obj));
 409                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 410         }
 411         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 412
 413         /* open slave index copy */
 414         LASSERT(qqi->qqi_slv_obj == NULL);
 415         obd_str2uuid(&uuid, qsd->qsd_svname);
 416         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 417                                                        qsd->qsd_root,
 418                                                        &qqi->qqi_fid, &uuid,
 419                                                        true);
 420         if (IS_ERR(qqi->qqi_slv_obj)) {
 421                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 422                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 423                        PTR_ERR(qqi->qqi_slv_obj));
 424                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 425         }
 426         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 427
 428         /* allocate site */
 429         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 430         if (IS_ERR(qqi->qqi_site)) {
 431                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 432                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 433                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 434         }
 435
 436         /* register proc entry for accounting & global index copy objects */
 437         rc = lprocfs_seq_create(qsd->qsd_proc,
 438                                 qtype == USRQUOTA ? "acct_user" : "acct_group",
 439                                 0444, &lprocfs_quota_seq_fops,
 440                                 qqi->qqi_acct_obj);
 441         if (rc) {
 442                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 443                        qsd->qsd_svname, rc);
 444                 GOTO(out, rc);
 445         }
 446
 447         rc = lprocfs_seq_create(qsd->qsd_proc,
 448                                 qtype == USRQUOTA ? "limit_user" : "limit_group",
 449                                 0444, &lprocfs_quota_seq_fops,
 450                                 qqi->qqi_glb_obj);
 451         if (rc) {
 452                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 453                        qsd->qsd_svname, rc);
 454                 GOTO(out, rc);
 455         }
 456         EXIT;
 457 out:
 458         if (rc)
 459                 qsd_qtype_fini(env, qsd, qtype);
 460         return rc;
 461 }
 462
 463 /*
 464  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 465  * structures associated with the quota slave (on-disk objects, lquota entry
 466  * tables, ...).
 467  * This function should be called when the OSD is shutting down.
 468  *
 469  * \param env - is the environment passed by the caller
 470  * \param qsd - is the qsd instance to shutdown
 471  */
 472 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 473 {
 474         int     qtype;
 475         ENTRY;
 476
 477         if (unlikely(qsd == NULL))
 478                 RETURN_EXIT;
 479
 480         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 481         write_lock(&qsd->qsd_lock);
 482         qsd->qsd_stopping = true;
 483         write_unlock(&qsd->qsd_lock);
 484
 485         /* remove qsd proc entry */
 486         if (qsd->qsd_proc != NULL) {
 487                 lprocfs_remove(&qsd->qsd_proc);
 488                 qsd->qsd_proc = NULL;
 489         }
 490
 491         /* stop the writeback thread */
 492         qsd_stop_upd_thread(qsd);
 493
 494         /* shutdown the reintegration threads */
 495         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 496                 if (qsd->qsd_type_array[qtype] == NULL)
 497                         continue;
 498                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 499         }
 500
 501         if (qsd->qsd_ns != NULL) {
 502                 qsd->qsd_ns = NULL;
 503         }
 504
 505         /* free per-quota type data */
 506         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
 507                 qsd_qtype_fini(env, qsd, qtype);
 508
 509         /* deregister connection to the quota master */
 510         qsd->qsd_exp_valid = false;
 511         lustre_deregister_lwp_item(&qsd->qsd_exp);
 512
 513         /* release per-filesystem information */
 514         if (qsd->qsd_fsinfo != NULL) {
 515                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 516                 /* remove from the list of fsinfo */
 517                 list_del_init(&qsd->qsd_link);
 518                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 519                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 520                 qsd->qsd_fsinfo = NULL;
 521         }
 522
 523         /* release quota root directory */
 524         if (qsd->qsd_root != NULL) {
 525                 lu_object_put(env, &qsd->qsd_root->do_lu);
 526                 qsd->qsd_root = NULL;
 527         }
 528
 529         /* release reference on dt_device */
 530         if (qsd->qsd_dev != NULL) {
 531                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 532                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 533                 qsd->qsd_dev = NULL;
 534         }
 535
 536         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 537         OBD_FREE_PTR(qsd);
 538         EXIT;
 539 }
 540 EXPORT_SYMBOL(qsd_fini);
 541
 542 /*
 543  * Create a new qsd_instance to be associated with backend osd device
 544  * identified by \dev.
 545  *
 546  * \param env    - the environment passed by the caller
 547  * \param svname - is the service name of the OSD device creating this instance
 548  * \param dev    - is the dt_device where to store quota index files
 549  * \param osd_proc - is the procfs parent directory where to create procfs file
 550  *                   related to this new qsd instance
 551  *
 552  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 553  *           appropriate error on failure
 554  */
 555 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 556                               struct dt_device *dev,
 557                               struct proc_dir_entry *osd_proc)
 558 {
 559         struct qsd_thread_info  *qti = qsd_info(env);
 560         struct qsd_instance     *qsd;
 561         int                      rc, type, idx;
 562         ENTRY;
 563
 564         /* only configure qsd for MDT & OST */
 565         type = server_name2index(svname, &idx, NULL);
 566         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 567                 RETURN(NULL);
 568
 569         /* allocate qsd instance */
 570         OBD_ALLOC_PTR(qsd);
 571         if (qsd == NULL)
 572                 RETURN(ERR_PTR(-ENOMEM));
 573
 574         /* generic initializations */
 575         rwlock_init(&qsd->qsd_lock);
 576         INIT_LIST_HEAD(&qsd->qsd_link);
 577         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 578         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 579         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 580         spin_lock_init(&qsd->qsd_adjust_lock);
 581         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 582         qsd->qsd_prepared = false;
 583         qsd->qsd_started = false;
 584
 585         /* copy service name */
 586         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 587             >= sizeof(qsd->qsd_svname))
 588                 GOTO(out, rc = -E2BIG);
 589
 590         /* grab reference on osd device */
 591         lu_device_get(&dev->dd_lu_dev);
 592         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 593         qsd->qsd_dev = dev;
 594
 595         /* we only support pool ID 0 (default data or metadata pool) for the
 596          * time being. A different pool ID could be assigned to this target via
 597          * the configuration log in the future */
 598         qsd->qsd_pool_id  = 0;
 599
 600         /* get fsname from svname */
 601         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 602         if (rc) {
 603                 CERROR("%s: fail to extract filesystem name\n", svname);
 604                 GOTO(out, rc);
 605         }
 606
 607         /* look up quota setting for the filesystem the target belongs to */
 608         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 609         if (qsd->qsd_fsinfo == NULL) {
 610                 CERROR("%s: failed to locate filesystem information\n", svname);
 611                 GOTO(out, rc = -EINVAL);
 612         }
 613
 614         /* add in the list of lquota_fsinfo */
 615         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 616         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 617         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 618
 619         /* register procfs directory */
 620         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 621                                          lprocfs_quota_qsd_vars, qsd);
 622         if (IS_ERR(qsd->qsd_proc)) {
 623                 rc = PTR_ERR(qsd->qsd_proc);
 624                 qsd->qsd_proc = NULL;
 625                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 626                        svname, rc);
 627                 GOTO(out, rc);
 628         }
 629         EXIT;
 630 out:
 631         if (rc) {
 632                 qsd_fini(env, qsd);
 633                 return ERR_PTR(rc);
 634         }
 635         RETURN(qsd);
 636 }
 637 EXPORT_SYMBOL(qsd_init);
 638
 639 /*
 640  * Initialize on-disk structures in order to manage quota enforcement for
 641  * the target associated with the qsd instance \qsd and starts the reintegration
 642  * procedure for each quota type as soon as possible.
 643  * The last step of the reintegration will be completed once qsd_start() is
 644  * called, at which points the space reconciliation with the master will be
 645  * executed.
 646  * This function must be called when the server stack is fully configured,
 647  * typically when ->ldo_prepare is called across the stack.
 648  *
 649  * \param env - the environment passed by the caller
 650  * \param qsd - is qsd_instance to prepare
 651  *
 652  * \retval - 0 on success, appropriate error on failure
 653  */
 654 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 655 {
 656         struct qsd_thread_info  *qti = qsd_info(env);
 657         int                      qtype, rc = 0;
 658         ENTRY;
 659
 660         if (unlikely(qsd == NULL))
 661                 RETURN(0);
 662
 663         read_lock(&qsd->qsd_lock);
 664         if (qsd->qsd_prepared) {
 665                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 666                 rc = -EALREADY;
 667         }
 668         read_unlock(&qsd->qsd_lock);
 669         if (rc)
 670                 RETURN(rc);
 671
 672         /* Record whether this qsd instance is managing quota enforcement for a
 673          * MDT (i.e. inode quota) or OST (block quota) */
 674         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 675                 qsd->qsd_is_md = true;
 676                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 677         } else {
 678                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 679         }
 680
 681         /* look-up on-disk directory for the quota slave */
 682         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 683                                                     QSD_DIR);
 684         if (IS_ERR(qsd->qsd_root)) {
 685                 rc = PTR_ERR(qsd->qsd_root);
 686                 qsd->qsd_root = NULL;
 687                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 688                        qsd->qsd_svname, rc);
 689                 RETURN(rc);
 690         }
 691
 692         /* initialize per-quota type data */
 693         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 694                 rc = qsd_qtype_init(env, qsd, qtype);
 695                 if (rc)
 696                         RETURN(rc);
 697         }
 698
 699         /* pools successfully setup, mark the qsd as prepared */
 700         write_lock(&qsd->qsd_lock);
 701         qsd->qsd_prepared = true;
 702         write_unlock(&qsd->qsd_lock);
 703
 704         /* start reintegration thread for each type, if required */
 705         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 706                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 707
 708                 if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
 709                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 710                                        "since space accounting isn't functional"
 711                                        ". Please run tunefs.lustre --quota on "
 712                                        "an unmounted filesystem if not done "
 713                                        "already\n", qsd->qsd_svname);
 714                         break;
 715                 }
 716
 717                 rc = qsd_start_reint_thread(qqi);
 718                 if (rc) {
 719                         CERROR("%s: failed to start reint thread for type %s "
 720                                "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
 721                                rc);
 722                         RETURN(rc);
 723                 }
 724         }
 725
 726         /* start writeback thread */
 727         rc = qsd_start_upd_thread(qsd);
 728         if (rc) {
 729                 CERROR("%s: failed to start writeback thread (%d)\n",
 730                        qsd->qsd_svname, rc);
 731                 RETURN(rc);
 732         }
 733
 734         /* generate osp name */
 735         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 736                                MTI_NAME_MAXLEN, 0);
 737         if (rc) {
 738                 CERROR("%s: failed to generate ospname (%d)\n",
 739                        qsd->qsd_svname, rc);
 740                 RETURN(rc);
 741         }
 742
 743         /* the connection callback will start the reintegration
 744          * procedure if quota is enabled */
 745         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 746                                       qsd_conn_callback, (void *)qsd);
 747         if (rc) {
 748                 CERROR("%s: fail to get connection to master (%d)\n",
 749                        qsd->qsd_svname, rc);
 750                 RETURN(rc);
 751         }
 752
 753         RETURN(0);
 754 }
 755 EXPORT_SYMBOL(qsd_prepare);
 756
 757 /*
 758  * Start a qsd instance. This will complete the last step of the reintegration
 759  * procedure as soon as possible (provided that the master is reachable).
 760  * This should be called when recovery has been completed and quota should now
 761  * be enforced on every operations.
 762  *
 763  * \param env - the environment passed by the caller
 764  * \param qsd - is the qsd instance associated with the osd device to start
 765  */
 766 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 767 {
 768         int     type, rc = 0;
 769         ENTRY;
 770
 771         if (unlikely(qsd == NULL))
 772                 RETURN(0);
 773
 774         write_lock(&qsd->qsd_lock);
 775         if (!qsd->qsd_prepared) {
 776                 CERROR("%s: can't start qsd instance since it wasn't properly "
 777                        "initialized\n", qsd->qsd_svname);
 778                 rc = -EFAULT;
 779         } else if (qsd->qsd_started) {
 780                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 781                 rc = -EALREADY;
 782         } else {
 783                 /* notify that the qsd_instance is now started */
 784                 qsd->qsd_started = true;
 785         }
 786         write_unlock(&qsd->qsd_lock);
 787
 788         if (rc)
 789                 RETURN(rc);
 790
 791         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 792          * up to usage; If usage < granted, release down to usage.  */
 793         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 794                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 795                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 796         }
 797
 798         RETURN(rc);
 799 }
 800 EXPORT_SYMBOL(qsd_start);
 801
 802 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 803
 804 /*
 805  * Global initialization performed at module load time
 806  */
 807 int qsd_glb_init(void)
 808 {
 809         int     rc;
 810
 811         rc = lu_kmem_init(qsd_caches);
 812         if (rc)
 813                 return rc;
 814
 815         qsd_key_init_generic(&qsd_thread_key, NULL);
 816         lu_context_key_register(&qsd_thread_key);
 817         lustre_register_quota_process_config(qsd_process_config);
 818
 819         return 0;
 820 }
 821
 822 /*
 823  * Companion of qsd_glb_init() called at module unload time
 824  */
 825 void qsd_glb_fini(void)
 826 {
 827         lustre_register_quota_process_config(NULL);
 828         lu_kmem_fini(qsd_caches);
 829         lu_context_key_degister(&qsd_thread_key);
 830 }