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, 2016, 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
  81         LASSERT(qsd != NULL);
  82
  83         memset(enabled, 0, sizeof(enabled));
  84         if (qsd_type_enabled(qsd, USRQUOTA))
  85                 strcat(enabled, "u");
  86         if (qsd_type_enabled(qsd, GRPQUOTA))
  87                 strcat(enabled, "g");
  88         if (qsd_type_enabled(qsd, PRJQUOTA))
  89                 strncat(enabled, "p", 1);
  90         if (strlen(enabled) == 0)
  91                 strcat(enabled, "none");
  92
  93         seq_printf(m, "target name:    %s\n"
  94                    "pool ID:        %d\n"
  95                    "type:           %s\n"
  96                    "quota enabled:  %s\n"
  97                    "conn to master: %s\n",
  98                    qsd->qsd_svname, qsd->qsd_pool_id,
  99                    qsd->qsd_is_md ? "md" : "dt", enabled,
 100                    qsd->qsd_exp_valid ? "setup" : "not setup yet");
 101
 102         if (qsd->qsd_prepared) {
 103                 memset(enabled, 0, sizeof(enabled));
 104                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 105                         strcat(enabled, "u");
 106                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 107                         strcat(enabled, "g");
 108                 if (qsd->qsd_type_array[PRJQUOTA]->qqi_acct_obj != NULL)
 109                         strncat(enabled, "p", 1);
 110                 if (strlen(enabled) == 0)
 111                         strcat(enabled, "none");
 112                 seq_printf(m, "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                            "project uptodate: glb[%d],slv[%d],reint[%d]\n",
 116                            enabled,
 117                            qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 118                            qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 119                            qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 120                            qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 121                            qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 122                            qsd->qsd_type_array[GRPQUOTA]->qqi_reint,
 123                            qsd->qsd_type_array[PRJQUOTA]->qqi_glb_uptodate,
 124                            qsd->qsd_type_array[PRJQUOTA]->qqi_slv_uptodate,
 125                            qsd->qsd_type_array[PRJQUOTA]->qqi_reint);
 126         }
 127         return 0;
 128 }
 129 LPROC_SEQ_FOPS_RO(qsd_state);
 130
 131 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
 132 {
 133         struct qsd_instance     *qsd = m->private;
 134         char                     enabled[5];
 135
 136         LASSERT(qsd != NULL);
 137
 138         memset(enabled, 0, sizeof(enabled));
 139         if (qsd_type_enabled(qsd, USRQUOTA))
 140                 strcat(enabled, "u");
 141         if (qsd_type_enabled(qsd, GRPQUOTA))
 142                 strcat(enabled, "g");
 143         if (qsd_type_enabled(qsd, PRJQUOTA))
 144                 strncat(enabled, "p", 1);
 145         if (strlen(enabled) == 0)
 146                 strcat(enabled, "none");
 147
 148         seq_printf(m, "%s\n", enabled);
 149         return 0;
 150 }
 151 LPROC_SEQ_FOPS_RO(qsd_enabled);
 152
 153 /* force reintegration procedure to be executed.
 154  * Used for test/debugging purpose */
 155 static ssize_t
 156 lprocfs_force_reint_seq_write(struct file *file, const char __user *buffer,
 157                                 size_t count, loff_t *off)
 158 {
 159         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 160         int                  rc = 0, qtype;
 161
 162         LASSERT(qsd != NULL);
 163
 164         write_lock(&qsd->qsd_lock);
 165         if (qsd->qsd_stopping) {
 166                 /* don't mess up with shutdown procedure, it is already
 167                  * complicated enough */
 168                 rc = -ESHUTDOWN;
 169         } else if (!qsd->qsd_prepared) {
 170                 rc = -EAGAIN;
 171         } else {
 172                 /* mark all indexes as stale */
 173                 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 174                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 175                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 176                 }
 177         }
 178         write_unlock(&qsd->qsd_lock);
 179
 180         if (rc)
 181                 return rc;
 182
 183         /* kick off reintegration */
 184         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 185                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 186                 if (rc)
 187                         break;
 188         }
 189         return rc == 0 ? count : rc;
 190 }
 191 LPROC_SEQ_FOPS_WO_TYPE(qsd, force_reint);
 192
 193 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 194 {
 195         struct qsd_instance *qsd = m->private;
 196         LASSERT(qsd != NULL);
 197
 198         seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 199         return 0;
 200 }
 201
 202 static ssize_t
 203 qsd_timeout_seq_write(struct file *file, const char __user *buffer,
 204                         size_t count, loff_t *off)
 205 {
 206         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 207         int rc;
 208         __s64 timeout;
 209         LASSERT(qsd != NULL);
 210
 211         rc = lprocfs_str_to_s64(buffer, count, &timeout);
 212         if (rc)
 213                 return rc;
 214         if (timeout < 0 || timeout > INT_MAX)
 215                 return -EINVAL;
 216
 217         qsd->qsd_timeout = timeout;
 218         return count;
 219 }
 220 LPROC_SEQ_FOPS(qsd_timeout);
 221
 222 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 223         { .name =       "info",
 224           .fops =       &qsd_state_fops         },
 225         { .name =       "enabled",
 226           .fops =       &qsd_enabled_fops       },
 227         { .name =       "force_reint",
 228           .fops =       &qsd_force_reint_fops   },
 229         { .name =       "timeout",
 230           .fops =       &qsd_timeout_fops       },
 231         { NULL }
 232 };
 233
 234 /*
 235  * Callback function invoked by the OSP layer when the connection to the master
 236  * has been set up.
 237  *
 238  * \param data - is a pointer to the qsd_instance
 239  *
 240  * \retval - 0 on success, appropriate error on failure
 241  */
 242 static int qsd_conn_callback(void *data)
 243 {
 244         struct qsd_instance *qsd = (struct qsd_instance *)data;
 245         int                  type;
 246         ENTRY;
 247
 248         /* qsd_exp should now be valid */
 249         LASSERT(qsd->qsd_exp);
 250
 251         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 252
 253         write_lock(&qsd->qsd_lock);
 254         /* notify that qsd_exp is now valid */
 255         qsd->qsd_exp_valid = true;
 256         write_unlock(&qsd->qsd_lock);
 257
 258         /* Now that the connection to master is setup, we can initiate the
 259          * reintegration procedure for quota types which are enabled.
 260          * It is worth noting that, if the qsd_instance hasn't been started
 261          * already, then we can only complete the first two steps of the
 262          * reintegration procedure (i.e. global lock enqueue and slave
 263          * index transfer) since the space usage reconciliation (i.e.
 264          * step 3) will have to wait for qsd_start() to be called */
 265         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 266                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 267                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 268         }
 269
 270         RETURN(0);
 271 }
 272
 273 /*
 274  * Release qsd_qtype_info structure which contains data associated with a
 275  * given quota type. This releases the accounting objects.
 276  * It's called on OSD cleanup when the qsd instance is released.
 277  *
 278  * \param env - is the environment passed by the caller
 279  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 280  *              to be released
 281  * \param qtype - is the quota type to be shutdown
 282  */
 283 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 284                            int qtype)
 285 {
 286         struct qsd_qtype_info   *qqi;
 287         int repeat = 0;
 288         ENTRY;
 289
 290         if (qsd->qsd_type_array[qtype] == NULL)
 291                 RETURN_EXIT;
 292         qqi = qsd->qsd_type_array[qtype];
 293         qsd->qsd_type_array[qtype] = NULL;
 294
 295         /* all deferred work lists should be empty */
 296         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 297         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 298
 299         /* shutdown lquota site */
 300         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 301                 lquota_site_free(env, qqi->qqi_site);
 302                 qqi->qqi_site = NULL;
 303         }
 304
 305         /* The qqi may still be holding by global locks which are being
 306          * canceled asynchronously (LU-4365), see the following steps:
 307          *
 308          * - On server umount, we try to clear all quota locks first by
 309          *   disconnecting LWP (which will invalidate import and cleanup
 310          *   all locks on it), however, if quota reint process is holding
 311          *   the global lock for reintegration at that time, global lock
 312          *   will fail to be cleared on LWP disconnection.
 313          *
 314          * - Umount process goes on and stops reint process, the global
 315          *   lock will be dropped on reint process exit, however, the lock
 316          *   cancel in done in asynchronous way, so the
 317          *   qsd_glb_blocking_ast() might haven't been called yet when we
 318          *   get here.
 319          */
 320         while (atomic_read(&qqi->qqi_ref) > 1) {
 321                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 322                        atomic_read(&qqi->qqi_ref), repeat);
 323                 repeat++;
 324                 set_current_state(TASK_INTERRUPTIBLE);
 325                 schedule_timeout(cfs_time_seconds(1));
 326         }
 327
 328         /* by now, all qqi users should have gone away */
 329         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 330         lu_ref_fini(&qqi->qqi_reference);
 331
 332         /* release accounting object */
 333         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 334                 dt_object_put(env, qqi->qqi_acct_obj);
 335                 qqi->qqi_acct_obj = NULL;
 336         }
 337
 338         /* release slv index */
 339         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 340                 dt_object_put(env, qqi->qqi_slv_obj);
 341                 qqi->qqi_slv_obj = NULL;
 342                 qqi->qqi_slv_ver = 0;
 343         }
 344
 345         /* release global index */
 346         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 347                 dt_object_put(env, qqi->qqi_glb_obj);
 348                 qqi->qqi_glb_obj = NULL;
 349                 qqi->qqi_glb_ver = 0;
 350         }
 351
 352         OBD_FREE_PTR(qqi);
 353         EXIT;
 354 }
 355
 356 static const char *qtype2acct_name(int qtype)
 357 {
 358         switch (qtype) {
 359         case USRQUOTA:
 360                 return "acct_user";
 361         case GRPQUOTA:
 362                 return "acct_group";
 363         case PRJQUOTA:
 364                 return "acct_project";
 365         }
 366
 367         LASSERTF(0, "invalid quota type: %d", qtype);
 368         return NULL;
 369 }
 370
 371 static const char *qtype2glb_name(int qtype)
 372 {
 373         switch (qtype) {
 374         case USRQUOTA:
 375                 return "limit_user";
 376         case GRPQUOTA:
 377                 return "limit_group";
 378         case PRJQUOTA:
 379                 return "limit_project";
 380         }
 381
 382         LASSERTF(0, "invalid quota type: %d", qtype);
 383         return NULL;
 384 }
 385
 386 /*
 387  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 388  * This opens the accounting object and initializes the proc file.
 389  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 390  * instance.
 391  *
 392  * \param env  - the environment passed by the caller
 393  * \param qsd  - is the qsd instance which will be in charge of the new
 394  *               qsd_qtype_info instance.
 395  * \param qtype - is quota type to set up
 396  *
 397  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 398  *           appropriate error on failure
 399  */
 400 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 401                           int qtype)
 402 {
 403         struct qsd_qtype_info   *qqi;
 404         int                      rc;
 405         struct obd_uuid          uuid;
 406         ENTRY;
 407
 408         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 409
 410         /* allocate structure for this quota type */
 411         OBD_ALLOC_PTR(qqi);
 412         if (qqi == NULL)
 413                 RETURN(-ENOMEM);
 414         qsd->qsd_type_array[qtype] = qqi;
 415         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 416
 417         /* set backpointer and other parameters */
 418         qqi->qqi_qsd   = qsd;
 419         qqi->qqi_qtype = qtype;
 420         lu_ref_init(&qqi->qqi_reference);
 421         qqi->qqi_glb_uptodate = false;
 422         qqi->qqi_slv_uptodate = false;
 423         qqi->qqi_reint        = false;
 424         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 425         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 426         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 427         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 428         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id,
 429                             QSD_RES_TYPE(qsd), qtype);
 430
 431         /* open accounting object */
 432         LASSERT(qqi->qqi_acct_obj == NULL);
 433         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 434         if (IS_ERR(qqi->qqi_acct_obj)) {
 435                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 436                        qsd->qsd_svname, qtype_name(qtype),
 437                        PTR_ERR(qqi->qqi_acct_obj));
 438                 qqi->qqi_acct_obj = NULL;
 439                 qqi->qqi_acct_failed = true;
 440         }
 441
 442         /* open global index copy */
 443         LASSERT(qqi->qqi_glb_obj == NULL);
 444         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 445                                                        qsd->qsd_root,
 446                                                        &qqi->qqi_fid, true);
 447         if (IS_ERR(qqi->qqi_glb_obj)) {
 448                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 449                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 450                        PTR_ERR(qqi->qqi_glb_obj));
 451                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 452         }
 453         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 454
 455         /* open slave index copy */
 456         LASSERT(qqi->qqi_slv_obj == NULL);
 457         obd_str2uuid(&uuid, qsd->qsd_svname);
 458         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 459                                                        qsd->qsd_root,
 460                                                        &qqi->qqi_fid, &uuid,
 461                                                        true);
 462         if (IS_ERR(qqi->qqi_slv_obj)) {
 463                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 464                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 465                        PTR_ERR(qqi->qqi_slv_obj));
 466                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 467         }
 468         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 469
 470         /* allocate site */
 471         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 472         if (IS_ERR(qqi->qqi_site)) {
 473                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 474                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 475                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 476         }
 477
 478         /* register proc entry for accounting & global index copy objects */
 479         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 480                                 0444, &lprocfs_quota_seq_fops,
 481                                 qqi->qqi_acct_obj);
 482         if (rc) {
 483                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 484                        qsd->qsd_svname, rc);
 485                 GOTO(out, rc);
 486         }
 487
 488         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 489                                 0444, &lprocfs_quota_seq_fops,
 490                                 qqi->qqi_glb_obj);
 491         if (rc) {
 492                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 493                        qsd->qsd_svname, rc);
 494                 GOTO(out, rc);
 495         }
 496         EXIT;
 497 out:
 498         if (rc)
 499                 qsd_qtype_fini(env, qsd, qtype);
 500         return rc;
 501 }
 502
 503 /*
 504  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 505  * structures associated with the quota slave (on-disk objects, lquota entry
 506  * tables, ...).
 507  * This function should be called when the OSD is shutting down.
 508  *
 509  * \param env - is the environment passed by the caller
 510  * \param qsd - is the qsd instance to shutdown
 511  */
 512 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 513 {
 514         int     qtype;
 515         ENTRY;
 516
 517         if (unlikely(qsd == NULL))
 518                 RETURN_EXIT;
 519
 520         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 521         write_lock(&qsd->qsd_lock);
 522         qsd->qsd_stopping = true;
 523         write_unlock(&qsd->qsd_lock);
 524
 525         /* remove qsd proc entry */
 526         if (qsd->qsd_proc != NULL) {
 527                 lprocfs_remove(&qsd->qsd_proc);
 528                 qsd->qsd_proc = NULL;
 529         }
 530
 531         /* stop the writeback thread */
 532         qsd_stop_upd_thread(qsd);
 533
 534         /* shutdown the reintegration threads */
 535         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 536                 if (qsd->qsd_type_array[qtype] == NULL)
 537                         continue;
 538                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 539         }
 540
 541         if (qsd->qsd_ns != NULL) {
 542                 qsd->qsd_ns = NULL;
 543         }
 544
 545         /* free per-quota type data */
 546         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 547                 qsd_qtype_fini(env, qsd, qtype);
 548
 549         if (qsd->qsd_exp) {
 550                 /* deregister connection to the quota master */
 551                 qsd->qsd_exp_valid = false;
 552                 lustre_deregister_lwp_item(&qsd->qsd_exp);
 553         }
 554
 555         /* release per-filesystem information */
 556         if (qsd->qsd_fsinfo != NULL) {
 557                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 558                 /* remove from the list of fsinfo */
 559                 list_del_init(&qsd->qsd_link);
 560                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 561                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 562                 qsd->qsd_fsinfo = NULL;
 563         }
 564
 565         /* release quota root directory */
 566         if (qsd->qsd_root != NULL) {
 567                 dt_object_put(env, qsd->qsd_root);
 568                 qsd->qsd_root = NULL;
 569         }
 570
 571         /* release reference on dt_device */
 572         if (qsd->qsd_dev != NULL) {
 573                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 574                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 575                 qsd->qsd_dev = NULL;
 576         }
 577
 578         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 579         OBD_FREE_PTR(qsd);
 580         EXIT;
 581 }
 582 EXPORT_SYMBOL(qsd_fini);
 583
 584 /*
 585  * Create a new qsd_instance to be associated with backend osd device
 586  * identified by \dev.
 587  *
 588  * \param env    - the environment passed by the caller
 589  * \param svname - is the service name of the OSD device creating this instance
 590  * \param dev    - is the dt_device where to store quota index files
 591  * \param osd_proc - is the procfs parent directory where to create procfs file
 592  *                   related to this new qsd instance
 593  *
 594  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 595  *           appropriate error on failure
 596  */
 597 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 598                               struct dt_device *dev,
 599                               struct proc_dir_entry *osd_proc)
 600 {
 601         struct qsd_thread_info  *qti = qsd_info(env);
 602         struct qsd_instance     *qsd;
 603         int                      rc, type, idx;
 604         ENTRY;
 605
 606         /* only configure qsd for MDT & OST */
 607         type = server_name2index(svname, &idx, NULL);
 608         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 609                 RETURN(NULL);
 610
 611         /* allocate qsd instance */
 612         OBD_ALLOC_PTR(qsd);
 613         if (qsd == NULL)
 614                 RETURN(ERR_PTR(-ENOMEM));
 615
 616         /* generic initializations */
 617         rwlock_init(&qsd->qsd_lock);
 618         INIT_LIST_HEAD(&qsd->qsd_link);
 619         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 620         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 621         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 622         spin_lock_init(&qsd->qsd_adjust_lock);
 623         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 624         qsd->qsd_prepared = false;
 625         qsd->qsd_started = false;
 626
 627         /* copy service name */
 628         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 629             >= sizeof(qsd->qsd_svname))
 630                 GOTO(out, rc = -E2BIG);
 631
 632         /* grab reference on osd device */
 633         lu_device_get(&dev->dd_lu_dev);
 634         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 635         qsd->qsd_dev = dev;
 636
 637         /* we only support pool ID 0 (default data or metadata pool) for the
 638          * time being. A different pool ID could be assigned to this target via
 639          * the configuration log in the future */
 640         qsd->qsd_pool_id  = 0;
 641
 642         /* get fsname from svname */
 643         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 644         if (rc) {
 645                 CERROR("%s: fail to extract filesystem name\n", svname);
 646                 GOTO(out, rc);
 647         }
 648
 649         /* look up quota setting for the filesystem the target belongs to */
 650         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 651         if (qsd->qsd_fsinfo == NULL) {
 652                 CERROR("%s: failed to locate filesystem information\n", svname);
 653                 GOTO(out, rc = -EINVAL);
 654         }
 655
 656         /* add in the list of lquota_fsinfo */
 657         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 658         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 659         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 660
 661         /* register procfs directory */
 662         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 663                                          lprocfs_quota_qsd_vars, qsd);
 664         if (IS_ERR(qsd->qsd_proc)) {
 665                 rc = PTR_ERR(qsd->qsd_proc);
 666                 qsd->qsd_proc = NULL;
 667                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 668                        svname, rc);
 669                 GOTO(out, rc);
 670         }
 671         EXIT;
 672 out:
 673         if (rc) {
 674                 qsd_fini(env, qsd);
 675                 return ERR_PTR(rc);
 676         }
 677         RETURN(qsd);
 678 }
 679 EXPORT_SYMBOL(qsd_init);
 680
 681 /*
 682  * Initialize on-disk structures in order to manage quota enforcement for
 683  * the target associated with the qsd instance \qsd and starts the reintegration
 684  * procedure for each quota type as soon as possible.
 685  * The last step of the reintegration will be completed once qsd_start() is
 686  * called, at which points the space reconciliation with the master will be
 687  * executed.
 688  * This function must be called when the server stack is fully configured,
 689  * typically when ->ldo_prepare is called across the stack.
 690  *
 691  * \param env - the environment passed by the caller
 692  * \param qsd - is qsd_instance to prepare
 693  *
 694  * \retval - 0 on success, appropriate error on failure
 695  */
 696 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 697 {
 698         struct qsd_thread_info  *qti = qsd_info(env);
 699         int                      qtype, rc = 0;
 700         ENTRY;
 701
 702         if (unlikely(qsd == NULL))
 703                 RETURN(0);
 704
 705         read_lock(&qsd->qsd_lock);
 706         if (qsd->qsd_prepared) {
 707                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 708                 rc = -EALREADY;
 709         }
 710         read_unlock(&qsd->qsd_lock);
 711         if (rc)
 712                 RETURN(rc);
 713
 714         /* Record whether this qsd instance is managing quota enforcement for a
 715          * MDT (i.e. inode quota) or OST (block quota) */
 716         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 717                 qsd->qsd_is_md = true;
 718                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 719         } else {
 720                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 721         }
 722
 723         /* look-up on-disk directory for the quota slave */
 724         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 725                                                     QSD_DIR);
 726         if (IS_ERR(qsd->qsd_root)) {
 727                 rc = PTR_ERR(qsd->qsd_root);
 728                 qsd->qsd_root = NULL;
 729                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 730                        qsd->qsd_svname, rc);
 731                 RETURN(rc);
 732         }
 733
 734         /* initialize per-quota type data */
 735         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 736                 rc = qsd_qtype_init(env, qsd, qtype);
 737                 if (rc)
 738                         RETURN(rc);
 739         }
 740
 741         /* pools successfully setup, mark the qsd as prepared */
 742         write_lock(&qsd->qsd_lock);
 743         qsd->qsd_prepared = true;
 744         write_unlock(&qsd->qsd_lock);
 745
 746         if (qsd->qsd_dev->dd_rdonly)
 747                 RETURN(0);
 748
 749         /* start reintegration thread for each type, if required */
 750         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 751                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 752
 753                 if (qsd_type_enabled(qsd, qtype) &&
 754                     qqi->qqi_acct_failed) {
 755                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 756                                        "since space accounting isn't functional"
 757                                        ". Please run tunefs.lustre --quota on "
 758                                        "an unmounted filesystem if not done "
 759                                        "already\n", qsd->qsd_svname);
 760                         break;
 761                 }
 762
 763                 rc = qsd_start_reint_thread(qqi);
 764                 if (rc) {
 765                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 766                                 qsd->qsd_svname, qtype_name(qtype), rc);
 767                         RETURN(rc);
 768                 }
 769         }
 770
 771         /* start writeback thread */
 772         rc = qsd_start_upd_thread(qsd);
 773         if (rc) {
 774                 CERROR("%s: failed to start writeback thread (%d)\n",
 775                        qsd->qsd_svname, rc);
 776                 RETURN(rc);
 777         }
 778
 779         /* generate osp name */
 780         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 781                                MTI_NAME_MAXLEN, 0);
 782         if (rc) {
 783                 CERROR("%s: failed to generate ospname (%d)\n",
 784                        qsd->qsd_svname, rc);
 785                 RETURN(rc);
 786         }
 787
 788         /* the connection callback will start the reintegration
 789          * procedure if quota is enabled */
 790         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 791                                       qsd_conn_callback, (void *)qsd);
 792         if (rc) {
 793                 CERROR("%s: fail to get connection to master (%d)\n",
 794                        qsd->qsd_svname, rc);
 795                 RETURN(rc);
 796         }
 797
 798         RETURN(0);
 799 }
 800 EXPORT_SYMBOL(qsd_prepare);
 801
 802 /*
 803  * Start a qsd instance. This will complete the last step of the reintegration
 804  * procedure as soon as possible (provided that the master is reachable).
 805  * This should be called when recovery has been completed and quota should now
 806  * be enforced on every operations.
 807  *
 808  * \param env - the environment passed by the caller
 809  * \param qsd - is the qsd instance associated with the osd device to start
 810  */
 811 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 812 {
 813         int     type, rc = 0;
 814         ENTRY;
 815
 816         if (unlikely(qsd == NULL))
 817                 RETURN(0);
 818
 819         write_lock(&qsd->qsd_lock);
 820         if (!qsd->qsd_prepared) {
 821                 CERROR("%s: can't start qsd instance since it wasn't properly "
 822                        "initialized\n", qsd->qsd_svname);
 823                 rc = -EFAULT;
 824         } else if (qsd->qsd_started) {
 825                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 826                 rc = -EALREADY;
 827         } else {
 828                 /* notify that the qsd_instance is now started */
 829                 qsd->qsd_started = true;
 830         }
 831         write_unlock(&qsd->qsd_lock);
 832
 833         if (rc)
 834                 RETURN(rc);
 835
 836         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 837          * up to usage; If usage < granted, release down to usage.  */
 838         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 839                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 840                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 841         }
 842
 843         RETURN(rc);
 844 }
 845 EXPORT_SYMBOL(qsd_start);
 846
 847 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 848
 849 /*
 850  * Global initialization performed at module load time
 851  */
 852 int qsd_glb_init(void)
 853 {
 854         int     rc;
 855
 856         rc = lu_kmem_init(qsd_caches);
 857         if (rc)
 858                 return rc;
 859
 860         qsd_key_init_generic(&qsd_thread_key, NULL);
 861         lu_context_key_register(&qsd_thread_key);
 862         lustre_register_quota_process_config(qsd_process_config);
 863
 864         return 0;
 865 }
 866
 867 /*
 868  * Companion of qsd_glb_init() called at module unload time
 869  */
 870 void qsd_glb_fini(void)
 871 {
 872         lustre_register_quota_process_config(NULL);
 873         lu_kmem_fini(qsd_caches);
 874         lu_context_key_degister(&qsd_thread_key);
 875 }