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         static char unknown[24];
 359
 360         switch (qtype) {
 361         case USRQUOTA:
 362                 return "acct_user";
 363         case GRPQUOTA:
 364                 return "acct_group";
 365         case PRJQUOTA:
 366                 return "acct_project";
 367         }
 368
 369         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 370         return unknown;
 371 }
 372
 373 static const char *qtype2glb_name(int qtype)
 374 {
 375         static char unknown[24];
 376
 377         switch (qtype) {
 378         case USRQUOTA:
 379                 return "limit_user";
 380         case GRPQUOTA:
 381                 return "limit_group";
 382         case PRJQUOTA:
 383                 return "limit_project";
 384         }
 385
 386         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 387         return unknown;
 388 }
 389
 390 /*
 391  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 392  * This opens the accounting object and initializes the proc file.
 393  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 394  * instance.
 395  *
 396  * \param env  - the environment passed by the caller
 397  * \param qsd  - is the qsd instance which will be in charge of the new
 398  *               qsd_qtype_info instance.
 399  * \param qtype - is quota type to set up
 400  *
 401  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 402  *           appropriate error on failure
 403  */
 404 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 405                           int qtype)
 406 {
 407         struct qsd_qtype_info   *qqi;
 408         int                      rc;
 409         struct obd_uuid          uuid;
 410         ENTRY;
 411
 412         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 413
 414         /* allocate structure for this quota type */
 415         OBD_ALLOC_PTR(qqi);
 416         if (qqi == NULL)
 417                 RETURN(-ENOMEM);
 418         qsd->qsd_type_array[qtype] = qqi;
 419         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 420
 421         /* set backpointer and other parameters */
 422         qqi->qqi_qsd   = qsd;
 423         qqi->qqi_qtype = qtype;
 424         lu_ref_init(&qqi->qqi_reference);
 425         qqi->qqi_glb_uptodate = false;
 426         qqi->qqi_slv_uptodate = false;
 427         qqi->qqi_reint        = false;
 428         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 429         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 430         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 431         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 432         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id,
 433                             QSD_RES_TYPE(qsd), qtype);
 434
 435         /* open accounting object */
 436         LASSERT(qqi->qqi_acct_obj == NULL);
 437         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 438         if (IS_ERR(qqi->qqi_acct_obj)) {
 439                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 440                        qsd->qsd_svname, qtype_name(qtype),
 441                        PTR_ERR(qqi->qqi_acct_obj));
 442                 qqi->qqi_acct_obj = NULL;
 443                 qqi->qqi_acct_failed = true;
 444         }
 445
 446         /* open global index copy */
 447         LASSERT(qqi->qqi_glb_obj == NULL);
 448         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 449                                                        qsd->qsd_root,
 450                                                        &qqi->qqi_fid, true);
 451         if (IS_ERR(qqi->qqi_glb_obj)) {
 452                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 453                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 454                        PTR_ERR(qqi->qqi_glb_obj));
 455                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 456         }
 457         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 458
 459         /* open slave index copy */
 460         LASSERT(qqi->qqi_slv_obj == NULL);
 461         obd_str2uuid(&uuid, qsd->qsd_svname);
 462         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 463                                                        qsd->qsd_root,
 464                                                        &qqi->qqi_fid, &uuid,
 465                                                        true);
 466         if (IS_ERR(qqi->qqi_slv_obj)) {
 467                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 468                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 469                        PTR_ERR(qqi->qqi_slv_obj));
 470                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 471         }
 472         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 473
 474         /* allocate site */
 475         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 476         if (IS_ERR(qqi->qqi_site)) {
 477                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 478                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 479                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 480         }
 481
 482         /* register proc entry for accounting & global index copy objects */
 483         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 484                                 0444, &lprocfs_quota_seq_fops,
 485                                 qqi->qqi_acct_obj);
 486         if (rc) {
 487                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 488                        qsd->qsd_svname, rc);
 489                 GOTO(out, rc);
 490         }
 491
 492         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 493                                 0444, &lprocfs_quota_seq_fops,
 494                                 qqi->qqi_glb_obj);
 495         if (rc) {
 496                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 497                        qsd->qsd_svname, rc);
 498                 GOTO(out, rc);
 499         }
 500         EXIT;
 501 out:
 502         if (rc)
 503                 qsd_qtype_fini(env, qsd, qtype);
 504         return rc;
 505 }
 506
 507 /*
 508  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 509  * structures associated with the quota slave (on-disk objects, lquota entry
 510  * tables, ...).
 511  * This function should be called when the OSD is shutting down.
 512  *
 513  * \param env - is the environment passed by the caller
 514  * \param qsd - is the qsd instance to shutdown
 515  */
 516 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 517 {
 518         int     qtype;
 519         ENTRY;
 520
 521         if (unlikely(qsd == NULL))
 522                 RETURN_EXIT;
 523
 524         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 525         write_lock(&qsd->qsd_lock);
 526         qsd->qsd_stopping = true;
 527         write_unlock(&qsd->qsd_lock);
 528
 529         /* remove qsd proc entry */
 530         if (qsd->qsd_proc != NULL) {
 531                 lprocfs_remove(&qsd->qsd_proc);
 532                 qsd->qsd_proc = NULL;
 533         }
 534
 535         /* stop the writeback thread */
 536         qsd_stop_upd_thread(qsd);
 537
 538         /* shutdown the reintegration threads */
 539         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 540                 if (qsd->qsd_type_array[qtype] == NULL)
 541                         continue;
 542                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 543         }
 544
 545         if (qsd->qsd_ns != NULL) {
 546                 qsd->qsd_ns = NULL;
 547         }
 548
 549         /* free per-quota type data */
 550         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 551                 qsd_qtype_fini(env, qsd, qtype);
 552
 553         if (qsd->qsd_exp) {
 554                 /* deregister connection to the quota master */
 555                 qsd->qsd_exp_valid = false;
 556                 lustre_deregister_lwp_item(&qsd->qsd_exp);
 557         }
 558
 559         /* release per-filesystem information */
 560         if (qsd->qsd_fsinfo != NULL) {
 561                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 562                 /* remove from the list of fsinfo */
 563                 list_del_init(&qsd->qsd_link);
 564                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 565                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 566                 qsd->qsd_fsinfo = NULL;
 567         }
 568
 569         /* release quota root directory */
 570         if (qsd->qsd_root != NULL) {
 571                 dt_object_put(env, qsd->qsd_root);
 572                 qsd->qsd_root = NULL;
 573         }
 574
 575         /* release reference on dt_device */
 576         if (qsd->qsd_dev != NULL) {
 577                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 578                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 579                 qsd->qsd_dev = NULL;
 580         }
 581
 582         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 583         OBD_FREE_PTR(qsd);
 584         EXIT;
 585 }
 586 EXPORT_SYMBOL(qsd_fini);
 587
 588 /*
 589  * Create a new qsd_instance to be associated with backend osd device
 590  * identified by \dev.
 591  *
 592  * \param env    - the environment passed by the caller
 593  * \param svname - is the service name of the OSD device creating this instance
 594  * \param dev    - is the dt_device where to store quota index files
 595  * \param osd_proc - is the procfs parent directory where to create procfs file
 596  *                   related to this new qsd instance
 597  *
 598  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 599  *           appropriate error on failure
 600  */
 601 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 602                               struct dt_device *dev,
 603                               struct proc_dir_entry *osd_proc)
 604 {
 605         struct qsd_thread_info  *qti = qsd_info(env);
 606         struct qsd_instance     *qsd;
 607         int                      rc, type, idx;
 608         ENTRY;
 609
 610         /* only configure qsd for MDT & OST */
 611         type = server_name2index(svname, &idx, NULL);
 612         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 613                 RETURN(NULL);
 614
 615         /* allocate qsd instance */
 616         OBD_ALLOC_PTR(qsd);
 617         if (qsd == NULL)
 618                 RETURN(ERR_PTR(-ENOMEM));
 619
 620         /* generic initializations */
 621         rwlock_init(&qsd->qsd_lock);
 622         INIT_LIST_HEAD(&qsd->qsd_link);
 623         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 624         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 625         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 626         spin_lock_init(&qsd->qsd_adjust_lock);
 627         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 628         qsd->qsd_prepared = false;
 629         qsd->qsd_started = false;
 630
 631         /* copy service name */
 632         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 633             >= sizeof(qsd->qsd_svname))
 634                 GOTO(out, rc = -E2BIG);
 635
 636         /* grab reference on osd device */
 637         lu_device_get(&dev->dd_lu_dev);
 638         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 639         qsd->qsd_dev = dev;
 640
 641         /* we only support pool ID 0 (default data or metadata pool) for the
 642          * time being. A different pool ID could be assigned to this target via
 643          * the configuration log in the future */
 644         qsd->qsd_pool_id  = 0;
 645
 646         /* get fsname from svname */
 647         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 648         if (rc) {
 649                 CERROR("%s: fail to extract filesystem name\n", svname);
 650                 GOTO(out, rc);
 651         }
 652
 653         /* look up quota setting for the filesystem the target belongs to */
 654         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 655         if (qsd->qsd_fsinfo == NULL) {
 656                 CERROR("%s: failed to locate filesystem information\n", svname);
 657                 GOTO(out, rc = -EINVAL);
 658         }
 659
 660         /* add in the list of lquota_fsinfo */
 661         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 662         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 663         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 664
 665         /* register procfs directory */
 666         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 667                                          lprocfs_quota_qsd_vars, qsd);
 668         if (IS_ERR(qsd->qsd_proc)) {
 669                 rc = PTR_ERR(qsd->qsd_proc);
 670                 qsd->qsd_proc = NULL;
 671                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 672                        svname, rc);
 673                 GOTO(out, rc);
 674         }
 675         EXIT;
 676 out:
 677         if (rc) {
 678                 qsd_fini(env, qsd);
 679                 return ERR_PTR(rc);
 680         }
 681         RETURN(qsd);
 682 }
 683 EXPORT_SYMBOL(qsd_init);
 684
 685 /*
 686  * Initialize on-disk structures in order to manage quota enforcement for
 687  * the target associated with the qsd instance \qsd and starts the reintegration
 688  * procedure for each quota type as soon as possible.
 689  * The last step of the reintegration will be completed once qsd_start() is
 690  * called, at which points the space reconciliation with the master will be
 691  * executed.
 692  * This function must be called when the server stack is fully configured,
 693  * typically when ->ldo_prepare is called across the stack.
 694  *
 695  * \param env - the environment passed by the caller
 696  * \param qsd - is qsd_instance to prepare
 697  *
 698  * \retval - 0 on success, appropriate error on failure
 699  */
 700 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 701 {
 702         struct qsd_thread_info  *qti = qsd_info(env);
 703         int                      qtype, rc = 0;
 704         ENTRY;
 705
 706         if (unlikely(qsd == NULL))
 707                 RETURN(0);
 708
 709         read_lock(&qsd->qsd_lock);
 710         if (qsd->qsd_prepared) {
 711                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 712                 rc = -EALREADY;
 713         }
 714         read_unlock(&qsd->qsd_lock);
 715         if (rc)
 716                 RETURN(rc);
 717
 718         /* Record whether this qsd instance is managing quota enforcement for a
 719          * MDT (i.e. inode quota) or OST (block quota) */
 720         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 721                 qsd->qsd_is_md = true;
 722                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 723         } else {
 724                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 725         }
 726
 727         /* look-up on-disk directory for the quota slave */
 728         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 729                                                     QSD_DIR);
 730         if (IS_ERR(qsd->qsd_root)) {
 731                 rc = PTR_ERR(qsd->qsd_root);
 732                 qsd->qsd_root = NULL;
 733                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 734                        qsd->qsd_svname, rc);
 735                 RETURN(rc);
 736         }
 737
 738         /* initialize per-quota type data */
 739         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 740                 rc = qsd_qtype_init(env, qsd, qtype);
 741                 if (rc)
 742                         RETURN(rc);
 743         }
 744
 745         /* pools successfully setup, mark the qsd as prepared */
 746         write_lock(&qsd->qsd_lock);
 747         qsd->qsd_prepared = true;
 748         write_unlock(&qsd->qsd_lock);
 749
 750         if (qsd->qsd_dev->dd_rdonly)
 751                 RETURN(0);
 752
 753         /* start reintegration thread for each type, if required */
 754         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 755                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 756
 757                 if (qsd_type_enabled(qsd, qtype) &&
 758                     qqi->qqi_acct_failed) {
 759                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 760                                        "since space accounting isn't functional"
 761                                        ". Please run tunefs.lustre --quota on "
 762                                        "an unmounted filesystem if not done "
 763                                        "already\n", qsd->qsd_svname);
 764                         break;
 765                 }
 766
 767                 rc = qsd_start_reint_thread(qqi);
 768                 if (rc) {
 769                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 770                                 qsd->qsd_svname, qtype_name(qtype), rc);
 771                         RETURN(rc);
 772                 }
 773         }
 774
 775         /* start writeback thread */
 776         rc = qsd_start_upd_thread(qsd);
 777         if (rc) {
 778                 CERROR("%s: failed to start writeback thread (%d)\n",
 779                        qsd->qsd_svname, rc);
 780                 RETURN(rc);
 781         }
 782
 783         /* generate osp name */
 784         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 785                                MTI_NAME_MAXLEN, 0);
 786         if (rc) {
 787                 CERROR("%s: failed to generate ospname (%d)\n",
 788                        qsd->qsd_svname, rc);
 789                 RETURN(rc);
 790         }
 791
 792         /* the connection callback will start the reintegration
 793          * procedure if quota is enabled */
 794         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 795                                       qsd_conn_callback, (void *)qsd);
 796         if (rc) {
 797                 CERROR("%s: fail to get connection to master (%d)\n",
 798                        qsd->qsd_svname, rc);
 799                 RETURN(rc);
 800         }
 801
 802         RETURN(0);
 803 }
 804 EXPORT_SYMBOL(qsd_prepare);
 805
 806 /*
 807  * Start a qsd instance. This will complete the last step of the reintegration
 808  * procedure as soon as possible (provided that the master is reachable).
 809  * This should be called when recovery has been completed and quota should now
 810  * be enforced on every operations.
 811  *
 812  * \param env - the environment passed by the caller
 813  * \param qsd - is the qsd instance associated with the osd device to start
 814  */
 815 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 816 {
 817         int     type, rc = 0;
 818         ENTRY;
 819
 820         if (unlikely(qsd == NULL))
 821                 RETURN(0);
 822
 823         write_lock(&qsd->qsd_lock);
 824         if (!qsd->qsd_prepared) {
 825                 CERROR("%s: can't start qsd instance since it wasn't properly "
 826                        "initialized\n", qsd->qsd_svname);
 827                 rc = -EFAULT;
 828         } else if (qsd->qsd_started) {
 829                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 830                 rc = -EALREADY;
 831         } else {
 832                 /* notify that the qsd_instance is now started */
 833                 qsd->qsd_started = true;
 834         }
 835         write_unlock(&qsd->qsd_lock);
 836
 837         if (rc)
 838                 RETURN(rc);
 839
 840         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 841          * up to usage; If usage < granted, release down to usage.  */
 842         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 843                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 844                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 845         }
 846
 847         RETURN(rc);
 848 }
 849 EXPORT_SYMBOL(qsd_start);
 850
 851 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 852
 853 /*
 854  * Global initialization performed at module load time
 855  */
 856 int qsd_glb_init(void)
 857 {
 858         int     rc;
 859
 860         rc = lu_kmem_init(qsd_caches);
 861         if (rc)
 862                 return rc;
 863
 864         qsd_key_init_generic(&qsd_thread_key, NULL);
 865         lu_context_key_register(&qsd_thread_key);
 866         lustre_register_quota_process_config(qsd_process_config);
 867
 868         return 0;
 869 }
 870
 871 /*
 872  * Companion of qsd_glb_init() called at module unload time
 873  */
 874 void qsd_glb_fini(void)
 875 {
 876         lustre_register_quota_process_config(NULL);
 877         lu_kmem_fini(qsd_caches);
 878         lu_context_key_degister(&qsd_thread_key);
 879 }