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