lustre/quota/qsd_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2012, 2014, Intel Corporation.
  25  * Use is subject to license terms.
  26  *
  27  * Author: Johann Lombardi <johann.lombardi@intel.com>
  28  * Author: Niu    Yawei    <yawei.niu@intel.com>
  29  */
  30
  31 /*
  32  * Quota Slave Driver (QSD) management.
  33  *
  34  * The quota slave feature is implemented under the form of a library called
  35  * QSD. Each OSD device should create a QSD instance via qsd_init() which will
  36  * be used to manage quota enforcement for this device. This implies:
  37  * - completing the reintegration procedure with the quota master (aka QMT, see
  38  *   qmt_dev.c) to retrieve the latest quota settings and space distribution.
  39  * - managing quota locks in order to be notified of configuration changes.
  40  * - acquiring space from the QMT when quota space for a given user/group is
  41  *   close to exhaustion.
  42  * - allocating quota space to service threads for local request processing.
  43  *
  44  * Once the QSD instance created, the OSD device should invoke qsd_start()
  45  * when recovery is completed. This notifies the QSD that we are about to
  46  * process new requests on which quota should be strictly enforced.
  47  * Then, qsd_op_begin/end can be used to reserve/release/pre-acquire quota space
  48  * for/after each operation until shutdown where the QSD instance should be
  49  * freed via qsd_fini().
  50  */
  51
  52 #define DEBUG_SUBSYSTEM S_LQUOTA
  53
  54 #include <obd_class.h>
  55 #include "qsd_internal.h"
  56
  57 struct kmem_cache *upd_kmem;
  58
  59 struct lu_kmem_descr qsd_caches[] = {
  60         {
  61                 .ckd_cache = &upd_kmem,
  62                 .ckd_name  = "upd_kmem",
  63                 .ckd_size  = sizeof(struct qsd_upd_rec)
  64         },
  65         {
  66                 .ckd_cache = NULL
  67         }
  68 };
  69
  70 /* define qsd thread key */
  71 LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
  72 LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
  73 LU_KEY_INIT_GENERIC(qsd);
  74
  75 /* some procfs helpers */
  76 static int qsd_state_seq_show(struct seq_file *m, void *data)
  77 {
  78         struct qsd_instance     *qsd = m->private;
  79         char                     enabled[5];
  80
  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 < MAXQUOTAS; qtype++) {
 164                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 165                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 166                 }
 167         }
 168         write_unlock(&qsd->qsd_lock);
 169
 170         if (rc)
 171                 return rc;
 172
 173         /* kick off reintegration */
 174         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 175                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 176                 if (rc)
 177                         break;
 178         }
 179         return rc == 0 ? count : rc;
 180 }
 181 LPROC_SEQ_FOPS_WO_TYPE(qsd, force_reint);
 182
 183 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 184 {
 185         struct qsd_instance *qsd = m->private;
 186         LASSERT(qsd != NULL);
 187
 188         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                  timeout, rc;
 198         LASSERT(qsd != NULL);
 199
 200         rc = lprocfs_write_helper(buffer, count, &timeout);
 201         if (rc)
 202                 return rc;
 203         if (timeout < 0)
 204                 return -EINVAL;
 205
 206         qsd->qsd_timeout = timeout;
 207         return count;
 208 }
 209 LPROC_SEQ_FOPS(qsd_timeout);
 210
 211 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 212         { .name =       "info",
 213           .fops =       &qsd_state_fops         },
 214         { .name =       "enabled",
 215           .fops =       &qsd_enabled_fops       },
 216         { .name =       "force_reint",
 217           .fops =       &qsd_force_reint_fops   },
 218         { .name =       "timeout",
 219           .fops =       &qsd_timeout_fops       },
 220         { NULL }
 221 };
 222
 223 /*
 224  * Callback function invoked by the OSP layer when the connection to the master
 225  * has been set up.
 226  *
 227  * \param data - is a pointer to the qsd_instance
 228  *
 229  * \retval - 0 on success, appropriate error on failure
 230  */
 231 static int qsd_conn_callback(void *data)
 232 {
 233         struct qsd_instance *qsd = (struct qsd_instance *)data;
 234         int                  type;
 235         ENTRY;
 236
 237         /* qsd_exp should now be valid */
 238         LASSERT(qsd->qsd_exp);
 239
 240         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 241
 242         write_lock(&qsd->qsd_lock);
 243         /* notify that qsd_exp is now valid */
 244         qsd->qsd_exp_valid = true;
 245         write_unlock(&qsd->qsd_lock);
 246
 247         /* Now that the connection to master is setup, we can initiate the
 248          * reintegration procedure for quota types which are enabled.
 249          * It is worth noting that, if the qsd_instance hasn't been started
 250          * already, then we can only complete the first two steps of the
 251          * reintegration procedure (i.e. global lock enqueue and slave
 252          * index transfer) since the space usage reconciliation (i.e.
 253          * step 3) will have to wait for qsd_start() to be called */
 254         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 255                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 256                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 257         }
 258
 259         RETURN(0);
 260 }
 261
 262 /*
 263  * Release qsd_qtype_info structure which contains data associated with a
 264  * given quota type. This releases the accounting objects.
 265  * It's called on OSD cleanup when the qsd instance is released.
 266  *
 267  * \param env - is the environment passed by the caller
 268  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 269  *              to be released
 270  * \param qtype - is the quota type to be shutdown
 271  */
 272 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 273                            int qtype)
 274 {
 275         struct qsd_qtype_info   *qqi;
 276         int repeat = 0;
 277         ENTRY;
 278
 279         if (qsd->qsd_type_array[qtype] == NULL)
 280                 RETURN_EXIT;
 281         qqi = qsd->qsd_type_array[qtype];
 282         qsd->qsd_type_array[qtype] = NULL;
 283
 284         /* all deferred work lists should be empty */
 285         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 286         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 287
 288         /* shutdown lquota site */
 289         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 290                 lquota_site_free(env, qqi->qqi_site);
 291                 qqi->qqi_site = NULL;
 292         }
 293
 294         /* The qqi may still be holding by global locks which are being
 295          * canceled asynchronously (LU-4365), see the following steps:
 296          *
 297          * - On server umount, we try to clear all quota locks first by
 298          *   disconnecting LWP (which will invalidate import and cleanup
 299          *   all locks on it), however, if quota reint process is holding
 300          *   the global lock for reintegration at that time, global lock
 301          *   will fail to be cleared on LWP disconnection.
 302          *
 303          * - Umount process goes on and stops reint process, the global
 304          *   lock will be dropped on reint process exit, however, the lock
 305          *   cancel in done in asynchronous way, so the
 306          *   qsd_glb_blocking_ast() might haven't been called yet when we
 307          *   get here.
 308          */
 309         while (atomic_read(&qqi->qqi_ref) > 1) {
 310                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 311                        atomic_read(&qqi->qqi_ref), repeat);
 312                 repeat++;
 313                 set_current_state(TASK_INTERRUPTIBLE);
 314                 schedule_timeout(cfs_time_seconds(1));
 315         }
 316
 317         /* by now, all qqi users should have gone away */
 318         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 319         lu_ref_fini(&qqi->qqi_reference);
 320
 321         /* release accounting object */
 322         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 323                 lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
 324                 qqi->qqi_acct_obj = NULL;
 325         }
 326
 327         /* release slv index */
 328         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 329                 lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
 330                 qqi->qqi_slv_obj = NULL;
 331                 qqi->qqi_slv_ver = 0;
 332         }
 333
 334         /* release global index */
 335         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 336                 lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
 337                 qqi->qqi_glb_obj = NULL;
 338                 qqi->qqi_glb_ver = 0;
 339         }
 340
 341         OBD_FREE_PTR(qqi);
 342         EXIT;
 343 }
 344
 345 /*
 346  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 347  * This opens the accounting object and initializes the proc file.
 348  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 349  * instance.
 350  *
 351  * \param env  - the environment passed by the caller
 352  * \param qsd  - is the qsd instance which will be in charge of the new
 353  *               qsd_qtype_info instance.
 354  * \param qtype - is quota type to set up
 355  *
 356  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 357  *           appropriate error on failure
 358  */
 359 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 360                           int qtype)
 361 {
 362         struct qsd_qtype_info   *qqi;
 363         int                      rc;
 364         struct obd_uuid          uuid;
 365         ENTRY;
 366
 367         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 368
 369         /* allocate structure for this quota type */
 370         OBD_ALLOC_PTR(qqi);
 371         if (qqi == NULL)
 372                 RETURN(-ENOMEM);
 373         qsd->qsd_type_array[qtype] = qqi;
 374         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 375
 376         /* set backpointer and other parameters */
 377         qqi->qqi_qsd   = qsd;
 378         qqi->qqi_qtype = qtype;
 379         lu_ref_init(&qqi->qqi_reference);
 380         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id, QSD_RES_TYPE(qsd),
 381                             qtype);
 382         qqi->qqi_glb_uptodate = false;
 383         qqi->qqi_slv_uptodate = false;
 384         qqi->qqi_reint        = false;
 385         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 386         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 387         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 388         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 389
 390         /* open accounting object */
 391         LASSERT(qqi->qqi_acct_obj == NULL);
 392         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 393         if (IS_ERR(qqi->qqi_acct_obj)) {
 394                 CDEBUG(D_QUOTA, "%s: no %s space accounting support rc:%ld\n",
 395                        qsd->qsd_svname, QTYPE_NAME(qtype),
 396                        PTR_ERR(qqi->qqi_acct_obj));
 397                 qqi->qqi_acct_obj = NULL;
 398                 qsd->qsd_acct_failed = true;
 399         }
 400
 401         /* open global index copy */
 402         LASSERT(qqi->qqi_glb_obj == NULL);
 403         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 404                                                        qsd->qsd_root,
 405                                                        &qqi->qqi_fid, true);
 406         if (IS_ERR(qqi->qqi_glb_obj)) {
 407                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 408                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 409                        PTR_ERR(qqi->qqi_glb_obj));
 410                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 411         }
 412         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 413
 414         /* open slave index copy */
 415         LASSERT(qqi->qqi_slv_obj == NULL);
 416         obd_str2uuid(&uuid, qsd->qsd_svname);
 417         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 418                                                        qsd->qsd_root,
 419                                                        &qqi->qqi_fid, &uuid,
 420                                                        true);
 421         if (IS_ERR(qqi->qqi_slv_obj)) {
 422                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 423                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 424                        PTR_ERR(qqi->qqi_slv_obj));
 425                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 426         }
 427         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 428
 429         /* allocate site */
 430         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 431         if (IS_ERR(qqi->qqi_site)) {
 432                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 433                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 434                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 435         }
 436
 437         /* register proc entry for accounting & global index copy objects */
 438         rc = lprocfs_seq_create(qsd->qsd_proc,
 439                                 qtype == USRQUOTA ? "acct_user" : "acct_group",
 440                                 0444, &lprocfs_quota_seq_fops,
 441                                 qqi->qqi_acct_obj);
 442         if (rc) {
 443                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 444                        qsd->qsd_svname, rc);
 445                 GOTO(out, rc);
 446         }
 447
 448         rc = lprocfs_seq_create(qsd->qsd_proc,
 449                                 qtype == USRQUOTA ? "limit_user" : "limit_group",
 450                                 0444, &lprocfs_quota_seq_fops,
 451                                 qqi->qqi_glb_obj);
 452         if (rc) {
 453                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 454                        qsd->qsd_svname, rc);
 455                 GOTO(out, rc);
 456         }
 457         EXIT;
 458 out:
 459         if (rc)
 460                 qsd_qtype_fini(env, qsd, qtype);
 461         return rc;
 462 }
 463
 464 /*
 465  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 466  * structures associated with the quota slave (on-disk objects, lquota entry
 467  * tables, ...).
 468  * This function should be called when the OSD is shutting down.
 469  *
 470  * \param env - is the environment passed by the caller
 471  * \param qsd - is the qsd instance to shutdown
 472  */
 473 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 474 {
 475         int     qtype;
 476         ENTRY;
 477
 478         if (unlikely(qsd == NULL))
 479                 RETURN_EXIT;
 480
 481         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 482         write_lock(&qsd->qsd_lock);
 483         qsd->qsd_stopping = true;
 484         write_unlock(&qsd->qsd_lock);
 485
 486         /* remove qsd proc entry */
 487         if (qsd->qsd_proc != NULL) {
 488                 lprocfs_remove(&qsd->qsd_proc);
 489                 qsd->qsd_proc = NULL;
 490         }
 491
 492         /* stop the writeback thread */
 493         qsd_stop_upd_thread(qsd);
 494
 495         /* shutdown the reintegration threads */
 496         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 497                 if (qsd->qsd_type_array[qtype] == NULL)
 498                         continue;
 499                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 500         }
 501
 502         if (qsd->qsd_ns != NULL) {
 503                 qsd->qsd_ns = NULL;
 504         }
 505
 506         /* free per-quota type data */
 507         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
 508                 qsd_qtype_fini(env, qsd, qtype);
 509
 510         /* deregister connection to the quota master */
 511         qsd->qsd_exp_valid = false;
 512         lustre_deregister_lwp_item(&qsd->qsd_exp);
 513
 514         /* release per-filesystem information */
 515         if (qsd->qsd_fsinfo != NULL) {
 516                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 517                 /* remove from the list of fsinfo */
 518                 list_del_init(&qsd->qsd_link);
 519                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 520                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 521                 qsd->qsd_fsinfo = NULL;
 522         }
 523
 524         /* release quota root directory */
 525         if (qsd->qsd_root != NULL) {
 526                 lu_object_put(env, &qsd->qsd_root->do_lu);
 527                 qsd->qsd_root = NULL;
 528         }
 529
 530         /* release reference on dt_device */
 531         if (qsd->qsd_dev != NULL) {
 532                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 533                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 534                 qsd->qsd_dev = NULL;
 535         }
 536
 537         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 538         OBD_FREE_PTR(qsd);
 539         EXIT;
 540 }
 541 EXPORT_SYMBOL(qsd_fini);
 542
 543 /*
 544  * Create a new qsd_instance to be associated with backend osd device
 545  * identified by \dev.
 546  *
 547  * \param env    - the environment passed by the caller
 548  * \param svname - is the service name of the OSD device creating this instance
 549  * \param dev    - is the dt_device where to store quota index files
 550  * \param osd_proc - is the procfs parent directory where to create procfs file
 551  *                   related to this new qsd instance
 552  *
 553  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 554  *           appropriate error on failure
 555  */
 556 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 557                               struct dt_device *dev,
 558                               struct proc_dir_entry *osd_proc)
 559 {
 560         struct qsd_thread_info  *qti = qsd_info(env);
 561         struct qsd_instance     *qsd;
 562         int                      rc, type, idx;
 563         ENTRY;
 564
 565         /* only configure qsd for MDT & OST */
 566         type = server_name2index(svname, &idx, NULL);
 567         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 568                 RETURN(NULL);
 569
 570         /* allocate qsd instance */
 571         OBD_ALLOC_PTR(qsd);
 572         if (qsd == NULL)
 573                 RETURN(ERR_PTR(-ENOMEM));
 574
 575         /* generic initializations */
 576         rwlock_init(&qsd->qsd_lock);
 577         INIT_LIST_HEAD(&qsd->qsd_link);
 578         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 579         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 580         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 581         spin_lock_init(&qsd->qsd_adjust_lock);
 582         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 583         qsd->qsd_prepared = false;
 584         qsd->qsd_started = false;
 585
 586         /* copy service name */
 587         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 588             >= sizeof(qsd->qsd_svname))
 589                 GOTO(out, rc = -E2BIG);
 590
 591         /* grab reference on osd device */
 592         lu_device_get(&dev->dd_lu_dev);
 593         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 594         qsd->qsd_dev = dev;
 595
 596         /* we only support pool ID 0 (default data or metadata pool) for the
 597          * time being. A different pool ID could be assigned to this target via
 598          * the configuration log in the future */
 599         qsd->qsd_pool_id  = 0;
 600
 601         /* get fsname from svname */
 602         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 603         if (rc) {
 604                 CERROR("%s: fail to extract filesystem name\n", svname);
 605                 GOTO(out, rc);
 606         }
 607
 608         /* look up quota setting for the filesystem the target belongs to */
 609         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 610         if (qsd->qsd_fsinfo == NULL) {
 611                 CERROR("%s: failed to locate filesystem information\n", svname);
 612                 GOTO(out, rc = -EINVAL);
 613         }
 614
 615         /* add in the list of lquota_fsinfo */
 616         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 617         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 618         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 619
 620         /* register procfs directory */
 621         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 622                                          lprocfs_quota_qsd_vars, qsd);
 623         if (IS_ERR(qsd->qsd_proc)) {
 624                 rc = PTR_ERR(qsd->qsd_proc);
 625                 qsd->qsd_proc = NULL;
 626                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 627                        svname, rc);
 628                 GOTO(out, rc);
 629         }
 630         EXIT;
 631 out:
 632         if (rc) {
 633                 qsd_fini(env, qsd);
 634                 return ERR_PTR(rc);
 635         }
 636         RETURN(qsd);
 637 }
 638 EXPORT_SYMBOL(qsd_init);
 639
 640 /*
 641  * Initialize on-disk structures in order to manage quota enforcement for
 642  * the target associated with the qsd instance \qsd and starts the reintegration
 643  * procedure for each quota type as soon as possible.
 644  * The last step of the reintegration will be completed once qsd_start() is
 645  * called, at which points the space reconciliation with the master will be
 646  * executed.
 647  * This function must be called when the server stack is fully configured,
 648  * typically when ->ldo_prepare is called across the stack.
 649  *
 650  * \param env - the environment passed by the caller
 651  * \param qsd - is qsd_instance to prepare
 652  *
 653  * \retval - 0 on success, appropriate error on failure
 654  */
 655 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 656 {
 657         struct qsd_thread_info  *qti = qsd_info(env);
 658         int                      qtype, rc = 0;
 659         ENTRY;
 660
 661         if (unlikely(qsd == NULL))
 662                 RETURN(0);
 663
 664         read_lock(&qsd->qsd_lock);
 665         if (qsd->qsd_prepared) {
 666                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 667                 rc = -EALREADY;
 668         }
 669         read_unlock(&qsd->qsd_lock);
 670         if (rc)
 671                 RETURN(rc);
 672
 673         /* Record whether this qsd instance is managing quota enforcement for a
 674          * MDT (i.e. inode quota) or OST (block quota) */
 675         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 676                 qsd->qsd_is_md = true;
 677                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 678         } else {
 679                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 680         }
 681
 682         /* look-up on-disk directory for the quota slave */
 683         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 684                                                     QSD_DIR);
 685         if (IS_ERR(qsd->qsd_root)) {
 686                 rc = PTR_ERR(qsd->qsd_root);
 687                 qsd->qsd_root = NULL;
 688                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 689                        qsd->qsd_svname, rc);
 690                 RETURN(rc);
 691         }
 692
 693         /* initialize per-quota type data */
 694         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 695                 rc = qsd_qtype_init(env, qsd, qtype);
 696                 if (rc)
 697                         RETURN(rc);
 698         }
 699
 700         /* pools successfully setup, mark the qsd as prepared */
 701         write_lock(&qsd->qsd_lock);
 702         qsd->qsd_prepared = true;
 703         write_unlock(&qsd->qsd_lock);
 704
 705         /* start reintegration thread for each type, if required */
 706         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 707                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 708
 709                 if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
 710                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 711                                        "since space accounting isn't functional"
 712                                        ". Please run tunefs.lustre --quota on "
 713                                        "an unmounted filesystem if not done "
 714                                        "already\n", qsd->qsd_svname);
 715                         break;
 716                 }
 717
 718                 rc = qsd_start_reint_thread(qqi);
 719                 if (rc) {
 720                         CERROR("%s: failed to start reint thread for type %s "
 721                                "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
 722                                rc);
 723                         RETURN(rc);
 724                 }
 725         }
 726
 727         /* start writeback thread */
 728         rc = qsd_start_upd_thread(qsd);
 729         if (rc) {
 730                 CERROR("%s: failed to start writeback thread (%d)\n",
 731                        qsd->qsd_svname, rc);
 732                 RETURN(rc);
 733         }
 734
 735         /* generate osp name */
 736         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 737                                MTI_NAME_MAXLEN, 0);
 738         if (rc) {
 739                 CERROR("%s: failed to generate ospname (%d)\n",
 740                        qsd->qsd_svname, rc);
 741                 RETURN(rc);
 742         }
 743
 744         /* the connection callback will start the reintegration
 745          * procedure if quota is enabled */
 746         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 747                                       qsd_conn_callback, (void *)qsd);
 748         if (rc) {
 749                 CERROR("%s: fail to get connection to master (%d)\n",
 750                        qsd->qsd_svname, rc);
 751                 RETURN(rc);
 752         }
 753
 754         RETURN(0);
 755 }
 756 EXPORT_SYMBOL(qsd_prepare);
 757
 758 /*
 759  * Start a qsd instance. This will complete the last step of the reintegration
 760  * procedure as soon as possible (provided that the master is reachable).
 761  * This should be called when recovery has been completed and quota should now
 762  * be enforced on every operations.
 763  *
 764  * \param env - the environment passed by the caller
 765  * \param qsd - is the qsd instance associated with the osd device to start
 766  */
 767 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 768 {
 769         int     type, rc = 0;
 770         ENTRY;
 771
 772         if (unlikely(qsd == NULL))
 773                 RETURN(0);
 774
 775         write_lock(&qsd->qsd_lock);
 776         if (!qsd->qsd_prepared) {
 777                 CERROR("%s: can't start qsd instance since it wasn't properly "
 778                        "initialized\n", qsd->qsd_svname);
 779                 rc = -EFAULT;
 780         } else if (qsd->qsd_started) {
 781                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 782                 rc = -EALREADY;
 783         } else {
 784                 /* notify that the qsd_instance is now started */
 785                 qsd->qsd_started = true;
 786         }
 787         write_unlock(&qsd->qsd_lock);
 788
 789         if (rc)
 790                 RETURN(rc);
 791
 792         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 793          * up to usage; If usage < granted, release down to usage.  */
 794         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 795                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 796                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 797         }
 798
 799         RETURN(rc);
 800 }
 801 EXPORT_SYMBOL(qsd_start);
 802
 803 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 804
 805 /*
 806  * Global initialization performed at module load time
 807  */
 808 int qsd_glb_init(void)
 809 {
 810         int     rc;
 811
 812         rc = lu_kmem_init(qsd_caches);
 813         if (rc)
 814                 return rc;
 815
 816         qsd_key_init_generic(&qsd_thread_key, NULL);
 817         lu_context_key_register(&qsd_thread_key);
 818         lustre_register_quota_process_config(qsd_process_config);
 819
 820         return 0;
 821 }
 822
 823 /*
 824  * Companion of qsd_glb_init() called at module unload time
 825  */
 826 void qsd_glb_fini(void)
 827 {
 828         lustre_register_quota_process_config(NULL);
 829         lu_kmem_fini(qsd_caches);
 830         lu_context_key_degister(&qsd_thread_key);
 831 }