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, 2013, Intel Corporation.
  25  * Use is subject to license terms.
  26  *
  27  * Author: Johann Lombardi <johann.lombardi@intel.com>
  28  * Author: Niu    Yawei    <yawei.niu@intel.com>
  29  */
  30
  31 /*
  32  * Quota Slave Driver (QSD) management.
  33  *
  34  * The quota slave feature is implemented under the form of a library called
  35  * QSD. Each OSD device should create a QSD instance via qsd_init() which will
  36  * be used to manage quota enforcement for this device. This implies:
  37  * - completing the reintegration procedure with the quota master (aka QMT, see
  38  *   qmt_dev.c) to retrieve the latest quota settings and space distribution.
  39  * - managing quota locks in order to be notified of configuration changes.
  40  * - acquiring space from the QMT when quota space for a given user/group is
  41  *   close to exhaustion.
  42  * - allocating quota space to service threads for local request processing.
  43  *
  44  * Once the QSD instance created, the OSD device should invoke qsd_start()
  45  * when recovery is completed. This notifies the QSD that we are about to
  46  * process new requests on which quota should be strictly enforced.
  47  * Then, qsd_op_begin/end can be used to reserve/release/pre-acquire quota space
  48  * for/after each operation until shutdown where the QSD instance should be
  49  * freed via qsd_fini().
  50  */
  51
  52 #define DEBUG_SUBSYSTEM S_LQUOTA
  53
  54 #include <obd_class.h>
  55 #include "qsd_internal.h"
  56
  57 struct kmem_cache *upd_kmem;
  58
  59 struct lu_kmem_descr qsd_caches[] = {
  60         {
  61                 .ckd_cache = &upd_kmem,
  62                 .ckd_name  = "upd_kmem",
  63                 .ckd_size  = sizeof(struct qsd_upd_rec)
  64         },
  65         {
  66                 .ckd_cache = NULL
  67         }
  68 };
  69
  70 /* define qsd thread key */
  71 LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
  72 LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
  73 LU_KEY_INIT_GENERIC(qsd);
  74
  75 /* some procfs helpers */
  76 static int qsd_state_seq_show(struct seq_file *m, void *data)
  77 {
  78         struct qsd_instance     *qsd = m->private;
  79         char                     enabled[5];
  80         int                      rc;
  81
  82         LASSERT(qsd != NULL);
  83
  84         memset(enabled, 0, sizeof(enabled));
  85         if (qsd_type_enabled(qsd, USRQUOTA))
  86                 strcat(enabled, "u");
  87         if (qsd_type_enabled(qsd, GRPQUOTA))
  88                 strcat(enabled, "g");
  89         if (strlen(enabled) == 0)
  90                 strcat(enabled, "none");
  91
  92         rc = seq_printf(m, "target name:    %s\n"
  93                         "pool ID:        %d\n"
  94                         "type:           %s\n"
  95                         "quota enabled:  %s\n"
  96                         "conn to master: %s\n",
  97                         qsd->qsd_svname, qsd->qsd_pool_id,
  98                         qsd->qsd_is_md ? "md" : "dt", enabled,
  99                         qsd->qsd_exp_valid ? "setup" : "not setup yet");
 100
 101         if (qsd->qsd_prepared) {
 102                 memset(enabled, 0, sizeof(enabled));
 103                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 104                         strcat(enabled, "u");
 105                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 106                         strcat(enabled, "g");
 107                 if (strlen(enabled) == 0)
 108                         strcat(enabled, "none");
 109                 rc += seq_printf(m, "space acct:     %s\n"
 110                                 "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 111                                 "group uptodate: glb[%d],slv[%d],reint[%d]\n",
 112                                 enabled,
 113                                 qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 114                                 qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 115                                 qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 116                                 qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 117                                 qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 118                                 qsd->qsd_type_array[GRPQUOTA]->qqi_reint);
 119         }
 120         return rc;
 121 }
 122 LPROC_SEQ_FOPS_RO(qsd_state);
 123
 124 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
 125 {
 126         struct qsd_instance     *qsd = m->private;
 127         char                     enabled[5];
 128
 129         LASSERT(qsd != NULL);
 130
 131         memset(enabled, 0, sizeof(enabled));
 132         if (qsd_type_enabled(qsd, USRQUOTA))
 133                 strcat(enabled, "u");
 134         if (qsd_type_enabled(qsd, GRPQUOTA))
 135                 strcat(enabled, "g");
 136         if (strlen(enabled) == 0)
 137                 strcat(enabled, "none");
 138
 139         return seq_printf(m, "%s\n", enabled);
 140 }
 141 LPROC_SEQ_FOPS_RO(qsd_enabled);
 142
 143 /* force reintegration procedure to be executed.
 144  * Used for test/debugging purpose */
 145 static ssize_t
 146 lprocfs_force_reint_seq_write(struct file *file, const char *buffer,
 147                                 size_t count, loff_t *off)
 148 {
 149         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 150         int                  rc = 0, qtype;
 151
 152         LASSERT(qsd != NULL);
 153
 154         write_lock(&qsd->qsd_lock);
 155         if (qsd->qsd_stopping) {
 156                 /* don't mess up with shutdown procedure, it is already
 157                  * complicated enough */
 158                 rc = -ESHUTDOWN;
 159         } else if (!qsd->qsd_prepared) {
 160                 rc = -EAGAIN;
 161         } else {
 162                 /* mark all indexes as stale */
 163                 for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 164                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 165                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 166                 }
 167         }
 168         write_unlock(&qsd->qsd_lock);
 169
 170         if (rc)
 171                 return rc;
 172
 173         /* kick off reintegration */
 174         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 175                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 176                 if (rc)
 177                         break;
 178         }
 179         return rc == 0 ? count : rc;
 180 }
 181 LPROC_SEQ_FOPS_WO_TYPE(qsd, force_reint);
 182
 183 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 184 {
 185         struct qsd_instance *qsd = m->private;
 186         LASSERT(qsd != NULL);
 187
 188         return seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 189 }
 190
 191 static ssize_t
 192 qsd_timeout_seq_write(struct file *file, const char *buffer,
 193                         size_t count, loff_t *off)
 194 {
 195         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 196         int                  timeout, rc;
 197         LASSERT(qsd != NULL);
 198
 199         rc = lprocfs_write_helper(buffer, count, &timeout);
 200         if (rc)
 201                 return rc;
 202         if (timeout < 0)
 203                 return -EINVAL;
 204
 205         qsd->qsd_timeout = timeout;
 206         return count;
 207 }
 208 LPROC_SEQ_FOPS(qsd_timeout);
 209
 210 static struct lprocfs_seq_vars lprocfs_quota_qsd_vars[] = {
 211         { "info",               &qsd_state_fops         },
 212         { "enabled",            &qsd_enabled_fops       },
 213         { "force_reint",        &qsd_force_reint_fops   },
 214         { "timeout",            &qsd_timeout_fops       },
 215         { NULL }
 216 };
 217
 218 /*
 219  * Callback function invoked by the OSP layer when the connection to the master
 220  * has been set up.
 221  *
 222  * \param data - is a pointer to the qsd_instance
 223  *
 224  * \retval - 0 on success, appropriate error on failure
 225  */
 226 static int qsd_conn_callback(void *data)
 227 {
 228         struct qsd_instance *qsd = (struct qsd_instance *)data;
 229         int                  type;
 230         ENTRY;
 231
 232         /* qsd_exp should now be valid */
 233         LASSERT(qsd->qsd_exp);
 234
 235         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 236
 237         write_lock(&qsd->qsd_lock);
 238         /* notify that qsd_exp is now valid */
 239         qsd->qsd_exp_valid = true;
 240         write_unlock(&qsd->qsd_lock);
 241
 242         /* Now that the connection to master is setup, we can initiate the
 243          * reintegration procedure for quota types which are enabled.
 244          * It is worth noting that, if the qsd_instance hasn't been started
 245          * already, then we can only complete the first two steps of the
 246          * reintegration procedure (i.e. global lock enqueue and slave
 247          * index transfer) since the space usage reconciliation (i.e.
 248          * step 3) will have to wait for qsd_start() to be called */
 249         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 250                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 251                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 252         }
 253
 254         RETURN(0);
 255 }
 256
 257 /*
 258  * Release qsd_qtype_info structure which contains data associated with a
 259  * given quota type. This releases the accounting objects.
 260  * It's called on OSD cleanup when the qsd instance is released.
 261  *
 262  * \param env - is the environment passed by the caller
 263  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 264  *              to be released
 265  * \param qtype - is the quota type to be shutdown
 266  */
 267 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 268                            int qtype)
 269 {
 270         struct qsd_qtype_info   *qqi;
 271         int repeat = 0;
 272         ENTRY;
 273
 274         if (qsd->qsd_type_array[qtype] == NULL)
 275                 RETURN_EXIT;
 276         qqi = qsd->qsd_type_array[qtype];
 277         qsd->qsd_type_array[qtype] = NULL;
 278
 279         /* all deferred work lists should be empty */
 280         LASSERT(cfs_list_empty(&qqi->qqi_deferred_glb));
 281         LASSERT(cfs_list_empty(&qqi->qqi_deferred_slv));
 282
 283         /* shutdown lquota site */
 284         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 285                 lquota_site_free(env, qqi->qqi_site);
 286                 qqi->qqi_site = NULL;
 287         }
 288
 289         /* The qqi may still be holding by global locks which are being
 290          * canceled asynchronously (LU-4365), see the following steps:
 291          *
 292          * - On server umount, we try to clear all quota locks first by
 293          *   disconnecting LWP (which will invalidate import and cleanup
 294          *   all locks on it), however, if quota reint process is holding
 295          *   the global lock for reintegration at that time, global lock
 296          *   will fail to be cleared on LWP disconnection.
 297          *
 298          * - Umount process goes on and stops reint process, the global
 299          *   lock will be dropped on reint process exit, however, the lock
 300          *   cancel in done in asynchronous way, so the
 301          *   qsd_glb_blocking_ast() might haven't been called yet when we
 302          *   get here.
 303          */
 304         while (cfs_atomic_read(&qqi->qqi_ref) > 1) {
 305                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 306                        cfs_atomic_read(&qqi->qqi_ref), repeat);
 307                 repeat++;
 308                 schedule_timeout_and_set_state(TASK_INTERRUPTIBLE,
 309                                                 cfs_time_seconds(1));
 310         }
 311
 312         /* by now, all qqi users should have gone away */
 313         LASSERT(cfs_atomic_read(&qqi->qqi_ref) == 1);
 314         lu_ref_fini(&qqi->qqi_reference);
 315
 316         /* release accounting object */
 317         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 318                 lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
 319                 qqi->qqi_acct_obj = NULL;
 320         }
 321
 322         /* release slv index */
 323         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 324                 lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
 325                 qqi->qqi_slv_obj = NULL;
 326                 qqi->qqi_slv_ver = 0;
 327         }
 328
 329         /* release global index */
 330         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 331                 lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
 332                 qqi->qqi_glb_obj = NULL;
 333                 qqi->qqi_glb_ver = 0;
 334         }
 335
 336         OBD_FREE_PTR(qqi);
 337         EXIT;
 338 }
 339
 340 /*
 341  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 342  * This opens the accounting object and initializes the proc file.
 343  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 344  * instance.
 345  *
 346  * \param env  - the environment passed by the caller
 347  * \param qsd  - is the qsd instance which will be in charge of the new
 348  *               qsd_qtype_info instance.
 349  * \param qtype - is quota type to set up
 350  *
 351  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 352  *           appropriate error on failure
 353  */
 354 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 355                           int qtype)
 356 {
 357         struct qsd_qtype_info   *qqi;
 358         int                      rc;
 359         struct obd_uuid          uuid;
 360         ENTRY;
 361
 362         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 363
 364         /* allocate structure for this quota type */
 365         OBD_ALLOC_PTR(qqi);
 366         if (qqi == NULL)
 367                 RETURN(-ENOMEM);
 368         qsd->qsd_type_array[qtype] = qqi;
 369         cfs_atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 370
 371         /* set backpointer and other parameters */
 372         qqi->qqi_qsd   = qsd;
 373         qqi->qqi_qtype = qtype;
 374         lu_ref_init(&qqi->qqi_reference);
 375         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id, QSD_RES_TYPE(qsd),
 376                             qtype);
 377         qqi->qqi_glb_uptodate = false;
 378         qqi->qqi_slv_uptodate = false;
 379         qqi->qqi_reint        = false;
 380         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 381         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 382         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 383         CFS_INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 384
 385         /* open accounting object */
 386         LASSERT(qqi->qqi_acct_obj == NULL);
 387         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 388         if (IS_ERR(qqi->qqi_acct_obj)) {
 389                 CDEBUG(D_QUOTA, "%s: no %s space accounting support rc:%ld\n",
 390                        qsd->qsd_svname, QTYPE_NAME(qtype),
 391                        PTR_ERR(qqi->qqi_acct_obj));
 392                 qqi->qqi_acct_obj = NULL;
 393                 qsd->qsd_acct_failed = true;
 394         }
 395
 396         /* open global index copy */
 397         LASSERT(qqi->qqi_glb_obj == NULL);
 398         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 399                                                        qsd->qsd_root,
 400                                                        &qqi->qqi_fid, true);
 401         if (IS_ERR(qqi->qqi_glb_obj)) {
 402                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 403                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 404                        PTR_ERR(qqi->qqi_glb_obj));
 405                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 406         }
 407         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 408
 409         /* open slave index copy */
 410         LASSERT(qqi->qqi_slv_obj == NULL);
 411         obd_str2uuid(&uuid, qsd->qsd_svname);
 412         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 413                                                        qsd->qsd_root,
 414                                                        &qqi->qqi_fid, &uuid,
 415                                                        true);
 416         if (IS_ERR(qqi->qqi_slv_obj)) {
 417                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 418                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 419                        PTR_ERR(qqi->qqi_slv_obj));
 420                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 421         }
 422         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 423
 424         /* allocate site */
 425         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 426         if (IS_ERR(qqi->qqi_site)) {
 427                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 428                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 429                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 430         }
 431
 432         /* register proc entry for accounting & global index copy objects */
 433         rc = lprocfs_seq_create(qsd->qsd_proc,
 434                                 qtype == USRQUOTA ? "acct_user" : "acct_group",
 435                                 0444, &lprocfs_quota_seq_fops,
 436                                 qqi->qqi_acct_obj);
 437         if (rc) {
 438                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 439                        qsd->qsd_svname, rc);
 440                 GOTO(out, rc);
 441         }
 442
 443         rc = lprocfs_seq_create(qsd->qsd_proc,
 444                                 qtype == USRQUOTA ? "limit_user" : "limit_group",
 445                                 0444, &lprocfs_quota_seq_fops,
 446                                 qqi->qqi_glb_obj);
 447         if (rc) {
 448                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 449                        qsd->qsd_svname, rc);
 450                 GOTO(out, rc);
 451         }
 452         EXIT;
 453 out:
 454         if (rc)
 455                 qsd_qtype_fini(env, qsd, qtype);
 456         return rc;
 457 }
 458
 459 /*
 460  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 461  * structures associated with the quota slave (on-disk objects, lquota entry
 462  * tables, ...).
 463  * This function should be called when the OSD is shutting down.
 464  *
 465  * \param env - is the environment passed by the caller
 466  * \param qsd - is the qsd instance to shutdown
 467  */
 468 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 469 {
 470         int     qtype;
 471         ENTRY;
 472
 473         if (unlikely(qsd == NULL))
 474                 RETURN_EXIT;
 475
 476         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 477         write_lock(&qsd->qsd_lock);
 478         qsd->qsd_stopping = true;
 479         write_unlock(&qsd->qsd_lock);
 480
 481         /* remove qsd proc entry */
 482         if (qsd->qsd_proc != NULL) {
 483                 lprocfs_remove(&qsd->qsd_proc);
 484                 qsd->qsd_proc = NULL;
 485         }
 486
 487         /* stop the writeback thread */
 488         qsd_stop_upd_thread(qsd);
 489
 490         /* shutdown the reintegration threads */
 491         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 492                 if (qsd->qsd_type_array[qtype] == NULL)
 493                         continue;
 494                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 495         }
 496
 497         if (qsd->qsd_ns != NULL) {
 498                 qsd->qsd_ns = NULL;
 499         }
 500
 501         /* free per-quota type data */
 502         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++)
 503                 qsd_qtype_fini(env, qsd, qtype);
 504
 505         /* deregister connection to the quota master */
 506         qsd->qsd_exp_valid = false;
 507         lustre_deregister_lwp_item(&qsd->qsd_exp);
 508
 509         /* release per-filesystem information */
 510         if (qsd->qsd_fsinfo != NULL) {
 511                 down(&qsd->qsd_fsinfo->qfs_sem);
 512                 /* remove from the list of fsinfo */
 513                 cfs_list_del_init(&qsd->qsd_link);
 514                 up(&qsd->qsd_fsinfo->qfs_sem);
 515                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 516                 qsd->qsd_fsinfo = NULL;
 517         }
 518
 519         /* release quota root directory */
 520         if (qsd->qsd_root != NULL) {
 521                 lu_object_put(env, &qsd->qsd_root->do_lu);
 522                 qsd->qsd_root = NULL;
 523         }
 524
 525         /* release reference on dt_device */
 526         if (qsd->qsd_dev != NULL) {
 527                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 528                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 529                 qsd->qsd_dev = NULL;
 530         }
 531
 532         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 533         OBD_FREE_PTR(qsd);
 534         EXIT;
 535 }
 536 EXPORT_SYMBOL(qsd_fini);
 537
 538 /*
 539  * Create a new qsd_instance to be associated with backend osd device
 540  * identified by \dev.
 541  *
 542  * \param env    - the environment passed by the caller
 543  * \param svname - is the service name of the OSD device creating this instance
 544  * \param dev    - is the dt_device where to store quota index files
 545  * \param osd_proc - is the procfs parent directory where to create procfs file
 546  *                   related to this new qsd instance
 547  *
 548  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 549  *           appropriate error on failure
 550  */
 551 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 552                               struct dt_device *dev,
 553                               cfs_proc_dir_entry_t *osd_proc)
 554 {
 555         struct qsd_thread_info  *qti = qsd_info(env);
 556         struct qsd_instance     *qsd;
 557         int                      rc, type, idx;
 558         ENTRY;
 559
 560         /* only configure qsd for MDT & OST */
 561         type = server_name2index(svname, &idx, NULL);
 562         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 563                 RETURN(NULL);
 564
 565         /* allocate qsd instance */
 566         OBD_ALLOC_PTR(qsd);
 567         if (qsd == NULL)
 568                 RETURN(ERR_PTR(-ENOMEM));
 569
 570         /* generic initializations */
 571         rwlock_init(&qsd->qsd_lock);
 572         CFS_INIT_LIST_HEAD(&qsd->qsd_link);
 573         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 574         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 575         CFS_INIT_LIST_HEAD(&qsd->qsd_upd_list);
 576         spin_lock_init(&qsd->qsd_adjust_lock);
 577         CFS_INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 578         qsd->qsd_prepared = false;
 579         qsd->qsd_started = false;
 580
 581         /* copy service name */
 582         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 583             >= sizeof(qsd->qsd_svname))
 584                 GOTO(out, rc = -E2BIG);
 585
 586         /* grab reference on osd device */
 587         lu_device_get(&dev->dd_lu_dev);
 588         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 589         qsd->qsd_dev = dev;
 590
 591         /* we only support pool ID 0 (default data or metadata pool) for the
 592          * time being. A different pool ID could be assigned to this target via
 593          * the configuration log in the future */
 594         qsd->qsd_pool_id  = 0;
 595
 596         /* get fsname from svname */
 597         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 598         if (rc) {
 599                 CERROR("%s: fail to extract filesystem name\n", svname);
 600                 GOTO(out, rc);
 601         }
 602
 603         /* look up quota setting for the filesystem the target belongs to */
 604         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 605         if (qsd->qsd_fsinfo == NULL) {
 606                 CERROR("%s: failed to locate filesystem information\n", svname);
 607                 GOTO(out, rc = -EINVAL);
 608         }
 609
 610         /* add in the list of lquota_fsinfo */
 611         down(&qsd->qsd_fsinfo->qfs_sem);
 612         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 613         up(&qsd->qsd_fsinfo->qfs_sem);
 614
 615         /* register procfs directory */
 616         qsd->qsd_proc = lprocfs_seq_register(QSD_DIR, osd_proc,
 617                                                 lprocfs_quota_qsd_vars, qsd);
 618         if (IS_ERR(qsd->qsd_proc)) {
 619                 rc = PTR_ERR(qsd->qsd_proc);
 620                 qsd->qsd_proc = NULL;
 621                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 622                        svname, rc);
 623                 GOTO(out, rc);
 624         }
 625         EXIT;
 626 out:
 627         if (rc) {
 628                 qsd_fini(env, qsd);
 629                 return ERR_PTR(rc);
 630         }
 631         RETURN(qsd);
 632 }
 633 EXPORT_SYMBOL(qsd_init);
 634
 635 /*
 636  * Initialize on-disk structures in order to manage quota enforcement for
 637  * the target associated with the qsd instance \qsd and starts the reintegration
 638  * procedure for each quota type as soon as possible.
 639  * The last step of the reintegration will be completed once qsd_start() is
 640  * called, at which points the space reconciliation with the master will be
 641  * executed.
 642  * This function must be called when the server stack is fully configured,
 643  * typically when ->ldo_prepare is called across the stack.
 644  *
 645  * \param env - the environment passed by the caller
 646  * \param qsd - is qsd_instance to prepare
 647  *
 648  * \retval - 0 on success, appropriate error on failure
 649  */
 650 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 651 {
 652         struct qsd_thread_info  *qti = qsd_info(env);
 653         int                      qtype, rc = 0;
 654         ENTRY;
 655
 656         if (unlikely(qsd == NULL))
 657                 RETURN(0);
 658
 659         read_lock(&qsd->qsd_lock);
 660         if (qsd->qsd_prepared) {
 661                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 662                 rc = -EALREADY;
 663         }
 664         read_unlock(&qsd->qsd_lock);
 665         if (rc)
 666                 RETURN(rc);
 667
 668         /* Record whether this qsd instance is managing quota enforcement for a
 669          * MDT (i.e. inode quota) or OST (block quota) */
 670         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 671                 qsd->qsd_is_md = true;
 672                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 673         } else {
 674                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 675         }
 676
 677         /* look-up on-disk directory for the quota slave */
 678         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 679                                                     QSD_DIR);
 680         if (IS_ERR(qsd->qsd_root)) {
 681                 rc = PTR_ERR(qsd->qsd_root);
 682                 qsd->qsd_root = NULL;
 683                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 684                        qsd->qsd_svname, rc);
 685                 RETURN(rc);
 686         }
 687
 688         /* initialize per-quota type data */
 689         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 690                 rc = qsd_qtype_init(env, qsd, qtype);
 691                 if (rc)
 692                         RETURN(rc);
 693         }
 694
 695         /* pools successfully setup, mark the qsd as prepared */
 696         write_lock(&qsd->qsd_lock);
 697         qsd->qsd_prepared = true;
 698         write_unlock(&qsd->qsd_lock);
 699
 700         /* start reintegration thread for each type, if required */
 701         for (qtype = USRQUOTA; qtype < MAXQUOTAS; qtype++) {
 702                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 703
 704                 if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
 705                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 706                                        "since space accounting isn't functional"
 707                                        ". Please run tunefs.lustre --quota on "
 708                                        "an unmounted filesystem if not done "
 709                                        "already\n", qsd->qsd_svname);
 710                         break;
 711                 }
 712
 713                 rc = qsd_start_reint_thread(qqi);
 714                 if (rc) {
 715                         CERROR("%s: failed to start reint thread for type %s "
 716                                "(%d)\n", qsd->qsd_svname, QTYPE_NAME(qtype),
 717                                rc);
 718                         RETURN(rc);
 719                 }
 720         }
 721
 722         /* start writeback thread */
 723         rc = qsd_start_upd_thread(qsd);
 724         if (rc) {
 725                 CERROR("%s: failed to start writeback thread (%d)\n",
 726                        qsd->qsd_svname, rc);
 727                 RETURN(rc);
 728         }
 729
 730         /* generate osp name */
 731         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 732                                MTI_NAME_MAXLEN, 0);
 733         if (rc) {
 734                 CERROR("%s: failed to generate ospname (%d)\n",
 735                        qsd->qsd_svname, rc);
 736                 RETURN(rc);
 737         }
 738
 739         /* the connection callback will start the reintegration
 740          * procedure if quota is enabled */
 741         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 742                                       qsd_conn_callback, (void *)qsd);
 743         if (rc) {
 744                 CERROR("%s: fail to get connection to master (%d)\n",
 745                        qsd->qsd_svname, rc);
 746                 RETURN(rc);
 747         }
 748
 749         RETURN(0);
 750 }
 751 EXPORT_SYMBOL(qsd_prepare);
 752
 753 /*
 754  * Start a qsd instance. This will complete the last step of the reintegration
 755  * procedure as soon as possible (provided that the master is reachable).
 756  * This should be called when recovery has been completed and quota should now
 757  * be enforced on every operations.
 758  *
 759  * \param env - the environment passed by the caller
 760  * \param qsd - is the qsd instance associated with the osd device to start
 761  */
 762 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 763 {
 764         int     type, rc = 0;
 765         ENTRY;
 766
 767         if (unlikely(qsd == NULL))
 768                 RETURN(0);
 769
 770         write_lock(&qsd->qsd_lock);
 771         if (!qsd->qsd_prepared) {
 772                 CERROR("%s: can't start qsd instance since it wasn't properly "
 773                        "initialized\n", qsd->qsd_svname);
 774                 rc = -EFAULT;
 775         } else if (qsd->qsd_started) {
 776                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 777                 rc = -EALREADY;
 778         } else {
 779                 /* notify that the qsd_instance is now started */
 780                 qsd->qsd_started = true;
 781         }
 782         write_unlock(&qsd->qsd_lock);
 783
 784         if (rc)
 785                 RETURN(rc);
 786
 787         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 788          * up to usage; If usage < granted, release down to usage.  */
 789         for (type = USRQUOTA; type < MAXQUOTAS; type++) {
 790                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 791                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 792         }
 793
 794         RETURN(rc);
 795 }
 796 EXPORT_SYMBOL(qsd_start);
 797
 798 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 799
 800 /*
 801  * Global initialization performed at module load time
 802  */
 803 int qsd_glb_init(void)
 804 {
 805         int     rc;
 806
 807         rc = lu_kmem_init(qsd_caches);
 808         if (rc)
 809                 return rc;
 810
 811         qsd_key_init_generic(&qsd_thread_key, NULL);
 812         lu_context_key_register(&qsd_thread_key);
 813         lustre_register_quota_process_config(qsd_process_config);
 814
 815         return 0;
 816 }
 817
 818 /*
 819  * Companion of qsd_glb_init() called at module unload time
 820  */
 821 void qsd_glb_fini(void)
 822 {
 823         lustre_register_quota_process_config(NULL);
 824         lu_kmem_fini(qsd_caches);
 825         lu_context_key_degister(&qsd_thread_key);
 826 }